Skip to content

Latest commit

 

History

History
1868 lines (1273 loc) · 158 KB

CHANGELOG.md

File metadata and controls

1868 lines (1273 loc) · 158 KB

Changelog

All notable changes to CITY_NET are documented here. Format follows Keep a Changelog.


[1.14.10] - 2026-10-22

Node.js's October 2026 security fixes.

Security

  • The server runs Node.js 24. Node.js's security release of 21 October 2026 fixed ten flaws, the worst an out-of-bounds write in its Brotli decoder that could let an attacker run code on the server (CVE-2026-111497). The fixes came only for Node 22, 24 and 26; CITY_NET's server ran Node 20, which is past its end and gets no more fixes. The server's image now runs Node 24 (24.22.1 or newer), as do the build and the tests. That also covers the replies to the server's own requests (the in-app updater asking Docker Hub, and the sheet import), which Node decodes too.
  • Compressed request bodies are refused. The server used to decompress any request body labelled as compressed, with that same decoder, from anyone who could reach it, signed in or not. No browser compresses what it sends to CITY_NET, so the server now answers such a request with 415 instead of decompressing it, whatever Node it runs on.
  • Manual installs need a fixed Node. setup.sh and setup.ps1 now accept only Node.js 24.22.1+ or 22.23.4+ (they asked for 18 or newer), and the server says at start when the Node it runs on lacks the fixes. No exploitation in the wild had been reported when this was released.

PHP's flaws of the same week (PHP2Shell, CVE-2026-109618 and -109619) don't touch CITY_NET: nothing in it runs PHP.


[1.14.9] - 2026-10-15

Dev builds reach installs set up before 1.8.1.

Fixed

  • Dev builds on an older docker-compose.yml. An install set up before 1.8.1 still has a docker-compose.yml that names its images :latest (an in-app update replaces the images, never that file). On one, ticking LOOK FOR DEV BUILDS changed nothing that mattered: every update downloaded and restarted on the stable release again. Now, when the server is on dev builds and its compose file doesn't follow IMAGE_TAG, ticking the box or updating changes just its two CITY_NET image lines to follow it, keeps the original beside it (docker-compose.yml.before-dev-builds-<time>.bak), and touches nothing else in the file.

Under the hood

  • The real-Docker updater check starts from an old install. It now begins with a compose file from before 1.8.1, as many real installs have, and recreates a real install's state: dev builds turned on, the server restarted, still on the stable release. From there one in-app update must land on the dev build. It failed exactly as the real install did before this fix, and passes with it on Docker Desktop for Windows.

[1.14.8] - 2026-10-15

In-app updates work one after another.

Fixed

  • A second in-app update in a row failed. After one successful in-app update, the next one stopped with "read /tmp/docker-compose.yml: is a directory", on any OS. The update recorded the compose file under a name that only existed inside its own helper, and the next update looked for it there. Updates now find the compose file where it really is, and record it there.

    If an in-app update has already failed with "is a directory", update to 1.14.8 by hand once: in your CITY_NET folder, docker compose pull and then docker compose up -d. If yours hasn't, the update button installs 1.14.8 as usual.

Under the hood

  • The updater is tested on real Docker. A new check builds the real server image, starts a throwaway install from the real docker-compose.yml, and runs two in-app updates in a row and dev builds on and off, checking the data survives each one. It runs on every pull request, and by hand on Windows before a release. It fails on the previous updater exactly as a real install did.

[1.14.7] - 2026-10-15

In-app updates and LOOK FOR DEV BUILDS work on Windows.

Fixed

  • CLICK TO UPDATE and LOOK FOR DEV BUILDS on Windows. On a server running Docker Desktop for Windows, both failed: the app handed Docker your CITY_NET folder by its Windows name (X:\CITY_NET), and Docker reads the colon after the drive letter as a separator, so it refused. LOOK FOR DEV BUILDS said "The channel could not be changed. Nothing was switched." The app now gives Docker the same folder in a form it accepts (/x/CITY_NET). Linux and Mac servers are unchanged. Checked on Docker Desktop for Windows: the switch to dev builds, and an update restarting the stack on the same data.

    If your server is on Windows, update to 1.14.7 by hand once, since the update button in your current version is the part that's broken: in your CITY_NET folder, run docker compose pull and then docker compose up -d. After that, updates work from the app.


[1.14.6] - 2026-10-14

Development builds, from inside the app.

Added

  • LOOK FOR DEV BUILDS. A box under "Check for update" in SYSTEM_INFO, for the main admin on a Docker install. Ticking it asks first, makes a copy of your database beside it (city.db.before-dev-builds-<time>.bak), and sets IMAGE_TAG=dev in backend/.env and in the .env beside docker-compose.yml, changing only that line. "Check for update" then offers the newest development build. Unticking sets it back to latest. Before this, it meant editing both files by hand and pulling from a terminal. Where the app can't change them (an install without the Docker socket), the box says which line to set by hand.

Fixed

  • The compose file's note about development builds no longer mentions a DEV=true setting that no longer exists; IMAGE_TAG alone chooses.

[1.14.5] - 2026-10-02

Database backups stay out of the server image.

Security

  • Database backups no longer go into a Docker image built from your own copy. The server image left out the database itself, but not the copies beside it: city.db.bak and the backups made before a migration were built into an image made on your machine, and with them every account, sheet and bank. Those, and the backend's test files, are now left out. The images published to Docker Hub are built from the repository, which never held them, so they were never affected. Git now also ignores the before-migration backups.

[1.14.4] - 2026-09-29

NPC sheets and hidden faces stay with the GM.

Security

  • NPC sheets no longer go to every player. The map list sent each player the whole sheet of every NPC linked to a token: stats, description, GM notes, everything. It now only goes to the GM, or someone the GM has given editing rights, who can open those sheets anyway. The GM's initiative rolls still read it.

  • A silhouetted NPC's face is actually hidden. The silhouette was a filter on the player's own screen, with the real portrait one right-click away. Players now get no portrait at all for a silhouetted NPC, and the token window shows the token in its side's color instead. The GM still sees the silhouetted portrait.

  • Initiative no longer shows a silhouetted face. Rolling enemies or friendlies, ADD TO INIT and ROLL INIT on an NPC sheet all put the portrait in the tracker, which every player sees. A silhouetted NPC now enters with the NPC marker instead.

  • The shop's old buy and sell messages are gone. Since the cart, every purchase and sale goes through one checkout, but the server still answered the two older one-item messages. Nothing in the app sent them, yet they still moved money for anyone who sent one by hand. Their tests now run against the checkout, so every case they covered is still checked.


[1.14.3] - 2026-09-29

Deleting a map-sized city.

Fixed

  • Deleting after generating a huge city no longer crashes the server. A city generated across the whole map could not be deleted: the server stopped with SQLITE_FULL, and every delete after that failed the same way. Large deletes now go through in pieces, still all or nothing, and a delete that fails reports an error instead of taking the server down.

  • The undo history no longer grows forever. It kept a full copy of everything ever deleted, which is what filled the database. It now keeps the last 50 actions. A change too large to keep a copy of, like deleting or regenerating a whole city, is recorded as too large to undo: UNDO says so once, and the next UNDO reaches the change before it.

  • A database that already filled up recovers on restart. Update and restart, and the oversized history is trimmed on boot. Nothing else in the database is touched.

  • Undoing a map-sized city generation works, where it used to fail on the number of buildings.


[1.14.2] - 2026-09-25

The shop, on the terminal layout, with a cart.

Added

  • A cart in every shop. + CART on the BUY or SELL list puts things in a CART folder, which blinks until you open it and shows how many things are in it. Open it and press CHECK OUT to buy and sell everything at once.

  • The cart is itemized. Something bought twice is one line with ×2, and − / + / ✕ to change it. Each thing sold is its own line, shown as minus money. Installed cyberware says so on its line: it comes out with no surgery roll. CLEAR CART empties it. Closing the shop, or opening another one, starts a fresh cart.

  • One total, in words. "YOU PAY 420cr" or "THE SHOP PAYS YOU 90cr", with your balance after. Your bank moves once, by the difference between what you buy and what you sell.

  • What you'd be carrying (CWN). The cart shows your Readied and Stowed load after checkout, and turns red past your limit. It only stops the checkout when the encumbrance house rule is on.

  • A receipt. After checking out, the cart shows a receipt: the shop, the date and time, what was bought and sold, the total, where the money went and your new balance. It goes away when you open BUY or SELL.

  • A filter on the SELL list, like the one on BUY.

  • Windows behave like a desktop's. The window in front is the focused one and keeps the solid title bar; the others dim to an outlined bar. Click anywhere in a window to bring it to the front. Esc closes the window in front, except while you are typing in a field.

  • Windows remember where you left them. Each kind of window opens where you last dragged it, in your browser: every shop in one spot, every token window in another.

  • A boot screen. Opening the app shows a BIOS-style self test typing out before the login window fades in. It shows once per browser tab, and the SKIP button in the corner skips it. If the page has been clicked, you hear the PC speaker's beep and the drive clicking as each line prints (browsers keep a page silent until it is clicked).

Changed

  • The shop window looks like the building and token windows. The building in the corner, BUY / SELL / CART down the left, and the shop line (type, lines, your balance) on top. The shelf tabs and the filter stay at the top of the list while it scrolls. The window is as tall as what it holds. Prices and ranges no longer wrap in the middle.

  • Checking out is all or nothing. If any line can't go through, nothing is charged and nothing leaves your sheet, and the cart stays as it was. Short of credits with overdraft allowed, you're asked once for the whole cart whether to take debt or go negative. If a price changed while things sat in the cart, nothing is charged and the new total is shown to check out again.

  • BUY and SELL no longer charge or pay on their own. Everything goes through the cart.

  • Every window has the solid title bar the building, token and shop windows have: the theme's color with the window's name in black. The update notice matches.

  • Windows are named like programs: SHOP.EXE · VIC'S ARMS, BANK.EXE, SHEET.EXE · GHOST, ID.EXE for a token, SITE.EXE for a building, and so on.

  • The startup sound plays as the login window appears, after the boot, instead of on logging in. The ambient hum eases up slowly after the boot, or quickly on a refresh, and is a quarter quieter than before.

  • The login screen has the CRT scanlines over it, like the rest of the app.

  • The version on the login screen sits below the THEMES button instead of beside it, where it was starting to run into it.

Fixed

  • Buying two of the same weapon at once can no longer put both in one row. Bought things are placed one at a time after checkout, each seeing the one before.

  • A checkout the server never answers stops waiting after 15 seconds, keeps your cart, and says to check your balance before trying again.

  • Moving the volume slider no longer restarts the ambient hum. It changes its volume in place, and muting and unmuting pick it up where it was.

  • After a refresh, the hum starts on your first key press or click. It used to stay silent until you muted and unmuted, and never started at all if you logged in with the keyboard. (Browsers keep a page silent until you touch it, Firefox on every load.)

  • A big balance change no longer flattens the bank's chart. A 150,000cr sale used to jump the ticker by about 1,800 and squash every other candle into a line for a minute. Big changes now make a tall candle, and small ones still a small tick.


[1.14.1] - 2026-09-25

The building and token windows, redone as terminals.

Added

  • A new building window. Clicking a building opens a terminal-style window: folders down the left, the open folder's text on the right, and the building in the top corner. INFO holds the description, headed with the shop type and district. RESIDENTS holds the NPC list. Switch folders by clicking or with the up and down arrow keys. The buttons along the bottom (SHOP, ENTER BATTLE MAP, BROADCAST PING and the rest) are the ones each viewer had before, and wrap onto a second row when there are many. Line breaks in a description or residents list are kept now, instead of running together.

  • The building, turning, in the corner of its window. It's drawn from its own parts as a wireframe in your theme's color. If this machine can't draw 3D, the CITY_NET badge shows instead.

  • Building photos. A GM can give a building a photo, and everyone then sees it in place of the turning building. Upload it in the admin edit view under BUILDING PHOTO.

  • GM notes on buildings. A GM NOTES folder, visible only to the main admin, holds what the players shouldn't know. Players never receive the notes at all, and neither do players who have been granted editing rights. Edit them in the folder itself, or in the admin edit view.

  • UPDATE_DATA_POINT saves the photo and GM notes too. In the admin edit view, choosing a photo, removing one or typing notes is saved along with the rest of the building. If any of it fails, the alert says what failed instead of CHANGES_SAVED.

  • A new token window. Clicking a token opens the same terminal-style window, with the portrait (or a token in its side's color) in the corner. INFO is headed with the player's name and starts with their handle and role. HEALTH is where health lives now. The buttons along the bottom are the ones each viewer had before.

  • HEALTH on your own token, and on any token for the GM: the numbers, HEAL, DAMAGE, TEMP HP, the injury map and STIM_HEAL in CWN, under a live heart monitor. The GM also gets MAX HP, the token's AC or DV to edit, and ADD TO INIT for an NPC without a sheet.

  • HEALTH on anyone else's token is a picture, not numbers. A live heart monitor that beats, changes color as they get hurt and flatlines when they go down, plus the injury map. Temp HP shows as a shield, and Shadowrun stun as a bar, with no figures.

  • QUICK ACTIONS on your own token. Your AC or DV, and rolls straight off your sheet without opening it: CWN's saves, CP:R's stats, Shadowrun's initiative and composure, and a skill picker for every system. They are the sheet's own rolls, with the same penalties, and land in the dice tray.

  • GM notes on NPC tokens. The same GM NOTES folder buildings have, on enemy and friendly tokens, for the main admin only.

Changed

  • Clicking another player's token opens its window. It used to open a separate health window. Now it is the token window, with INFO and HEALTH, the same as an NPC.

  • HIT_POINTS opens the token window on HEALTH. A player gets their own token. A player who hasn't placed one gets a CREATE HEALTH RECORD button there. The GM gets the selected token. The sheet's linked HP field goes to the same place.

  • The token's remove button says REMOVE_MY_TOKEN or REMOVE_TOKEN, like the sidebar, instead of PURGE_DATA_POINT.

  • Saved maps carry the GM notes. Loading a map brings back the notes written for its buildings, and never leaves one map's notes on another map's buildings.

Removed

  • The token window's ID and COMBAT tabs. Handle and role moved to INFO. An attack you are setting up shows as a line above the buttons. The last attack's result is no longer repeated there: the dice tray and the map already show it.

  • The UPDATE_HEALTH and CHECK_HEALTH buttons, and the separate HIT_POINTS and health windows. All of it is the token window's HEALTH folder now.

Fixed

  • Loading a saved map lost every building's shop type and buy-back rate. It also lost their AC, sidewalk and signage settings, and whether they were hidden. A loaded map now brings back everything each building was saved with.

  • Undoing a building delete brought it back without those settings too. It now restores everything the building had, and its GM notes are still there.


[1.14.0] - 2026-09-24

Shops open in every game system.

Added

  • Shops work in Cyberpunk RED, Shadowrun and generic games. Building types, the buy-back rate and SHOP_CATALOGUES are now available in every system. Outside Cities Without Number a shop sells only what its GM uploads; the CWN book's items stay in CWN.

    Bought items land in that game's own sheet rows. A Cyberpunk RED gun keeps its ROF, and a Shadowrun gun gets its DV and AR. Cyberware goes into the cyberware table on sheets that have one (CWN and Cyberpunk RED) and into the inventory on those that do not. A purchase with nowhere to go, such as a full weapon rack, is refused before any money moves. Selling clears exactly what that sheet's row holds, so nothing is left half-sold.

  • Shop names that fit the game. A Ripperdoc is a Street Doc in Shadowrun and a Cyber Clinic in generic games, and outside CWN "Operator Gear" is just Gear. The building types themselves are the same everywhere, so shops stay where they are when you switch systems.

  • A new shop tells the GM how to stock it. Set a building to a shop that has nothing to sell and the edit window lists the steps: open SHOP_CATALOGUES (an OPEN IT button goes straight there), download the example, fill in the sections this shop sells, then upload it or type the lines into the box, preview and save. It shows the exact header line for that shop's section in your game. A GM opening an empty shop sees the same steps; a player is told the GM adds the stock.

Changed

  • The buy-back rate says "shelf price" outside CWN. It read "% of book price" in every game, and only CWN has a book. The SELL tab says the same.

  • An empty shelf says so. It used to read NOTHING MATCHES THAT, as if a filter had found nothing. It now says the shelves are empty and that the GM adds stock in SHOP_CATALOGUES.

Fixed

  • Players never saw the items a GM uploaded after loading the page. The window asked for them before the player had signed in, the server ignored the request, and nothing asked again. In CWN the built-in items hid it; in any other game every shop looked empty. They are now sent as soon as a player signs in.

  • Selling a CWN vehicle left half of it on the sheet. The sale cleared mount columns a vehicle does not have and missed the ones it does, along with its fittings. Selling now clears exactly the fields the sheet draws for that row, read from the sheet itself.


[1.13.0] - 2026-09-24

Shops that take your money, and buy your old gear back.

Added

  • Every priced table in the Cities Without Number book now has a storefront. Armor (p53) and Common Operator Gear (p50) had never been transcribed at all, and now are. Two new shop types sell them, the Armorer and the General Store. The Garage sells vehicles, fittings and vehicle weapons, and the gun shop and armorer sell the mods that go on what they stock. A shop that carries more than one table shows a tab for each.

  • Buying charges your bank account. What you buy only turns up on your sheet once the payment has gone through. The price comes from the server rather than your screen, so the number you see is the number you pay. Each shop shows your balance, in red when it is below zero, with any debt beside it.

  • New house rule: BUY WITH MONEY YOU DO NOT HAVE. It is off by default, which refuses a purchase you cannot afford. With it on, you are asked how to cover the shortfall: take debt, go negative, or cancel. Nothing penalises a negative balance automatically; that is left to the GM. The rule applies to every game system.

  • Selling. The SELL tab lists what you own that this shop deals in: a gun shop buys guns, a ripperdoc buys chrome. Add things to a sell list, check the total, and confirm once. What you sell leaves your sheet, the money reaches your bank, and a line vanishes when you sell the last of it. It covers inventory, weapons (carried or stashed), vehicles, and cyberware.

    Selling installed cyberware warns you first. The book calls for surgery and a complications roll to take chrome out, and CITY_NET does not model either. So an installed piece is flagged on its row, in the confirmation and on the receipt, with a reminder to square the surgery with your GM. KEEP IMPLANTS takes the installed pieces off the list and leaves the rest of the sale alone. If you own a spare boxed piece and an installed one, selling one sells the spare.

    Items no catalogue recognises, such as renamed rows, homebrew and quest items, sell for nothing, and the GM settles up directly.

  • Buy-back rate. What a shop pays for second-hand goods, as a percentage of the book price. Set it for the whole game in the GAME tab under AWARD_EXPERIENCE (45% by default). Give a single shop its own rate where you set its building type. Leave that box blank to use the global rate; setting it to 0 makes a shop that buys nothing.

  • GMs can add their own items to the shops. SHOP_CATALOGUES, in the GAME tab under the buy-back rate, opens a window where you can:

    • download an example file shaped for your game system, with a few real rows to copy
    • download the current list of everything already on sale
    • paste a table or upload a file, then preview it: rows read, problems by line number, and anything that would replace a built-in price
    • save it

    Your items are added to the ones that come with the app and never replace them. A row with the same name as a built-in item house-rules that one price, and the preview says so first. In the downloaded current list, the built-in rows are commented out with #, so uploading it again unchanged changes nothing. Delete a row's # to override it. CSV, tab-separated and JSON are all accepted, and so are names containing commas, Excel's semicolons and a leading byte-order mark.

Changed

  • Dropdowns across the whole app now match your theme. Every one of them was drawing the browser's own grey control. Now they follow whichever of the seven themes you use.

  • Shop notices say where a purchase lands, not that BUY costs money. For example: a weapon goes into a weapon slot stowed, an augment arrives owned but not installed, and armor goes into your inventory without changing your AC. The line only turns amber when something is actually wrong.

  • Shops are still Cities Without Number only, on purpose. Uploaded catalogues work for any system. Buying and selling, though, still assume a CWN sheet: a Cyberpunk RED weapon would be written with the wrong fields and lose its ROF. The note at the shop gate says what has to change before it opens to the other systems.

Fixed

  • Anyone connected could move money in anybody's bank account. Withdrawing, borrowing and paying off debt all acted on whichever account name arrived with the request, and nothing checked that the sender owned it. Each now acts only on the signed-in player's own account. Nothing changes for legitimate use, because the bank window only ever asked about your own account.

  • Selling updated only the window you sold from. If your sheet was also open in its own browser tab, that tab kept showing the item you had sold. It now updates everywhere.


[1.12.1] - 2026-09-17

A structure now saves at the angle you actually turned it to.

Fixed

  • Rotating a structure on more than one axis saved the wrong angle. Tilt a building and spin it, save, and it came back facing somewhere you never put it. Measured at just over 22 degrees out for a moderate turn.

    Rotations are stored as three numbers, and three numbers only describe an orientation alongside the order they are applied in. Everything that draws a structure reads them in one order; the editor was writing them in another, so the same figures meant a different facing. Saving now goes through the orientation itself rather than the three numbers, so what is written is what will be read.

    Turning a building on its vertical axis alone was always correct, which is why this went unnoticed: it is what almost everyone does, and it is the one case where the two orders cannot disagree.

    Structures you have already placed are left exactly as they are. They keep their stored numbers and draw precisely as they do today. Rewriting them would have quietly rotated whole cities on a guess about what was originally meant. Re-saving a structure corrects it, so a building that has always looked slightly wrong can be nudged and saved to put it right.

Changed

  • The test suite now catches a whole class of break before it reaches CI. A frontend test may borrow a backend module - that is how the rules stay mirrored in both languages - but the frontend installs only its own dependencies, so borrowing one that needs an extra package passes locally and fails the build. That is now checked, naming the file and the package rather than failing somewhere confusing half an hour later.

[1.12.0] - 2026-09-15

Cities Without Number, finished. Every gap the rulebook audit found is closed: armor that actually soaks, the two Armor Classes the book prints, three tables of modifications, a countable inventory, encumbrance, experience, the drug table, skillplugs — and an import form that finally asks for all of it.

Nothing on an existing sheet changes meaning. Every new field is blank on sheets written before it, and blank keeps reading the way it always did: a weapon with no Melee AC defends at its ranged value, a weapon nobody has filed as Readied or Stowed is neither, and an armor with no Damage Soak soaks nothing. Two free-text boxes are retired rather than deleted — see Changed.

Added

  • Armor soaks damage before you do. The book's Damage Soak column was printed on every armor and modelled nowhere, so an armored character was over-damaged on every hit — up to 15 points a scene in a heavy suit. Damage now comes off the soak first, with a REFILL button on the pool, because only the table knows when a scene ends.

  • Armor has two Armor Classes, because the book gives it two. A War Harness is 13 against ranged and 14 against melee. The sheet had one number and wrote it to both, so whenever they differed one of your defences was wrong and the app was deciding hits with it. Shields split the same way.

  • Three tables of modifications, all applied rather than printed. All 11 armor mods and 13 weapon mods (p58-59), and all 10 cyberware mods (p71) fitted per implant. Everything is overlaid when a roll is made and never written into the sheet, so taking a mod off gives back exactly what it gave. The +3 ceiling on hit and damage is enforced, and the line under each list says when you are over it and what will actually be rolled.

  • Weapons have their own attribute, and body weaponry is a weapon. Str, Dex, Str/Dex, Wis or none, with a pair taking whichever is better at the time. Cyber blades carried a name and a strain cost and no stats, so a character who had paid for them attacked as if they had not; they now appear in the attack picker with the damage and skill they will really roll.

  • A character's Move rate is worked out and shown. The book does not derive Move from an attribute, which is worth saying because everyone assumes it does — it is a flat 10m, and exactly one implant adds to it.

  • Experience, and a way to hand it out. An EXP bar under the HP bar, filling from what this level cost toward the next. AWARD_EXPERIENCE in the admin panel gives points per character rather than splitting a pot, since dividing experience would mean a full party earned less each than a pair. The advancement rate is a house rule, not a per-character field: one table advances at one pace.

  • Encumbrance, counted always and charged only where asked. A READIED/STOWED line at the top of GEAR on every sheet, and a house rule — off by default — that turns the Move penalty on. The book is explicit that these rules are optional, so the count is shown either way.

  • A countable inventory, on every system. Structured rows with quantities, replacing the free-text Gear box that could describe ammunition but never add it up. Cities Without Number gets an Encumbrance column and the book's bundling rule, so six stims cost 2 rather than 6.

  • Weapons come from the catalogue, with a stash for what you are not carrying. 32 weapons with their real damage, trauma, shock and Encumbrance. The stash holds what you own but have not got on you: no Encumbrance, a location note, and one move each way.

  • Shops sell things. The gun shop stocks weapons and the clinic stocks the drug table — no charging, the same way the ripperdoc works. Sortable headers, an OWNED column that reads your sheet, and a refusal that says why when a weapon will not fit.

  • The whole pharmaceutical table (p60-61), and the three drugs that change a roll. Boneshaker, Olympus and Avalanche are modelled; the other thirteen carry the book's own sentence for the table to rule on. Doses are inventory rows, CONSUME is on the row, and what is running sits in the sheet header where it follows you onto every tab. Multiple drugs take the highest bonus and pay every price, which is the book's own disincentive.

  • Skillplugs (p64), and the price the book attaches to them. A plug grants a skill while it is loaded, taken as the better of the plug and what you earned. The cost is enforced: the worst the dice can do becomes an automatic failure no reroll can save, and the jack locks up for the scene. Each plug past the first widens that band.

  • Languages, and the allowance that pays for them. Thirty-four real languages grouped by region, plus typed entry — the two a character starts with are their city's tongue and their enclave's, both invented per campaign, so no fixed list could ever hold them.

  • Import a Characters Without Number export (characterswithoutnumber.app). An additional way in, beside the fillable PDF, the Companion code and a pasted stat block — none of which change. Brings attributes, skills, weapons, chrome and inventory; hit points, AC and cash live elsewhere in this app and are reported rather than written.

Changed

  • The downloadable import form now asks for what the sheet holds. It went from 103 boxes to 332, and two of the gaps were much older than this release: it had never printed a SKILLS section at all — nineteen skills, no boxes, while the importer had understood every one of them since it was written — and it offered four weapon rows for a sheet that holds six. Both now come from the same source the rest of the app uses, so neither can fall behind again.

  • Two free-text boxes are retired rather than deleted. The Gear box and, on Cities Without Number, Weapon Notes. Both still show whatever was typed into them and disappear once emptied, so nothing is lost and nobody new is handed a box that something else replaced. The importer and the printed form still accept both.

  • A raised attribute moves everything hanging off it. Modifiers, saving throws, the System Strain maximum and every skill that keys off it.

Fixed

  • Derived fields were computed when a sheet was saved and never when it was read, so a sheet last saved before a field existed carried nothing for it — and an empty number reads as 0, which is a wrong answer rather than an absent one. Every character read MOVE 0.

  • No number field on any sheet could be given a negative by typing. Every system, every sheet; it only surfaced on MOVE MOD, where negatives are the point.

  • A Cities Without Number rule was discounting Cyberpunk RED's Humanity Loss. One system's table had reached another's numbers.

  • Installing cyberware charged the undiscounted strain, so a piece the sheet had just said there was room for was refused.

  • The test suite failed about one run in three under load, somewhere unrelated to whatever was being changed. Fourteen files slept a fixed number of milliseconds waiting for database work and then asserted. They now wait for the work itself, which is both correct and faster.


[1.11.6] - 2026-09-02

A clearer way to put buildings in districts, and friendly NPCs a GM can let players move.

Existing districts are untouched. Nothing about this release changes what is already in your database: the districts you have, their colors, and every building assigned to one all come across as they are. The only schema change is a new column for NPC control, which is empty on every existing token and means "admin only" — exactly how they behaved before.

Added

  • Districts have an ASSIGN button of their own. Creating a district and choosing what goes in it used to be the same screen, so it was never clear which you were doing. They are separate now: ASSIGN picks structures on the map, EDIT changes the district itself. The list explains the flow and shows how many structures each district holds.

  • Districts can be renamed and recolored. Neither was possible before — the only route was delete and recreate, which dropped every assignment. Both carry the buildings with them, and a rename onto a name another district already uses is refused rather than quietly merging the two.

  • The edit screen lists what is in the district. Each structure by name, with its own REMOVE. A building with no name of its own shows the label the rest of the panel gives it rather than a blank row, and hovering one lights it up on the map so you can see what you are about to take out. The list stops at five and scrolls, so the controls under it stay on screen.

  • A GM can give players control of a friendly NPC. Specific players, everyone, or both, set from the NPC's own panel while placing it or any time after. The GM never loses control: this shares a token rather than handing it over. Friendly NPCs only — enemies stay with the GM.

Changed

  • Assigning structures to a district only ever adds. Saving used to clear the whole district and re-add whatever was selected, so the selection was the district and a stale one silently unassigned buildings with nothing to undo from. Only the structures you picked are touched. One already in another district moves, and the panel says how many will before you save.

  • Deleting a district says what it will cost. The confirmation names how many structures will be left with no district.


[1.11.5] - 2026-08-30

Cyberware for Cities Without Number, shops on the map, and two fixes to the initiative tracker.

Added

  • Cities Without Number has cyberware. The section was a single free-text box; it is now the same rows and body diagram Cyberpunk RED has, with the book's five install types, System Strain per piece, price and concealment. The old notes box is kept as a section of its own, so anything already typed there is still on the sheet.

  • All sixty implants from the book, as a catalogue. Adding a piece no longer means copying six fields off a page: pick it from the list and the form fills itself, modifiers included. Everything stays editable, because a piece can be salvaged or house-ruled and the book entry is a starting point rather than a contract.

  • Buildings can be shops. An admin gives a building a type, and the ones that trade get a SHOP button in its info window. BUY puts a piece on your sheet as an unplaced row and stops there: no money moves and no stock is kept yet, which the window says out loud. Buying and selling are separate tabs, since selling reads what you are carrying rather than what the shop stocks.

  • An END button on every combat in the initiative menu. Combats running elsewhere on the map can be closed from the list without navigating to them first. A combat spanning several scenes ends everywhere in one press.

  • Trauma Target and the System Strain maximum are now worked out for you. Trauma Target is 6 plus your armor's mod, and cyberware raises it further. The strain maximum is your CON plus a LIFESTYLE modifier, which the book sets by where a character lives.

Changed

  • Reordering the turn order no longer drags the current turn along with the character. Moving whoever is up next used to carry their turn down the list with them, so nobody else advanced. The turn now belongs to the position: move yourself down the order and whoever takes your place inherits it.

  • Installing cyberware is refused when it would exceed your System Strain. The book allows as much chrome as your maximum strain covers, and nothing was checking. The refusal says what the piece needs and how much you have free. Owning a piece still costs nothing; only fitting it does.

  • A raised attribute now moves everything that hangs off it. Cities Without Number rolls skills off attribute modifiers rather than the attributes themselves, so an implant that raised DEX and stopped there changed no Dex roll. Modifiers, saves and the strain maximum now follow.

Fixed

  • Cyberware effects reached only some of what the book gives them. Modifiers were read from each implant's one-line summary, where the descriptions beside them carry the actual numbers - the Aesthetic Augmentation Suite says only "Cha bonus" in the table and "a Charisma score of 14" in its description. Ten pieces gained the effects they always should have had.

[1.11.4] - 2026-08-28

Stronger Secure Mode.

Changed

  • Secure Mode checks accounts more strictly when a connection joins. The account check now runs against fully verified credentials rather than partly trusting what the connecting client says about itself, so it holds in cases where it previously could be sidestepped. Recommended for anyone running with Secure Mode on. Nothing changes day to day: admins, approved players and temporary admin access all work exactly as before, and servers running with Secure Mode off are unaffected.

[1.11.3] - 2026-08-28

Security patch for the realtime connection.

Fixed

  • A player could exhaust the server's memory through the realtime connection. The websocket library underneath Socket.IO mishandled floods of very small message fragments, so anyone able to open a connection could push the server into swapping or an out-of-memory kill. Nothing about how the connection behaves has changed, and a browser tab left open across the update reconnects on its own.

Changed

  • The realtime stack moved to patched releases. engine.io, socket.io-adapter and socket.io-parser on the server, engine.io-client and socket.io-parser in the browser, all patch-level and all still speaking the same protocol version. Only the lockfiles changed; no dependency was added, removed, or moved to a new major.

[1.11.2] - 2026-08-27

Custom signs.

Fixed

  • The move and rotate gizmo works on every sign, not just the last one added. Selecting an earlier sign left it with no gizmo at all, so the only sign you could reposition was the newest one. Every sign was reporting itself to the same slot on each selection change, and the ones later in the list overwrote the one you had actually picked.

  • Structures stay put while you are editing signs. Clicking near a sign no longer selects the building behind it and swaps the panel out mid-edit. Signs hang flat against whatever is behind them, so aiming at one and catching the structure underneath was easy.

Changed

  • Signs live in their own module. The meshes, the procedural signage, the editor and the state behind placing them were spread across three files including a 2500-line one; they are together now, in the same shape the initiative tracker uses. Nothing about how signs are stored or loaded changed, so signs already placed are exactly where they were.

[1.11.1] - 2026-08-27

The themes, reaching the parts they were missing.

Fixed

  • Structures stay put while you are editing signs. Clicking near a sign no longer selects the building behind it and swaps the panel out mid-edit. Signs hang flat against whatever is behind them, so aiming at one and hitting the structure underneath was easy.

  • The move and rotate gizmo works on every sign, not just the last one added. Selecting an earlier sign left it with no gizmo at all, so the only sign you could reposition was the newest. Every sign was reporting itself to the same slot on each selection change, and the ones later in the list overwrote the one you had actually picked.

  • Themes reach the text that was only inheriting a colour. Labels, headings and the readouts beside sliders stayed green on every theme while the values next to them changed. The base text colour was set on body, which sits outside the element the theme class is on, so it always resolved to the default green. Anything that set its own colour was already fine, which is why half of each line themed and half did not.

  • A new Vaporwave. Cream, two pinks and a purple, over a ground that drifts toward indigo as it darkens rather than staying the same magenta — which is what stops the pinks sinking into it. Alerts moved with it: the cream is a yellow, so the old amber warning would have read as ordinary text.

  • Input boxes, panels and tabs stop being green-tinted. Every translucent surface in the app was a fixed green wash, so a text box on Crimson sat on a green ground and the selected tab glowed green whatever theme was on. They take the theme's own dark ground now: red on Crimson, purple on Vaporwave, blue on Ocean.

    Modal backdrops are untouched. Those are a neutral black scrim by design rather than a tint that got missed.

  • Checkboxes and sliders take the theme's colour. All thirty-nine of them drew in the browser's own blue, which belonged to no theme.

  • Danger and warning are real colours now, tuned per theme. The palette had neither, while the code asked for --danger and fell through to a hardcoded red every time.

    Two themes needed them moved rather than merely defined. Crimson's own text colour is the exact red the code used for danger, and Solar's is the exact amber it used for warning — so on those themes a warning has been rendering the same colour as ordinary text, invisible as a warning rather than merely off-theme.


[1.11.0] - 2026-08-27

Cyberware, as something you can actually see.

Added

  • An augmentation window. A body diagram with your chrome arranged around it, wired to where each piece is installed: eyes, ears, neural link, arms, legs, and the categories that are not a place — Internal, External, Fashionware, Borgware. Open it from the CYBERWARE section of the GEAR tab.

    Underneath is the same list as a table you can sort by name, type, humanity loss or price. Two views because the questions differ: what is in my left arm is a diagram question, and what am I running and what did it cost is a table question. No single ordering answers both.

  • Say what a piece is without saying where it goes. The TYPE column in the table is a picker, so an imported piece can be marked a Cyberleg while you decide which leg. Those are two decisions and only the second one needs the diagram — and since an import carries no install location, every piece arrived Unfiled with no way to describe itself until someone dropped it on a limb.

    Once typed, it is offered first when you press + on a matching body part, instead of the list guessing from its name.

  • Put chrome you already have into a place on the body. Pressing + on an arm asks what to put there before offering a blank form, listing everything not yet placed — with the pieces whose names read like a match for that spot first, and marked. It is a suggestion rather than a decision: an imported piece knows only its name, so nothing files itself into the wrong arm on the strength of a word.

  • Taking a piece out again. Each entry in a body part has a − beside it that unplaces the piece without deleting it, so it goes back to the unplaced list and can be filed somewhere else. Deleting it outright is the × in the table — uninstalling chrome and never having owned it are different things.

  • Cyberware you can add by hand. Name, install type, humanity loss, price in eddies, and what it does. Adding from a panel fills the location in for you.

    It never asks which side. The same cybereye fits either socket, so left or right is a fact about where you install it rather than about the piece — you answer it by pressing + on the arm you mean. A paired piece added from the general form waits in the unplaced list until you put it somewhere.

  • Your chrome reaches your numbers. A Cyberpunk RED sheet counts its equipped cyberware: a skill of 3 with a +2 augment reads 5 on the page, with a badge saying so and a tooltip naming the piece responsible. The same total is what the dice use — the server applies it when it resolves the roll, and it appears in the breakdown as its own line rather than hiding inside a stat.

    The number you typed is still the number stored. Nothing is written back into your stats: take the chrome out and the stat is what it always was, and no amount of saving compounds the bonus.

    Two things worth knowing. A piece that sets a value is applied before pieces that adjust one, so setting COOL to 3 and adding 2 leaves 5 — the other order would let an augment do nothing at all. And an augment naming something this sheet has no field for is reported rather than silently ignored; the Companion's "Combat #" is the one that comes up.

    Cyberpunk RED only for now. The other systems record modifiers but do not apply them.

  • What a piece actually does, in numbers. A modifier is a stat, a skill or a roll type, and an amount — +6 Business, +2 Initiative — and a piece can carry as many as it needs. Setting a value is kept separate from adjusting one, since +3 Cool and Cool = 3 are different claims, and the table shows them differently.

    You pick what it changes and then pick the thing itself from a list of your system's stats and skills, read off the character sheet rather than typed in. A Cyberpunk character is offered Cyberpunk skills and a Shadowrun character Shadowrun ones, and the amount column says BY or TO depending on which you chose.

    These come across from a Companion import. They are the one mechanically real thing an export carries: the descriptions arrive blank because the Companion draws its flavour text from its own catalogue, but a modifier is something you typed into your own character, so it is really there and it comes with you.

  • Your chrome is on the sheet properly now. It was a single line of free text. It is a list, and the number of pieces is a property of your character rather than a number chosen in advance — a full borg does not run out of rows.

  • The printed import form has a cyberware table, twelve lines with the same columns. Paper has to stop somewhere; the app does not.

Fixed

  • Importing a character brought back none of its cyberware. A character with eight pieces of chrome imported zero. Every entry in a Companion export leaves the name blank and puts the identity in a type field, and the import read the name — so it found eight blanks and dropped them all.

    It now reads what the pieces actually are, and brings the humanity cost across with them, which was going to be typed in by hand.

    Two things the export has never carried, so neither can come across: where a piece is installed, and its description. The Companion shows that flavour text from its own catalogue rather than storing it on your character, so the fields arrive empty. Anything you typed there yourself does come through. Imported pieces are marked as not yet placed rather than being guessed into an arm.

  • Generated Cyberpunk RED NPCs come with chrome. GENERATE_SHEET fits cyberware to the tier — nothing on a mook, a piece or two on a skilled, a full set on an elite — already installed on the body rather than waiting to be placed. It is this app's kit, like the weapon rows, not a catalogue.

    Humanity pays for it, and current EMP follows from Humanity the way it does on every save, so a generated sheet does not claim full Humanity beside four implants.

  • A note modifier, for what the app cannot work out for you. Some chrome does something the sheet has no field for and no dice to roll. Written into the effect paragraph it gets missed at the table; as a modifier it is a chip everyone can see — Quickhack DV 10.

    You type both the label and the number, and the app never applies either. It is marked out from the modifiers that do change your rolls: dashed border, plain colour, and a tooltip saying so.

  • Edit anything on the cyberware list. A pencil beside each piece opens it in the form with everything filled in — name, type, humanity, price, effect and modifiers — so a mistyped name or a wrong effect can be corrected without deleting the piece and building it again. It keeps its place in the list and stays installed where it was.

  • TOKEN_PROTOCOLS is now CHARACTER_CONTROLS. The panel stopped being about the token a while ago: alongside placing and colouring it, it holds the character sheet, vehicles, enemy vehicles, health and armour. The name now says what it governs.

  • Setting a type no longer installs the piece. Naming something a Cyberleg in the list says what it is; it does not fit it. It stays in the unplaced list and is offered first when you press + on a matching body part, which is what naming the type is for. Fitting still happens on the diagram, which is the only place that knows where things go.

    This was most visible on the types with no left or right — marking something Fashionware installed it on the spot, and there was no way to own a piece you had not fitted yet.

  • Taking a piece out no longer forgets what it is. Unplacing a Cybereye left it with no type at all, so putting it back meant saying it was a Cybereye a second time. It now keeps its type and gives up only its side, which is the state a piece is in before anyone has chosen a socket.

  • The TYPE column shows the type, not the placement. Installing a Cybereye in the left eye made that column read "Cybereye L", which looked like installing it had rewritten the type you set. Which eye it is in is a fact about the body, and the diagram is where the body is.

  • Cyberware reaches your initiative. A cyberarm that raises REF now raises what you roll for turn order, and a piece whose modifier targets Initiative adds to it directly. Both previously showed on the sheet and reached every other roll while initiative alone read the unmodified stat. The chrome is named in the breakdown rather than folded into REF, so the number is explainable.

  • A modifier on Language (Streetslang) no longer lands on Language (Other). The two skills differ only inside the brackets, and the name matching stripped every bracket in order to cope with the "(x2)" that marks a double-cost skill. Only the cost marker comes off now.

    Found by testing every stat and skill the picker offers rather than one of each — the two Language skills were the only pair that collided, and nothing about the symptom would have pointed at the cause.

  • Cyberware only changes your numbers once it is actually installed. A piece you own but have not placed on the body no longer modifies a stat or a skill. The sheet was reading "0 INSTALLED" beside a stat one of those pieces had rewritten, which is the same fact contradicting itself — and since every import arrives unplaced, importing a character used to change their stats before anyone had said where anything went.

    Placed means placed: a piece typed as a Cyberleg but in neither leg is owned, not installed.

  • The sheet's cyberware summary counts what is installed, not everything on the list. It read "8 INSTALLED" over a body diagram with nothing on it.

  • The cyberware table's scrollbar matches the app, and keeps off the row buttons. That list scrolls itself rather than through the window around it, so it was drawing the browser's own scrollbar over the column holding the edit and remove buttons.

  • The character sheet's scrollbar stops covering the corner of a text box. The resize grip on a description or notes field sat four pixels from the bar, so reaching for it caught the bar instead. There is proper room now, and the gutter is reserved whether the sheet is scrolling or not, so the layout no longer shifts the moment a tab grows past the window.

    The sheet body scrolls itself rather than through the window around it, which is also why it had been drawing the browser's own scrollbar instead of the thin green one everything else uses. It matches now.

  • An edit made just before closing a sheet is no longer lost. Every change waits 400ms before being sent, so that typing is one save rather than one per keystroke. Closing the sheet inside that window used to discard whatever was waiting — the change was on screen and in local state and never reached the server, so it came back on the next load.

    It showed up as removing a piece of cyberware not sticking while adding one did: adding means typing, which always outlasts the delay, but removing is a single click and people close the window straight after. It applied to any field, not just cyberware.

Changed

  • Existing cyberware text becomes rows. Anything already written in the old box is read into the list on first start, one piece per comma. A parenthetical stays with its piece, so "Cybereye (Low Light)" is one entry rather than two. The original text is kept rather than deleted, in case the split read your notes badly.

[1.10.1] - 2026-08-22

One sheet, two people writing to it.

Fixed

  • A change to a character sheet could be silently undone by another change landing at the same moment. A sheet is stored as one block of data, so altering a single field means reading the whole sheet, changing that field, and writing the whole sheet back. Twenty-two places did that, and nothing stopped two of them overlapping — so the second one wrote a version of the sheet built before the first one's change existed, and that change simply vanished. No error, no warning: the hit points come back.

    Two different players were never at risk, since each write only ever touched that player's own row. The problem was one sheet with two writers, which is an ordinary moment at a table: a GM applying damage while that player edits their gear, an automatic effect landing mid-edit, someone with their sheet open in two tabs.

    Every writer now takes a turn per sheet. Nothing is rejected and nothing needs retrying — the second one simply reads after the first has finished, so both changes survive. Writers to different sheets never wait on each other.

  • Anything counted relative to its current value could come out wrong, not just missing. Spending LUCK, mage effort, system strain, the stun track and Edge grants were all computed from a copy of the sheet read earlier, so spending three LUCK while something else wrote the sheet could hand it back. These are now worked out against what the sheet actually says at the moment of the write. Two Edge grants arriving together are +2.

  • Damage to a vehicle could stop counting when two hits landed together. Both worked out the new hull value from the figure they read before writing, so both subtracted from the same starting number and one of them did nothing. Hull damage, stun drain and vehicle HP are now worked out as the write happens, so simultaneous hits add up.

  • A death save discarded five seconds of edits. The sheet was read, then the write waited out the client's dice animation before writing the whole sheet back — so anything changed while the dice were rolling was thrown away, at exactly the moment a table is busiest. Only the penalty is written now.

  • Importing a sheet could turn a player out of their vehicle. An import deliberately carries your seat across a replace, but it carried the version read before the import started — so being seated during the import undid it. Read at write time now.

  • A table-wide LUCK or Edge reset, or a CWN rest, could wipe an edit made while it ran. These scan every sheet and then write them all, so on a table of six the last write was built from a scan that had already gone stale. The scan now decides only who is affected; each value is worked out as its sheet is written.


[1.10.0] - 2026-08-22

Animated battle maps, and the second half of the security pass.

Two pieces of work that ended up in one release because they met in the same place: making animated maps usable meant reworking how uploads are stored and checked, which is where the security pass was already headed. Where 1.9.4 covered what the server says to the outside world, this covers what it does with the Docker socket and with files people give it.

Added

  • A battle map can be a looping video. Drop in a .webm, .mp4 or .m4v and it plays under the tokens — rain on the pavement, a flickering neon sign, water moving under a walkway. The animated map packs people buy work as they are; there is nothing to convert and nothing to configure.

    Maps can be up to 250MB now, raised from 25MB. An animated map is a video, and the loops people buy routinely run past the old ceiling — which would have made this feature something you could only use with files small enough not to be worth animating. The server writes uploads straight to disk rather than holding them in memory while it works, so a large map costs storage rather than a chunk of your server's RAM, and identical maps are stored once however many locations use them.

    It loops, and it is silent — every browser refuses to start a video with sound on its own, and a map that will not move until somebody clicks looks broken rather than deliberate. Playback stops while the tab is hidden, so a session left open overnight is not quietly decoding video at your laptop.

    Still maps take exactly the path they always did. Nothing about an existing map changes.

Security

  • Pressing Update twice started two updates. Nothing checked whether one was already running, so a double-click — or an impatient second press while the first was still pulling — ran two image pulls and then launched two privileged helper containers against the same stack, each recreating the containers the other was recreating. A second press is now refused with "an update is already running" rather than honoured. It is not a queue: two updates are not twice an update, and the run in flight is left alone to finish.

    A second press also used to wipe the first run's error before you had read it, so a failed update could quietly go back to looking healthy.

  • The update status page was publishing your update log. That page has no password on it by design — your browser has to keep reading it across the restart, when nobody is logged in — and it was answering with the last forty lines of update.log. That is docker compose output: image names, absolute paths, the layout of your server's filesystem. Nothing in CITY_NET has ever displayed it. It now returns the phase and a short reason, and the detail stays in backend/data/update.log on your machine, where it was already being written.

  • The update helper no longer runs through a shell. The command that recreates your stack was assembled as a line of shell text with your project's directory pasted into it inside double quotes. Double quotes look like protection and are not — they stop a path with spaces from splitting, but they do not stop a path containing $(...) from being executed. That path comes from Docker's own labels. The command is now passed as separate arguments with no shell involved at all, which removes the possibility rather than guarding one instance of it. As a side benefit, a project folder with a space in its name now works.

  • Two upload routes accepted any file type. Battle map images checked nothing at all, and music checked the Content-Type the uploader wrote into the request — a claim about the file rather than a reading of it. So a file named payload.html could be stored under that name in a folder served without a password, where the extension decides what a browser thinks it is. Both now check the extension, which is the part that actually determines how the file comes back out.

    The lists are drawn around what actually works, not around what is safe — every image format the map renderer can draw, SVG included, and every audio format a browser will play, including Opus and AAC that the old check turned away for having an unusual Content-Type rather than for being unplayable. What makes a generous list safe is the change below, not a short list.

  • Everything under /uploads is now served sandboxed. Files that came from outside are told, at the point they are served, that they may not act as part of this site — no cookies, no stored data, no reaching anything of yours — and browsers are told not to second-guess what type they are. This covers files already sitting on your disk from before, and it is what lets the upload rules stay generous instead of defensive.

Changed

  • Update both containers, not just the backend. The larger upload ceiling lives in the web server's config, which ships inside the frontend image. A normal docker compose pull takes care of it; an install that updated only the backend would find large maps refused by the proxy before the app ever saw them.

  • Running without the Docker socket is now a documented choice rather than a broken install. That socket is root on your host, so removing it is a sensible posture for a server exposed to the internet — everything keeps working except the update button, which refuses with an explanation. The refusal now offers updating from the host as an equal option instead of only telling you to add the socket back. See UPGRADE.md.

  • UPGRADE.md now states plainly what you are trusting when you press Update. Images are pulled from Docker Hub and run as root with the host socket, so that account's security is your install's security. That is true of any auto-updater; it should be written down rather than implied, along with the two ways to opt out.

Fixed

  • A rejected upload now tells you which file and why. Every refusal said a different amount and the most common one said nothing at all: a file over the size limit was never handled, so it came back as a web page the app then tried to read as data — what you actually saw for uploading a large map was a syntax error about an unexpected <. Refusals now name the file, say what was wrong with it, and list what would have worked, with the size limit included. The selected-file line names the file too, rather than only its size.

  • The web server in front of CITY_NET would have rejected a large map before the app saw it. It had its own 25MB ceiling on request bodies, separate from the app's, so raising the map limit alone would have achieved nothing in a Docker install — the upload would have failed with a bare "413" from the proxy, naming neither the file nor the reason. Both now agree, and a test checks that they still do.

  • A failed update could leave the app waiting for a restart that was never coming. The image pull already reported a bad exit, but the step after it — the helper that recreates your containers — did not, so if that failed the progress display sat on "restarting" indefinitely. It now says so. (On a successful update this never fires, because the helper replaces the very container watching it, which is the whole reason it exists.)

  • An upload interrupted halfway no longer leaves a file behind for ever. Uploads are written to a temporary file while they arrive; every normal outcome cleans up after itself, but a server killed mid-transfer cannot. Those leftovers — up to the full size limit each — are now swept at startup. Nothing that reached the map library is touched, and an upload in progress is left alone.

  • Animated maps show a preview in SELECT EXISTING. They are listed there now, and that gallery draws every entry as a still image — so a loop appeared as a broken icon.

  • The map uploader now offers video in the file picker. It was images only, so an animated map could not be selected without switching the dialog to show all files — and until now, doing that got you a successful upload and a blank map, because nothing in the app could draw one.


[1.9.4] - 2026-08-21

A pass over everything the server does that leaves the process.

Nothing here was being exploited. Two features reach services we do not own — the update check and the Companion character import — and they had grown two separate sets of gaps, each easy to get right once and easy to forget twice. Almost nothing here changes what you see; the one exception is an update check that used to lie to you, below.

Docker installs need do nothing. If you run your own reverse proxy in front of CITY_NET, see UPGRADE.md — there are two header lines to add.

Security

  • Anyone could stall the whole server by asking for the version. GET /api/version is open on purpose, so a client can watch for the restart an update causes — and it shelled out to docker info every single time it was asked. That call blocks everything else the server is doing until Docker answers, so a loop of requests to an unauthenticated endpoint was enough to hold the entire thing still. Whether a Docker socket is there cannot change while the process is running, so it is now asked once at first use and remembered, with a timeout for a socket that is present but not answering.

  • The Companion import has a ceiling now. It is the one open route that spends our outbound requests on an anonymous caller's word, which made it two things it should not be: a way to point this server's address at a third party as fast as you can send requests, and a way to walk the six-character code space and read back whichever characters answer — other people's characters, not ours to hand out.

    Ten attempts per ten minutes, per caller. Importing your own character is something you do once, and a couple more times if you mistype the code, so this is generous to a person and useless to a script. Requiring a login instead would have closed the same holes and taken the feature away from the people it is for: an open game has no player accounts to check.

  • The update check had no time limit at all. It stopped after five pages of tags, which bounds pages rather than time — so a registry that accepted the connection and then said nothing left the request hanging with no end to it. It also read the address of the next page out of the reply it had just been given, which let whoever answered choose where this server knocked next. Only the path is taken from the reply now; the destination is ours. And a response can no longer grow without limit while being read.

  • Errors from Docker Hub are no longer repeated back to the browser. They can carry host paths and internal addresses that were never ours to publish, and an admin can only retry or wait either way.

Changed

  • Every outbound request now goes through one place. A named destination, HTTPS, a deadline that covers the whole exchange rather than just the connection, a size limit, and no following of redirects — none of which a caller can opt out of and none of which has to be remembered twice. It means the answer to "what does this app talk to, and under what rules" is one short file rather than a search.

  • The server can tell your table apart. Every request arrives through the frontend's web server, which was not passing along who sent it — so as far as the backend could see, everyone at the table was one caller. Anything counting per person would have counted the whole house together, and the first player to import a character would have used up everyone's allowance.

Fixed

  • A throttled update check told you that you were up to date. Docker Hub rate-limits anonymous requests. When it did, its refusal was read as though it were a list of releases — an empty one — and an empty list is indistinguishable from there being nothing newer to install. So the check reported "You're up to date" while never having found out. It now says the registry refused the request, which is a thing you can act on: wait, and ask again.

    Whether you ever hit this depends on how often you press the button and what else shares your address, which is exactly why it was worth fixing rather than leaving to chance.

  • The update tests had been quietly talking to the real Docker Hub. They stubbed the network at a layer the code stopped using, so the stub silently stopped applying and the suite kept passing — because the real registry happened to agree with what the tests expected. It would have started failing on its own the day a release was published. The suite now refuses any real outbound call and names the address it caught, so this cannot happen a third time.


[1.9.3] - 2026-08-18

Cyberpunk RED characters can arrive from the Companion, by code.

Added

  • Import a character with a six-digit code. Build a character on the Cyberpunk RED Companion, export it, and type the code into IMPORT_SHEET — stats, skills, humanity, lifepath, and the names of your weapons, cyberware, gear and vehicles all come across. A third source beside the fillable PDF and the paste box, offered under Cyberpunk RED only, since the Companion is a Cyberpunk tool.

    The fetch happens on the server rather than in your browser, so your address never reaches a service you did not choose to contact. It fails into a sentence rather than a hang: a code that does not resolve reads differently from a service that is down, because those are two different things to do next about.

  • The preview is its own window. What an import would do is the thing you have to read before agreeing to it, and it was previously a block below three inputs in a pane that scrolls. Now it opens beside the dialog with everything visible at once: what was recognised, what was skipped, and — new — what the source never had. An export names your car but carries no SDP, and being told that is the difference between "the import worked" and "the import is broken".

  • IP and Reputation on the Cyberpunk RED sheet. Both are currencies a table spends between sessions and neither had anywhere to live but prose in the notes.

Changed

  • Applying an import replaces the sheet rather than merging into it. Merging quietly kept the past: a skill you dropped at the source stayed on the sheet, and a weapon row that no longer existed kept its damage.

    Because that is destructive, it goes through a confirmation that names the fields you will lose — not a warning, a list — so you can cancel and write them down first. Which vehicle you are riding in survives a replace: that is not character data, and re-importing a sheet should not turn you out of a car mid-session.

  • The icon rail labels show on hover. They were browser tooltips, which only appear once the pointer has rested — so the label came a beat late and never at all if you kept moving. Every rail button now has one, and they behave the same as each other, which they previously did not.


[1.9.2] - 2026-08-17

Enemy vehicles for the GM, and friendly NPCs who can ride along.

Added

  • An ENEMY VEHICLES window, for the GM. Every vehicle on an NPC sheet, grouped by that NPC's folder, with the hull bar, REPAIR and DAMAGE, and the MOVING toggle. Two ways in, both admin-only: the sidebar, and a button on a token menu.

    Nothing had to be stored to make them persist. NPC sheets already render the vehicle section and already live in folders, so an enemy van typed last session has been sitting there all along — what was missing was a read, since every roster query filtered NPCs out. Fill one in, and it is there next session.

    The roster is asked for rather than pushed, and refused to anyone but the GM, so a player's client never holds enemy hulls or armour at all. Attackers still see the cover badge on the token, which already worked.

  • Seating for enemy vehicles, from the tokens on the map you are looking at. Any of the GM's tokens can be put in a seat — including ones with no character sheet behind them, since a nameless ganger in the back of a van is just a token. The pickers filter to the city map or the battle-map floor currently in view, because running a rooftop fight should not mean scrolling past every ganger in the city.

    Friendlies are offered alongside hostiles but marked apart, in blue: both being the same green is how a body from your own side ends up in a hostile driver's seat in a hurry.

  • Players can invite friendly NPCs into their cars. Friendly tokens on the map level appear in the seat pickers beside the people at the table. Hostiles are not offered, and are refused by the server rather than merely hidden — a picker is a suggestion, the server is the rule.

    Anyone can turn an NPC out again, GM included. The rule that only you or the GM can empty your seat protects a person's autonomy; a token has none, so an invite is always undoable.

Changed

  • A seat holds one occupant, whichever kind. A person's seat and an NPC's are stored in different places — a player's on their own sheet, a token's in a new table — so taking a seat now turns out whoever was in it either way. Without that, a car quietly seats two bodies in one place and nobody notices until the shooting starts.
  • A passenger who walks off the current map level stays visible in their seat, marked OFF MAP, rather than reading as empty — which would have let the next change to that seat turn them out unintentionally.

Fixed

  • The vehicle roster reaches a player who has just logged in. Someone who had not added a vehicle of their own saw an empty window even with other people's cars on the roster. A client asks for the roster before the user logs in, so that request is dropped by design and a push at login covers it — but the push was still limited to Cities Without Number and still read CWN sheets whatever system was live. Nothing else refreshed the window until somebody happened to save a sheet. Applying an imported sheet had the same fault, one layer along.

[1.9.1] - 2026-08-16

Cyberpunk RED gets vehicles, sharing the machinery 1.9.0 built for Cities Without Number.

Added

  • A vehicle section on the Cyberpunk RED sheet. Name, SDP and its maximum, SP, seats, archetype, speed and cost as reference, and notes. Four vehicles, empty ones collapsed, SEATING in the header beside the collapse toggle — the same place, and the same order on the GEAR tab, as the CWN sheet.

    There is no picker for the book's vehicle table and there cannot be one: that table is not ours to distribute, where CWN's is published under a licence that permits it. ARCHETYPE fills the block instead, from seventeen archetypes of our own — BIKE, COMPACT CAR, SUPERCAR, SPEEDBOAT, CABIN CRUISER, AERODYNE, AIRSHIP and the rest — on a scale that belongs to nobody. They are approximate on purpose and every field stays editable, so a table wanting their own book's values types four numbers over the top and keeps the seats and the wireframe. The vehicle's name follows the archetype until someone names it themselves, and then it is theirs.

  • Eleven more wireframes, for the water and civil aircraft the CWN table has none of. Keyed by what the shape is rather than by any book's name for it, so a third system can reuse a speedboat without inheriting somebody else's vocabulary. Twenty-one shapes now.

  • Ramming, the driver's weapon. Offered in the attack panel as RAM · 6d6 to whoever holds the driver seat, and resolved whole rather than rolled to hit — driving into a thing is not an attack. Both vehicles take the same damage and armour does not apply, so ramming a parked car to death costs you most of a hull; the result says both hulls every time, because that is the half a table forgets. Everyone aboard both vehicles is named for the crash injury, which is what the seat roster is for.

    Targeting a token decides what you hit: aim at someone in a car and the car takes it, aim at someone on foot and they do. Resolved on the server, because a client deciding that could aim at a passenger and leave the car untouched.

  • A downloadable blank form for the sheet importer. The dialog always accepted a fillable PDF, but there was nowhere to get one — you needed a sheet whose field names happened to match. ↓ DOWNLOAD BLANK FORM gives you that sheet, for each of the three systems that have an importer. The field names are the contract, so a test walks every label on every form back through the importer: a box the importer cannot read is one a player fills in and loses.

  • The importer understands vehicles, for both systems. A car already written down elsewhere comes across rather than being retyped — which matters most in Cyberpunk, where there is no table to pick from. A SDP value also seeds the current pool, or every imported vehicle would arrive already wrecked.

Changed

  • The seating window no longer knows a game system. It read the wireframe and the seat names from the CWN book table directly, which meant a second system could only have a seating diagram by forking the component. Both now arrive through one resolver, and a test drives the window with no CWN behind it at all.
  • Vehicles are a registry rather than a hardcoded system. Which systems have them is now a list, and the active system travels with each call — so a table that has played both does not find last campaign's cars on this campaign's roster. Neither system lists the other's vehicles or players, or can seat into or damage the other's cars.
  • Cities Without Number gained vehicle import fields, which it never had. Less load-bearing there, since picking a book type fills the stat block, but a car edited away from its preset now survives a round trip.

[1.9.0] - 2026-08-15

Cities Without Number gets vehicles: on the sheet, in combat, and shared across the table.

Added

  • The book's vehicle table, as presets. Ten vehicles from p.82 — Motorcycle to Dropcraft — with their AC, HP, Armour, Trauma Target, Speed, Crew, Hardpoints, Power, Mass, Size and cost. Picking a TYPE fills the whole stat block, and the row then follows it: a Motorcycle shows no weapon mounts because it has no hardpoints, a Tank shows three.

    Armour is left unset on the Tank, APC and GEV. The book prints * and ** there rather than a number, meaning an immunity the GM rules on, and inventing a rating would be worse than writing the rule into the vehicle's notes — which the preset does.

    Every row is pinned to the book by a test holding the table verbatim, in its own column order. Five values had already been transcribed wrong from a screenshot before that existed.

  • Ten wireframes, one per vehicle. Drawn rather than illustrated: the city is wireframe with deliberately low segment counts, and a photo-real car dropped into that reads as a foreign object. Stroke-only and currentColor, so each takes the theme it is rendered in.

  • Mounted weapons from the book. Ten guns a hardpoint can carry, as a picker on each mount that fills damage, trauma and skill. Five are purpose-built vehicle weapons; the rest are Heavy weapons from the personal tables, two of which have no fixed damage at all — a grenade launcher fires whatever you loaded.

  • Vehicle fittings, against a Power and Mass budget. All 24 from p.84 as a per-vehicle list: install, remove, and a running POWER 2/8 · MASS 6/15 that turns red when overloaded. Mounted weapons count toward the same budget, because the book is explicit that a hardpoint costs Power and Mass just as a fitting does.

    A list rather than a dropdown filling fields, because a fitting comes off again. Several rewrite the stat block — Extra Durability is +25% HP, Hardpoint Support adds a mount — and a control that wrote those numbers in would have no way to take them back. The effects are printed; the numbers stay yours. Power Systems raise the pool rather than un-spending, so two of them read as POWER 0/14 (+6) rather than as minus six spent.

  • A shared VEHICLES window. A picker across every player's sheet, the vehicle's wireframe with a leader line and a dropdown per seat, a MOVING toggle, and its AC, AR, HP and crew count — so the table can read a car without opening anyone's sheet.

    Seats come from the book's Crew number, so a Motorcycle seats one and an APC sixteen. Guns are deliberately not seats: a Tank is crew 3 with three hardpoints and can never man every gun and drive at once, and making each hardpoint a seat would hand it a fourth body and quietly resolve a tension the rules intend.

    Anyone can seat anyone — piling into a car is a decision the table makes out loud — but only you or the GM can take you out, enforced on the server rather than by hiding a button.

    Three ways in, all the same window:

    • VEHICLES under GEOMETRY_PROTOCOLS in the sidebar
    • SEATING in the sheet's VEHICLES section header — beside the collapse toggle rather than inside the section, so folding it away does not take the button with it
    • VEHICLES on a token menu, shown only once some vehicle exists

    Players are listed by character name rather than login, since nobody at the table thinks of each other by account name.

  • A car badge on the sheet and the token menu, naming the vehicle you are in, with the way out. Shown on anyone's token but inert unless it is yours: seeing that someone is in a car is the reason it is there.

  • A hull bar in the VEHICLES window, with REPAIR and DAMAGE. Combat already wrote the vehicle's HP when someone shot the car; what it never covered was the repair afterwards, or a crash, or a ram — all of which meant opening the owner's sheet and editing a number. The bar grades green to amber to red at the same thresholds as the character health windows and reads WRECKED at zero.

    Both directions clamp to the hull on the server: a repair cannot exceed the maximum and damage cannot go below zero, because destroyed is derived from HP rather than stored, and an unclamped write could invent a state the rules have no name for. Everyone sees the bar; only the owner and the GM get the buttons, since taking someone else's car apart is what shooting it is for.

Changed

  • There is no CUSTOM vehicle. It was a hole rather than a feature: with no type a vehicle had no crew, no hardpoints and no Trauma Target, so it seated one person, refused to fire the mounts you could still fill in, and took traumatic hits twice as often as any real one — none of which said anything on screen. Every vehicle is one of the book's now; start from the nearest and edit the numbers.

    The section holds nothing until you add something, and a new vehicle starts as a Motorcycle. Changing the type renames it, unless you have given it a name of your own — a vehicle called MOTORCYCLE has not been named, one called Betty has.

  • Attacks on someone in a vehicle hit the vehicle. Armour Rating subtracts and the remainder comes off the vehicle's HP; the people inside are untouched until it is destroyed, at which point it stops being cover. A vehicle's AC is its own plus the driver's Drive while moving and four worse while stationary, and firing out of a moving one costs four either way.

    Damage lands on the owner's sheet, so four people in one car share one pool. Every unreadable state resolves to on foot — an owner who logged out, a row never filled in, a wreck — because a broken reference should cost someone their cover, never make them impossible to hit.

  • Mounts can be fired at all, from the weapon picker, keyed by the (vehicle, mount) pair so mount 1 on two different cars do not shadow each other. A gunner can fire the guns of a car they are riding in: the to-hit was already built from the attacker's sheet, so they fire someone else's cannon with their own skill.

  • Only weapons the book marks ! can traumatise a vehicle. Every trauma die was working on cars. A pistol's trauma die is devastating to a person and does nothing at all to a car. The marker rides on the trauma value — d20/x4! — because the damage string is fed to the dice roller and a marker in it would break the roll.

  • A vehicle rolls trauma against its own Trauma Target, not against that of whoever happens to be sitting in it.

  • A CWN attack reads the defender's sheet before rolling to hit, not after. Whether they are in a vehicle decides the AC. It was already being fetched twice further down, so this is one read where there were two.

  • Repeated sheet entries collapse. Six vehicles declared meant thirty rows of blank fields on a sheet with no vehicles on it. Only entries holding data render, plus one on an empty sheet, and + ADD reveals another. Each has a REMOVE behind a confirm click.

  • Themes carry a vehicle accent. Seat markers use each theme's own secondary colour rather than one imported hue, and only to mark a seat as occupied — the leader lines, the labels and an empty seat are drawn in the theme's own colour, because they are part of the diagram. Monochrome stays monochrome.

  • New column on locations: vehicle_state. Added by the usual startup migration — no action needed on upgrade, and empty for anyone not in a car.


[1.8.2] - 2026-08-14

Added

  • A roll button for BODY in Cyberpunk RED. It was the one stat with a modifier and no way to roll it. The roll is resolved server-side by field id like every other, which is the part adding a button alone would have missed — and a test now walks the template and fails if any roll button lacks a server-side roll behind it, so the next stat added cannot render a control that does nothing when pressed.

Changed

  • The Cyberpunk RED stats grid groups the rollable stats first. BODY moved up among the stats you roll, and MOVE and LUCK moved to the end, since neither is a check: MOVE is a movement allowance and LUCK is a pool you spend. The grid no longer mixes the two kinds together.

[1.8.1] - 2026-08-02

Added

  • An optional development channel, selected by one setting. IMAGE_TAG=latest by default, which is stable; IMAGE_TAG=dev follows development builds. That is the same variable docker-compose.yml interpolates to decide which images are pulled, so what the update check offers and what it installs cannot disagree — the check reads the deployment's own tag rather than a separate declaration of intent.

    Development builds are tagged X.Y.Z-dev, with a counter accepted but not required, so 1.9.0-dev and 1.9.0-dev.7 both work and introducing counters later is a build-workflow change rather than a code one. A dev build of a newer release is offered to someone on an older release, the release itself supersedes its own dev builds when it lands, and a release user is never dragged back onto a dev build of the same version. A pinned version tag counts as stable, since pinning is not a channel.

    A Dev Build to Docker Hub workflow publishes them, on manual dispatch or a push to a dev branch. It runs the test suites first, never touches latest, and pushes two tags per build: dev, which is what IMAGE_TAG=dev pulls, and X.Y.Z-dev.N, which is the only form the update check can see — dev is not a version and is filtered out of the tag listing. It refuses to run when package.json still holds an already-released version, since X.Y.Z-dev sorts below X.Y.Z and such a build could never be offered to anyone.

    IMAGE_TAG must also reach the .env beside docker-compose.yml, which the setup steps already cover by copying backend/.env to the project root: compose interpolates from the project file rather than from env_file.

Fixed

  • Dev builds could hide stable releases from everyone else. The update check read a single page of a hundred registry tags. Tags come back ordered by recency, so a run of development builds after a release fills that page with X.Y.Z-dev.N — every one of which a stable deployment filters out, leaving it with nothing and reporting no update available when one existed. It reads further pages now, but only while it has found nothing usable: the newest release is on the first page in the ordinary case, so this normally makes exactly one request as before. Bounded at five pages, so a registry that keeps offering another one cannot hang the check.

    It fails quietly and selectively, which is what makes it worth fixing ahead of the arithmetic — only people on older versions, only on the stable channel, with no error anywhere.

  • One dev tag on the registry would have silenced update notices for everyone. The tag filter was /^\d+\.\d+\.\d+/, unanchored, so 1.9.0-dev passed it and then parsed to NaN — which made the sort comparator return NaN, leaving the ordering undefined and letting a prerelease surface as the newest tag, whereupon the version check correctly refused it and reported no update at all. Version tags are matched strictly now, and sorted by a comparator that understands them.

  • The in-app updater could sit on WAITING FOR SERVER indefinitely. Every step failed silently — both child processes discarded their output, the route answered "Update started" before checking anything could work, and a non-zero exit from docker compose pull simply returned — so a stack that could not update was indistinguishable from one still working, and the client polled every three seconds forever with no deadline.

    The likeliest cause on a long-running instance is the compose file's own self-mount at /tmp/docker-compose.yml, which the update reads. A container started before that line existed does not have it, the pull fails, and everything above turns that into an endless wait — so the instances least able to update in place are exactly the ones that have been running longest.

    POST /api/update now checks the mount, the Docker socket and the compose project labels before answering, and returns 409 naming what is missing and what to do about it. Both steps append to backend/data/update.log, which lives on the data volume and so survives the container being replaced. GET /api/update/status reports phase and error, and the modal shows them, reassures at 45 seconds that a pull legitimately takes minutes, and gives up after six with the host command to fall back to.

  • A container too old to update itself now says so immediately. Such a container answers POST /api/update with "Update started" and then does nothing, so the client used to wait out the full deadline to learn what could be known at once. It is asked for GET /api/update/status first — a route that only exists in the self-checking build — and if that is missing the modal says the container predates it and shows the command to run on the host. The response shape is checked rather than just the status code, since a setup serving index.html for unknown paths answers 200 with a page. Nothing is POSTed to a server that cannot act on it.

  • The nav-panel update button had none of the above. There were two implementations of the update flow — the modal and the panel — and only the modal's was hardened. The panel ignored the server's refusal entirely, waited on the version rather than the restart, and polled every three seconds with no deadline, so the original symptom survived in the path the upgrade guide tells people to use. Both now drive one shared client, which is the only reason a second copy could go unfixed.

  • The "read more" link on the update panel pointed at a heading that does not exist. README.md#updating has no such anchor, so someone whose update just failed landed at the top of a 570-line README. Both links now go to UPGRADE.md, which is the actual guide.

  • A successful update could hang too. The client waited for the reported version to change, but a build without APP_VERSION reports dev before and after. /api/version now carries a boot id and the client waits for the restart itself.

  • The update check offered downgrades. hasUpdate was latest !== current, so a published tag trailing the running one counted as an update — a 1.8.0 instance was offered 1.7.4. It is a numeric version comparison now, and anything unparseable (dev, latest) is never offered.

Technical

  • The update logic moved out of the admin route into backend/updater.js. None of it was reachable from a test where it was, which is a large part of why three separate faults sat in it unnoticed.

[1.8.0] - 2026-08-01

Added

  • Drawn generation bounds — the city generator gains a DRAG_RECT / DRAW_AREA toggle. DRAW_AREA traces a boundary the way water is drawn, and generation is confined to that shape: blocks centred outside it are dropped, road seams are clipped to it, and a footprint straddling its edge is rejected. A concave shape generates nothing in its notch, so an L or a crescent works as drawn. The traced outline stays on screen until GENERATE, unlike water, which saves immediately.
  • Street layouts — a LAYOUT selector offering six distinct city types. Everything downstream of the block list is layout-agnostic, so a layout only has to produce blocks and the roads between them.
    • GRID — two perpendicular families of streets with avenues every fourth line. Reads as Manhattan or Chicago, and is genuinely distinct from the default, which always produces irregular rectangles however it is tuned. That road hierarchy is most of what makes a grid look designed rather than generated.
    • SUPERBLOCK — the same recursive split with a much larger floor: fewer roads, larger plots, open ground between them. Soviet microdistrict or corporate arcology.
    • RING — a beltway city, San Antonio being the reference: concentric loop roads with elevated arterials running out from downtown. The corners of a square selection are left empty on purpose, because a ring city is round.
    • ORGANIC_CELLS — a Voronoi diagram with streets along the cell boundaries. The only layout with no right angles in it: streets meet at odd angles and blocks are wedges and pentagons, which reads as a town that grew around footpaths rather than one a surveyor set out. Long cell boundaries become avenues, so the network gets a hierarchy without one being invented — the long runs across the diagram are the ones that would carry traffic anyway.
    • DOWNTOWN — an elongated street grid whose blocks are cut into lots around their rim, facing the street, with the middle of the block left as back lots. Every other layout hands the generator one block per city block, so a block gets one structure — right for a tower in a park, wrong for a downtown, where what makes a dense city look dense is many narrow buildings shouldering together along the street. Roughly four times the buildings of GRID over the same ground. Block sizes vary rather than repeating one cell — short blocks, long ones, and the occasional enormous one where a street was never cut through, Washington Square being the obvious example — and the depth of the built-up rim varies with them, since a constant depth reads as a machine even when the frontages differ. A large block is ringed inward more than once so it reads as built out rather than hollow, stopping while there is still a back lot behind the buildings; a block too thin to have a rim and a middle becomes a terrace rather than one monolith the length of the block. SUPERBLOCK is deliberately left alone as the opposite idea rather than being turned into this.
    • BSP stays the default and an unrecognised layout falls back to it, so existing generation is untouched and a stale saved option cannot produce an empty city.
  • Generated water — a WATER selector offering a RIVER across the region, a COAST cutting one edge off, or a LAKE inside it. Rivers and coastlines are most of why real cities look like themselves: they force asymmetry, cut districts apart, and give bridges a reason to exist, which until now only happened if a GM had drawn water first. NONE is the default and doubles as the off switch, so generation produces water only when asked and a GM who wants to draw their own is never overruled.
  • Park ponds — a PARK_PONDS toggle gives some parks water as well as trees, with the trees standing back from the edge. Separate from WATER because they are different scales of decision — a river reshapes the whole city, a pond is scenery in one plot — and either is wanted without the other. Off by default.
  • Roundabouts — a ROUNDABOUTS selector (OFF / SPARSE / NORMAL) putting a circus where major roads meet. It is an overlay on the finished road network rather than another layout, so one implementation serves all five: approaches are cut back to the ring, the ring is laid, and the island gets a monument where there is room for one or a stand of trees otherwise. An empty disc reads as a hole in the network rather than a junction, so every island gets something.
  • Seeded generation — an optional SEED field. The same seed over the same area with the same options rebuilds the same city, so a map can be recreated or shared as a short string. Leaving it blank rolls a fresh seed, and the seed actually used is reported back beneath the field rather than written into it.
  • REGENERATE — clears the previous generation in the selected area and builds afresh, for iterating on a district without hand-deleting it first. It keeps anything the GM authored: named structures, tokens, battle-map content and hand-drawn water all survive. Plain GENERATE still adds to what is there.
  • UNDO on the generator panel — the same server-side undo as the admin header, reachable without leaving the panel.
  • The panel now stays open after generating, instead of dropping back to the main admin list — generation is something you do repeatedly while tuning.

Changed

  • Road hierarchy, skyline taper and per-zone setbacks. Road width is graded by split depth, so arterials read as arterials and side streets as side streets. Building height now blends continuously with distance from the centre instead of stepping in bands, which turned the skyline from flat plateaus with hard seams into a taper. Corporate plots leave forecourts and slums and markets build to the lot line, via a per-zone lot coverage applied after the aspect clamp.

Fixed

  • Water bridges no longer pierce the buildings they pass over. Placement deliberately ignores overpasses so the ground beneath a deck stays buildable — that is what stops an elevated road sterilising every block it crosses — but nothing then stopped a tower rising straight through one. Anything under a deck is now capped just below it, and where the deck is too low to build under at all, near its ramps, the building is dropped rather than squashed to nothing.

Technical

  • A drawn boundary is a water polygon with the sign flipped — water keeps what falls outside, a boundary keeps what falls inside. clipSegmentToLand was generalised into clipSegmentToPolygons(seg, polys, keepInside) so the two share one implementation and cannot drift; there is a test asserting they are exact inverses. footprintOutsidePolygon mirrors footprintInWater but is stricter: water asks whether a footprint touches water at all, a boundary asks whether all of it is inside.
  • A boundary of fewer than three points cannot enclose an area and is treated as absent, falling back to the plain bounds rather than generating nothing and looking like a broken button.
  • Skipping a block draws no randomness, so generation without a boundary splits byte-identically to before. Tests pin that at both the split and the whole-city level.
  • RING fills the disc with a single sub-layout and lays its arterials over the top. A first version partitioned the disc into annular sectors and sub-laid each one, which produced a sparse, fragmented city — a sector's bounding box is far larger than the sector, so most of what each run generated fell outside its own region and was discarded. Density is now on par with the default over the same area.
  • RING elevates its spokes but leaves its loops on the ground. A closed loop has no ends to ramp down at, so an elevated ring either never meets the street network or does so at one arbitrary point. Spoke ramps are sized as a fraction of the spoke rather than a fixed length, so both ends reach the ground however large the city is — a fixed ramp longer than half the deck leaves it ending in mid-air.
  • Spokes run from the innermost loop outward rather than converging on a point, which removes a starburst of dead ground at the centre and is closer to how highways meet a downtown loop.
  • LayoutFn may return overpasses alongside blocks and roads, and generateCity merges them with whatever bridges the water needed. splitCity gained an optional minimum block size rather than SUPERBLOCK being a parallel implementation.
  • Block.lot lets a layout subdivide a block itself. The generator trims road padding off a block, clamps its aspect toward square and applies a per-zone setback — three rules that exist to turn a whole city block into one sensible plot. Applied to lots inside a subdivided block they pad the neighbours apart, square up the narrow frontages and pull the row back off the street, which is every ingredient of a street wall undone. A block flagged lot skips all three, its footprint being already decided. No existing layout sets it, so all five are untouched, and there is a test asserting that.
  • Three app-wide conventions a generator must not override. None is what its name suggests, each is asserted in a test, and getting any of them wrong produces a structure that does not look like it belongs to the same city. color: '#00ff00' is not green, it is the sentinel meaning "inherit the theme" — the renderer resolves anything else verbatim, so naming a real colour opts a structure out of theming entirely. polyCount: 5 is not a quality setting: everything is drawn as a wireframe, so the segment count is the look, and a 16-segment cylinder reads as a bright striped cage beside a city of pentagonal prisms. And shape: 'rhombus' is not an octahedron, it is a player or NPC token — TOKEN_SHAPES treats it as one on the server, a purge spares it as player content, and OverlapChecker publishes it in activeRhombuses, so a structure using one as a finial publishes a fake token inside itself.
  • A roundabout island is a tiny lake, as far as roads are concerned. clipSegmentToLand already cuts a segment out of a polygon and leaves the approaches stopping at its edge, which is exactly what a junction does to the roads meeting it — so trimming reuses the water clipper rather than a second implementation, and the ring reuses the arc sampling RING uses for its beltways. Siting has to handle two kinds of junction: a BSP or Voronoi network joins at shared endpoints, but GRID lays each street as one full-length span, so its crossings share no endpoint and are found only by intersecting segments — segmentCrossing was promoted out of the water clipper for that. The whole pass runs after consolidateRoads, which snaps nearby endpoints together and would otherwise snap a ring of short segments into a blob.
  • A Voronoi cell is reduced to the largest rectangle that fits it. Block is {x, z, w, d}, and the plot filler lays buildings out along a rectangle's axes — a pentagon has no axes to work with. Fitting a rectangle inside each cell keeps that 1180-line generator completely untouched while still delivering the irregular street pattern, which is where nearly all of the look comes from. The rectangle rarely fills its cell, so setbacks vary from plot to plot for free. Cells are built by half-plane clipping rather than a sweepline: for the hundred or so seeds a city needs it is fast enough, and Fortune's algorithm would be several hundred lines of beach line and event queue to save milliseconds nobody is waiting on. Seeds sit on a jittered lattice — a perfect lattice gives a honeycomb as machine-made as the grid, and fully random seeds clump into slivers too thin to build on.
  • Water is generated before the split; ponds after it. The split is already water-aware, so generating a river first means the road grid stops at the banks of its own accord and bridges are sited from the stubs left there — generating it afterwards would mean cutting finished roads, which is a different and worse problem. A park pond is the opposite case: the park only exists once the split has produced the block it sits in. That is safe because a pond is contained by its plot and never reaches a road, and ponds are kept out of the water array the split, the shoreline roads and the bridge siting were built from. A test pins it: a ponded and an unponded run of one seed give identical roads and overpasses.
  • water_bodies gains a generated column, so a regenerate can clear its own river without destroying a lake the GM drew. Existing rows default to 0, so everything already on a map counts as hand-drawn. The migration runs on startup — a server on older code will accept generated water but store it as hand-drawn.
  • Seeding reached the buildings, not just the layout. cityGen/ had a single Math.random (the injected default), but generateThemedBuildingsForPlot had forty of its own, so a seed reproduced the street layout while the buildings on it changed every run. The rng is threaded through to the plot filler. It is deliberately not crypto-backed: 1.7.1 moved outcome-deciding rolls to OS entropy, and city layout is cosmetic.
  • POST /purge-region clears a region's generated content in one transaction and emits a single update, rather than the panel issuing a delete per object. It distinguishes generated content from authored content by isUserDefinedName, token classification, battle-map membership and the new water flag.
  • Region membership and counting moved out of AdminPanel into cityGen/region.ts, and seeding into cityGen/rng.ts, so both are testable without rendering the panel.

[1.7.4] - 2026-07-28

Added

  • Map export — a new EXPORT tab in the admin panel writes the city out as a top-down PNG or a WebM video. Both frame the whole map from actual content bounds, so everything lands in one shot however far the city sprawls, and both carry a translucent CITY_NET watermark with the repo URL beneath it.
    • Resolution — 1080P / 2K / 4K / 8K. The names describe width; height follows the city's aspect ratio, and the pixel count is shown beside each label so the mapping is unambiguous. Video is capped at 2K.
    • Record length — 5 / 10 / 30 seconds, with a ● REC indicator and seconds-remaining readout while capturing.
    • INCLUDE_GRID (on) puts the ground grid in the shot, where it reads as map paper. INCLUDE_HIDDEN and INCLUDE_TOKENS (both off) keep GM-only structures and player tokens out, so a shared export cannot leak secrets and is a clean map rather than a snapshot of where everyone stood. Tokens never affect framing in either state, being mobile.
    • TRANSPARENT_BG renders the PNG without the theme background, for compositing over paper texture in an image editor. PNG only — WebM has no dependable alpha channel.
    • Files are named after the live map: nightcity-2026-07-28.png rather than city-map-1753718400000.png.
  • Export framing from real city bounds — bounds account for building rotation (via circumradius, correct for all three rotation axes), road width rather than just centrelines, water bodies, and overpasses. The camera centres on the city centroid, so a city generated off-origin is no longer cropped.
  • Sign rotation on all three axes — signs gain rotation_x and rotation_z alongside the existing yaw, with sliders and LAY_FLAT / STAND_UP presets. LAY_FLAT pitches a sign face-up with its text running north, so signs work as ground labels read from a top-down view.
  • UNDO in the DRAW_WATER panel — the same server-side undo as the admin header, so a mistraced body can be reverted without leaving the panel.
  • The live map is remembered — active_map_name is recorded in global_settings when a map is saved or loaded and cleared when the world is wiped. Nothing tracked this before: loading a map replaced the world and forgot where it came from.

Fixed

  • Signs no longer lose their rotation on save — a sign pitched flat sprang back upright on reload. Nothing in the stack handled pitch: the save path read only rotation.y, the schema stored only rotation_y, and the renderer hardcoded the other two axes to zero. All three now persist end to end. Existing signs are unaffected — the migration defaults to 0 and the renderer coalesces nulls.

Technical

  • Recording renders off-screen rather than mirroring the live canvas. Mirroring pinned the video to the window size and meant the resolution choice could not reach it. Rendering through its own camera also means recording never touches the live view — no camera fly-to, no locked controls — and it can reuse the PNG's orthographic camera, so the video is square-on rather than a perspective view leaning at the edges. makeExportCamera is shared by both paths so they cannot drift apart.
  • One off-screen WebGLRenderer is reused for the session. Building one per export exhausted the browser's WebGL context budget, at which point it evicted the live city canvas and the scene went black. forceContextLoss() was the cause of the loss it was meant to prevent.
  • PNG size is clamped to what the GPU can render, read from MAX_RENDERBUFFER_SIZE and MAX_TEXTURE_SIZE. The clamp considers both axes: a tall narrow city at 4096 wide implies a height several times that, and breaching the limit fails the render outright or returns a blank image. Aspect ratio is preserved when scaling down, and the reduction is logged rather than silent.
  • The recording countdown derives from a wall-clock deadline rather than decrementing per tick. Rendering the scene twice per frame starves timers, and a decrementing counter loses every dropped tick permanently — the display fell behind and then snapped from a few seconds straight to zero when auto-stop fired.
  • Frames are throttled to 30fps. requestAnimationFrame fires at the display rate, and anything above what the recorder samples is GPU work thrown away.
  • Recorder codec falls back vp9 → vp8 → webm → browser default, with onerror handling so a failed capture cannot wedge the button on STOP_RECORDING.
  • The grid's fade radius is widened for the duration of an export. fadeDistance is tuned for the interactive camera, and both paths frame the city from far above it, so the grid faded to nothing exactly when INCLUDE_GRID asked for it.
  • Hidden structures are baked into shared InstancedMesh draw calls and cannot be switched off by scene traversal, so the export suppresses them through renderLists and waits for the commit before capturing.
  • preserveDrawingBuffer removed from the main <Canvas>. It was required while recording read the live canvas; now that recording renders its own frames, it was a permanent cost to every player for nothing. overheadFlyHeight and the canvas-mirroring composite loop went with it.

[1.7.3] - 2026-07-27

Added

  • Custom dice — the GM can define dice with any number of sides and any face values, and every connected player can roll them.
    • CUSTOM_DIE.EXE — draggable build window opened from the dice roller. Name is required and must not collide with a standard die or an existing custom die; the side count is typed and confirmed with SET, which generates one input per face (20 visible before the list scrolls).
    • Faces accept anything — numbers, words, symbols. A roll is summed only when every face is numeric; otherwise the result lists the faces with no total. Signed values like +1 count as numeric, so ladder-style dice still total correctly.
    • Dice are stored server-side and shared with the table. GET /api/custom_dice is public so players see the GM's dice; create, edit and delete are admin-only. Each change broadcasts customDiceUpdated with the full list, so players already online pick up a new die without reloading.
    • Only admins see the create button, the edit cog and the delete control.
  • System dice — GET /api/system_dice/:system serves read-only dice that ship with a game system, merged into the roller ahead of the GM's own. They are defined in code rather than seeded into the database, so an app update changes the definitions with no migration and there is no row for an admin to edit or delete — the route exposes no write verbs at all. Seeded with the Fate ladder die (dF), dormant until a Fate system is added.

Changed

  • Roll definitions are resolved server-side. requestCustomDiceRoll now takes only a die id and reads the definition from the database (or the built-in manifest); it previously accepted a die object from the client, which meant face values could be forged.

Fixed

  • Custom dice rendered nothing in the tray. DiceScene infers the die shape from the results key parsed as a number, but custom rolls key by die name — parseInt('Fate') is NaN, which built a sphere with NaN segments and drew no dice. Rolls now carry a diceSides map and a sidesForKey helper resolves the shape, so a 4-sided die renders a tetrahedron. The helper falls back to a d6 rather than NaN for any unrecognised key, and sphere segments are capped so a 999-sided die cannot stall the tab.
  • Editing a die with the window already open left the form blank. The builder seeds its fields from useState initializers, which only run on mount, so clicking the cog changed the prop without reloading the form. App now keys the component on the die being edited; the window position moved up to App so it no longer jumps back to centre on the remount.

Tests

  • 585 backend / 844 frontend — new suites: custom_dice (API validation, auth, duplicate-name conflicts), system_dice (manifest integrity, absence of write routes), sockets.customdice (built-in vs database resolution, summing rules, count clamping, forged-payload rejection), CustomDieBuilder (create/edit modes, name-clash rules, reload-on-target-switch), useCustomDice (merge order, locked flag, broadcast handling, mutation errors), plus sidesForKey coverage in the dice tray suite.
  • Fixed a latent hook bug in useApi.test.ts: beforeEach(() => mockFetch.mockReset()) implicitly returns the mock, which Vitest then calls as a teardown function — invoking fetch() with no arguments.

[1.7.2] - 2026-07-27

Added

  • Basement floors in battle maps — floor designation now supports B# (Basement N) in addition to Penthouse, Level, and Lobby. The elevator button order is PH → L# (descending) → LBY → B# (descending), matching physical building layout top-to-bottom.
  • Hide / reveal structures — admin selection panel gains a HIDE_STRUCTURE / REVEAL_STRUCTURE toggle. Hidden structures are invisible to players; admins see them as a gray wireframe overlay with a [HIDDEN] label. State is persisted in the database and synced to all clients in real time.

Changed

  • Parks generated by city gen no longer show signage — holotree trunks are now created with has_signage: 0, and the location POST handler correctly stores the field so the setting survives a server restart.
  • Signage checkbox removed from the per-structure edit modal — the sidewalk option is no longer relevant per building and has been removed from the UI.

Fixed

  • Grid snap jump on enable — enabling GRID_SNAP no longer teleports the selected object. The object's position is rounded to the nearest integer when snap is turned on, so the first dragged frame starts from an already-aligned position.
  • Road traffic flicker — GhostTraffic cars were teleporting on every render because the weights array reference changed each tick, causing the packets memo to regenerate all car positions and speeds simultaneously. Route data is now held in a ref so packets only regenerates when the route count or car count actually changes.
  • Building flicker on hide/reveal — toggling hidden state no longer causes a visual pop. Both selectedLocation and the locations array are updated atomically in a single render, so the building never momentarily switches between the instanced and interactive render paths.
  • Auto-signage z-fighting — signs on building faces are pushed 0.04 units outward from the wall surface, eliminating the blink caused by coplanar geometry.
  • Signs one-sided — signs were invisible when viewed from behind. Each sign is now rendered as two FrontSide planes back-to-back (via a TwoFacedSign helper with a render-prop material), so text reads correctly from both directions without mirroring.

[1.7.1] - 2026-07-26

Added

  • Water-aware city generation — the generator predated water bodies entirely and built straight through lakes.
    • The BSP split clips each road seam to land as it lays it, so the grid stops at the shore rather than being drawn across the water and trimmed back.
    • Buildings, parks and landmarks all avoid water through a single check in the placement test. Plots centred in water are skipped; plots that merely touch a shoreline still build on their dry side.
    • Shoreline roads — each water body gets a road just inland of its edge, so approaches meet a waterfront junction instead of dead-ending at the water.
  • Generated bridges — roads that meet water can be carried across by an overpass, the first time generation has produced one.
    • Crossings are sited deliberately: a road end at the shore is probed across the water and kept only when it reaches a far bank in range, comes down on dry ground, and lands on a road facing back — so every bridge joins two pieces of the network.
    • Ramps run the full length of their approach road (capped at 120), touching down at that road's junction rather than part way along a block. The two approaches differ, so runs are set per end.
    • Deck levels are 6/9/12/15, assigned by graph colouring over which decks cross. A bridge takes the lowest level none of its crossing neighbours hold, and levels beyond what its approaches can reach at a fixed 1:3 grade are excluded — height is bought with ramp length, never steepness.
    • House rule: OVERPASS_DENSITY — new control in CITY_GENERATOR (off / sparse / normal / heavy) scaling both how many crossings are bridged and how far a span will reach (120 / 200 / 300 units).
  • Crypto-backed dice — every roll that decides an outcome now draws from OS entropy instead of Math.random. Backend rolls (roll engine, CP:R/CWN/SR6 attacks, dice tray, initiative) use crypto.randomInt; frontend initiative systems use crypto.getRandomValues. Visual and cosmetic randomness is unchanged.

Changed

  • City generation extracted to frontend/src/cityGen/ — roughly 420 lines moved out of a JSX click handler in AdminPanel into a pure module that takes bounds, options and world state and returns blocks, roads, buildings and overpasses. Persistence and UI state stay in the component. The random source is an injected parameter, which is what makes the module testable; behaviour is unchanged on a dry map.

Fixed

  • Buildings could sit inside existing structures — the collision grid filed each obstacle under the single cell holding its centre while lookups scan a 3×3 neighbourhood, so anything wider than one 20-unit cell was only detected near its middle. Obstacles are now registered across every cell their footprint covers, with an always-checked list for degenerate sizes.
  • Buildings could sit on roads — two causes. The road test compared a road's nearest point to a building's centre, which reads as clear whenever a road clips a corner; it is now an exact segment-versus-box clip. And the themed generators emit most of a structure relative to a cleared root without testing each piece, so wings and annexes escaped the check entirely — finished plots are now re-checked and rolled back whole if any piece landed badly.
  • Roads crossed under the waterfront road without joining it — approaches overshot it and stopped just past it with no shared point, so consolidation could not link them. Road ends near the waterfront are now pulled onto it.
  • Pillars speared overpasses passing beneath them — pillar placement checked roads only. It now also skips a column that would pass through a lower deck, while leaving pillars that merely stand under a higher one. Applies to hand-drawn overpasses too.

[1.7.0] - 2026-07-23

Added

  • Initiative tracker — full real-time initiative system across all supported TTRPG systems. Start/end initiative per scene, roll and re-roll, drag-to-reorder, admin remove, late-join flow, and multi-scene combats that share a single turn counter.
    • Generic — 1d20 roll; TURN counter; order held for entire combat.
    • Shadowrun 6e — REA + INT + Nd6 roll (Wired Reflexes extra dice selector); PASS counter; end-of-pass −10 score decay; survivors carry decayed scores into the next pass; new-round banner when all scores drop to zero.
    • Cyberpunk RED — REF + 1d10 roll; ROUND counter; order held for entire combat (RAW). Exploding d10 available as an opt-in house rule (rolling 10 adds 10 and triggers another roll, chaining while 10s keep coming).
    • Cities Without Number — side-based initiative (RAW default) and individual initiative (opt-in house rule). 1d8 + DEX mod per combatant; ROUND counter; PCs win ties. PLAYERS side score auto-derived from the best individual PC roll; NPC side auto-created on the first enemy roll with its own 1d8; friendly NPCs slot into the PLAYERS side. Sub-ordering within sides is drag-to-reorder. Individual initiative available as an opt-in house rule in AdminPanel.
  • Initiative sidebar panel — admin sees the tracker inline in the sidebar; players see it as a floating window or sidebar panel depending on layout.
  • Initiative nav button blink — the initiative button pulses green (matching the unread-chat animation) when the admin starts initiative or a new SR6 round begins, prompting players to roll.
  • Sheetless NPC manual roll — admin can type a score directly into the token context menu and add a sheetless NPC to the initiative order without a linked sheet.
  • Floor-aware NPC display — in building battle maps, the tracker only shows NPCs on the current floor; players and city-map combatants are always shown.
  • Roll breakdown in dice tray — initiative rolls appear in DICE_TRAY.EXE history with full breakdown: REA(5) + INT(2) + 1d6(4) = 11 for SR6, REF(6) + 1d10(8) = 14 for CP:R.
  • EXPLOD animation — when the CP:R exploding-die house rule is active and a 10 fires, the combatant row flashes white→green and displays an EXPLOD badge; the dice tray history appends 💥EXPLOD.
  • SR6 NPC sheet stats — bulk NPC rolls read REA/INT from the linked character sheet (sheet_data joined on location GET) rather than falling back to 3+3.
  • Player sheet stats — player JOIN rolls fetch the player's own sheet via GET /api/sheets/own so REA/INT/REF are used instead of defaults.
  • House rule: Exploding Initiative Die (CP:R) — added to the CP:R section of the House Rules panel in AdminPanel.

Changed

  • Initiative system registry — InitiativeSystem interface and getInitiativeSystem(key) registry extracted to frontend/src/modules/initiative/systems/; each system is an isolated module (generic.ts, sr6.ts, cpr.ts).
  • RollOptions interface — shared options bag (extraDice, explodingInitiative) replaces positional args; each system reads only its own fields.
  • diceResults keys — changed from Record<number, number[]> to Record<string, number[]> to match JSON serialization (object keys are always strings after a round-trip through SQLite).
  • Initiative system registry now includes cwn.ts; InitiativeSystem interface gains defaultMode?: 'individual' | 'side'.

Fixed

  • Combatant row drag — browser was intercepting the avatar portrait as a native image drag instead of triggering the row's drag handler. Fixed by adding draggable={false} to portrait <img> elements and setting a transparent ghost image on drag start.
  • Side-view drag-and-drop — dragSideId was stored as React state, causing it to be stale (null) when the first dragOver event fired immediately after dragStart. Converted to a ref so handlers are always synchronously current; visual indicator state (dragOverSideId) kept separate. Also tightened onDragLeave to only reset on genuine container exit (via relatedTarget check) rather than on every child-element transition.

[1.6.4] - 2026-07-20

Changed

  • Admin panel tabs — reorganized the admin panel into three tabs: CITY (map and structure tools), GAME (TTRPG system, currency, pay players, bank sounds, purge chat/rolls), and PLAYERS (online/offline player lists with grant/revoke admin, edit requests, active editor).

[1.6.3] - 2026-07-20

Fixed

  • Docker data persistence — all map and character data was lost on container restart or image update because DB_PATH was only set in docker-compose.yml. Running the container without compose fell back to the ephemeral /app/city.db inside the container. ENV DB_PATH=/app/data/city.db is now baked into Dockerfile.backend so the correct mounted path is always used.
  • Dirty image builds — backend/data/ was missing from .dockerignore, meaning a locally-built image could accidentally bundle the developer's database into the published artifact. Both backend/data/ and backend/uploads/ are now explicitly excluded.
  • DB path startup log — the resolved database path is now printed on boot ([db] opening database at: …) for easier deployment debugging.
  • In-app update wipes data — the self-update route mounted the host project directory at /project inside the helper container, so compose resolved ./backend/data to /project/backend/data and passed that to the host Docker daemon. The daemon found no such path and created a new empty bind mount, wiping all map data on every in-app update. Fixed by mounting at the host's own absolute path so the daemon receives paths it can actually resolve.

Added

  • Docker config regression tests — backend/__tests__/docker_config.test.js asserts that Dockerfile.backend contains the required ENV DB_PATH and mkdir directives and that .dockerignore excludes the data and uploads directories; failures block CI before any image is built.
  • CI Docker validation job — validate-docker-config job in ci.yml runs shell-level grep checks against Dockerfile.backend and .dockerignore on every PR and push to main, catching regressions before the release workflow runs.

[1.6.2] - 2026-07-19

Fixed

  • Docker build — extend npcSheetGenerated socket type in useSocket.ts to include system, portrait_url, sheet_name, and sheet_description; TypeScript was rejecting the fields added in 1.6.1 and failing the production build

[1.6.1] - 2026-07-19

SR6 polish: drain resistance, glitch feedback, Edge management, NPC sheet fixes, and name/description sync.

Added

  • Drain resistance — each spell row in the AWAKENED tab gains a yellow RES button (enabled when Drain Value and tradition attribute are set); clicking it rolls WIL + tradition attr, applies net drain to the Stun track, and overflows excess into Physical HP automatically
  • Glitch / Critical Glitch banner — dice tray displays a pulsing yellow GLITCH or red CRITICAL GLITCH banner after any SR6 pool roll that qualifies; clears on the next roll
  • SR6 Edge management (admin) — Admin Panel SR6 section adds REPLENISH ALL EDGE (resets every player's Edge to their max) and a per-player GIVE 1 EDGE button (online non-admin players only, capped at max)

Fixed

  • NPC sheet system mismatch — OPEN_SHEET now only shows when the linked sheet's system matches the active game system; a CWN sheet no longer surfaces under SR6 and vice versa
  • GENERATE_SHEET → OPEN_SHEET flip — button switches immediately after generation without requiring the token window to be closed and reopened; system is now included in the npcSheetGenerated socket event
  • Edge pips not rendering — pip display now uses the higher of current or max edge so pips appear even before edge_max has been explicitly saved
  • NPC sheet name/description sync — editing name or description on an NPC sheet updates the linked map token label live (sheet → token); saving an enemy/friendly token with a changed name or description patches the linked sheet (token → sheet); generation already seeded both fields from the token

[1.6.0] - 2026-07-18

Shadowrun 6E — third full game system, built entirely on the existing sheet/roll/attack architecture.

Added

Shadowrun 6E system

  • SR6 character sheet — 9 attributes (BOD–EDG), Essence/Magic/Resonance, 16 consolidated skills rolled as d6 pools, 4 weapon rows (DV / AR / mode / atk), Edge pips (spend-by-click, CP:R LUCK pattern), derived Physical/Stun monitors, initiative score and composure recomputed on every save
  • Pool dice engine — new pool roll shape: pool size = attribute + skill (+mods), 5s and 6s are hits, glitch when half or more of the pool shows 1, critical glitch on a glitch with zero hits; results land in the dice tray as N hits / M dice
  • Two damage tracks — Physical monitor lives on the token (HP bar as usual); Stun is a sheet track that clamps at the Stun Monitor and overflows the excess into Physical automatically, server-side
  • SR6 attacks — one ATTACK button: attack pool, weapon Attack Rating compared to the target token's Armor Rating shifts DV ±1, damage applies to Physical with a "GM: soak BOD+ARMOR" prompt (soak stays manual per RAW-lite)
  • Auto-rolled defense — defenders backed by a sheet dodge automatically (REA + INT pool, shown in the dice tray); net hits decide the hit and add to the DV; sheetless tokens don't defend
  • Stun bar in the health window — SR6 health review shows the Stun track (cyan bar, red + overflow warning when full) under the Physical monitor, live-updating; served by a public two-number stun endpoint
  • AWAKENED / EMERGED tabs — gated behind the new sr6_awakened and sr6_emerged house rules (spells/powers/tradition and complex forms/sprites)
  • SR6 NPC tiers — Ganger / Street Tough / Shadowrunner / Prime Runner for GENERATE_SHEET, with consistent monitors, armor and weapon rows
  • SR6 sheet import — attribute/skill aliases (BOD, AGI, cc, perc…), Edge max seeds current, plain-text stat block parser

Changed

Modularity pass

  • Gated sheet tabs are data — GATED_TABS/hiddenTabsFor config replaces per-system conditionals in the player sheet hook and the admin NPC window
  • One import mapper — makeMapFields factory: CP:R / CWN / SR6 importers are now alias tables + numeric sets around a single shared loop
  • Shared attack scaffolding — target lookup and attackResult emission extracted from the CWN and SR6 handlers; per-system attack code is rules only

[1.5.1] - 2026-07-17

NPC portraits with a glitchy TV/CRT effect, stock headshot library, and token info window improvements.

Added

NPC portraits

  • Stock headshot library — 16 bundled NPC headshots (new art by PaMuDA) served from frontend/public/npc-headshots/; GENERATE_SHEET auto-assigns a random headshot to new NPC sheets
  • HEADSHOTS picker — admin button on the NPC sheet window opens a < N/16 > prev/next navigator to browse and assign any stock headshot; seeds to the current portrait when it's a stock image
  • TvPortrait component — reusable, self-contained glitchy TV/CRT portrait effect (chromatic R/B fringe, scanlines, rolling refresh band, intermittent glitch jitter) usable on any portrait; BracketPortrait and the token info window both use it
  • FX toggle — per-sheet portrait_shadow_filter field with an FX ■/□ button under the sheet portrait enables/disables the TV effect (on by default)
  • Public token identity route — GET /api/sheets/npcs/link-public/:location_id returns sheet name + portrait only (no stats/description) for enemy and friendly tokens

Changed

Token info window

  • Title now reads ID: {sheet name} from the linked NPC sheet instead of HOSTILE_NODE
  • Portrait from the linked sheet displays in the window with the TV effect
  • DATA_DESCRIPTION pulls from the linked sheet's description (admin view); falls back to the data point's description
  • All players (not just admin) now see a linked token's name and portrait; description and stats remain admin-only — the GM builds mystery manually by leaving a token unlinked or unnamed
  • EDIT_DATA_POINT is hidden for tokens linked to a sheet — the sheet drives name/description
  • CHECK_HEALTH now opens only the health window; the quick sheet card no longer auto-opens with it

[1.5.0] - 2026-07-16

Cities Without Number (CWN) — second full game system, accessibility theme picker, and a major frontend deduplication pass.

Added

Cities Without Number system

  • CWN character sheet — 6 attributes with SWN modifier table (0/unset → neutral, not −2), 3 saving throws, 19 skills (2d6 plain sum), combat stats (BHB, AC as token-linked writable field), System Strain (current/max), conditions (Frail, auto-initiative), and a Deluxe tab (4 spell rows + summoning) gated behind the cwn_deluxe house rule
  • Armor section — players enter BASE_AC + DEX_CAP + SHIELD_BONUS; cwnEffectiveAc computes and pushes the result to the token on every save; field stays hand-managed when armor is blank
  • CWN combat — single ATTACK button (same UX as CP:R); 1d20 + BHB + skill + attr mod roll-to-hit vs target AC; damage roll + attr mod; trauma die checked against the defender's Trauma Target (default 6) — rating is the damage multiplier; shock damage on miss if target AC is low enough; all results land in the dice tray
  • Stabilize flow — MORTALLY WOUNDED banner at 0 HP; clicking STABILIZE rolls the clicking user's 2d6 + Heal + INT mod vs DC 8 + rounds since downed (+ 2 without tools); on success: target becomes Frail and gains 1 HP after the dice animation completes; on failure: rounds_since_downed increments; allies see the STABILIZE button on the Health Review window (their Main Action per RAW)
  • FRAIL banner — shown above 0 HP when the frail flag is set; CLEAR FRAIL button (GM-approved) clears it
  • STIM_HEAL — field-healing shortcut adds +1 System Strain then heals; refuses with STRAIN MAXED (409) when at max
  • LONG REST — admin panel button recovers 1 strain on every CWN sheet
  • Spell rows — manual name/effect/damage/cost fields; per-row CAST button rolls damage if present, spends Effort, flags OVERCAST! if insufficient
  • CWN house rules — cwn_trauma (default ON — gates trauma die and major-injury flow) and cwn_deluxe (default OFF — gates Deluxe tab); data-driven HouseRulesPanel component replaces both old bespoke IIFE blocks
  • CWN NPC tiers — Warrior / Expert / Boss / Elite combat tiers + 3 Spirit tiers; each stamped with trauma_target: 6; NPC melee DV formula is system-specific (CP:R formula only for cyberpunk_red)
  • CWN sheet import — alias mapping for all 19 skills, armor fields, spell rows, and trauma_target

Accessibility — login theme picker

  • Theme picker on login screen — custom dropdown ("THEMES ∨") lets players change the UI theme before logging in; each option is styled in its own theme's primary colour
  • Theme saved to localStorage immediately; written to player_accounts.theme on login
  • Theme returned in the JWT payload and applied on every session without a re-login
  • Logo scene glow, non-animated text, and version badge all follow the active theme
  • Standalone sheet tab reads theme from the socket auth handshake (currentTheme prop) or localStorage fallback

Changed

Frontend deduplication

  • usePlayerSheet hook — extracted shared sheet state, debounced saves, house-rule fetches, and action emitters (roll / deathSave / stabilize / castSpell) from both CharacterSheetWindow and SheetPage; both surfaces now share one implementation and have full CWN feature parity
  • SheetAttackPanel — unified component replaces the separate CprAttackPanel and CwnAttackPanel; driven by ATTACK_PANEL_CONFIG keyed by game system
  • hasSheetCombat(system) — exported from Sidebar.tsx; all gameSystem === 'cyberpunk_red' || gameSystem === 'cities_without_number' checks in App and Sidebar replaced with a single capability predicate; adding ruleset N requires only one config entry
  • HouseRulesPanel component — data-driven (accepts HouseRuleDef[]); CP:R and CWN house-rule panels are now one-liners

Dice tray

  • Back-to-back broadcasts (e.g. to-hit + damage from a single attack) are now queued and played sequentially instead of overwriting each other; dice rolls play for 5 s, dice-less results (shock) for 0.8 s

Fixed

  • Unset stats (value 0) now map to modifier 0 under the SWN table — previously bucketed as ≤ 3 → −2
  • AC on the character sheet now reads from and writes to the token's melee_ac / ranged_ac fields; unset token AC defaults to 10
  • Health review window was resolving to the enemy's token when both the enemy and the player share the same owner; now prefers the rhombus shape
  • Stabilize / death-save outcome (Frail flag, HP write, penalty) now applies after the dice animation completes (DICE_ANIM_MS delay), not before
  • NPC tiers generated for non-CP:R systems no longer receive the CP:R melee-DV formula
  • Trauma die now resolves against the defender's Trauma Target (default 6), not the weapon's rating; rating is the damage multiplier
  • Version text on the login screen now uses the theme's --green colour with bold weight and glow

[1.4.1] - 2026-07-14

Fixed

  • App version fallback — backend /api/check-update now reads the running version from package.json when the APP_VERSION env var is not set (manual installs and dev environments no longer show a false "update available" prompt)

[1.4.0] - 2026-07-13

Cyberpunk RED character sheets: the full Phase 3-5 sheet system, making CP:R the first feature-complete game system.

Added

  • Character sheet system — template-driven sheets (one renderer, per-system templates); player window, admin view of any player/NPC sheet, standalone browser tab (?sheet=true), quick-sheet card, portrait upload with TV-glitch effect, segmented HP bar (green/yellow/red)
  • Server-authoritative rolls — stat/skill rolls resolve against the stored sheet (exploding CP:R check die); results land in the dice tray and history
  • CP:R combat flow — single ATTACK button; weapon picker from structured sheet weapon rows (name/DMG/skill/ROF); to-hit vs token DV; aimed shots (−8, head, ×2 damage through armor); damage auto-rolled, soaked by defender SP, armor ablation on penetration; damage writes through to token HP; attack animation follows the weapon type (melee/ranged)
  • SP SHIELD — defender's shield intercepts damage first and breaks down point-for-point; overflow soaks against location SP
  • Critical injuries — two+ max-face damage dice trigger +5 direct damage (ignores armor/shield) and prompt the GM to roll the book's injury table (table not embedded)
  • Death saves — MORTALLY WOUNDED banner at 0 HP with DEATH SAVE button; 1d10 + escalating penalty vs BODY (natural 10 always fails); penalty resets on healing above 0
  • Seriously Wounded — banner at ≤ threshold HP; −2 to all checks applied server-side (−4 while mortally wounded)
  • Armor penalty — heavy-armor stat penalty applied to all REF/DEX-keyed checks and attacks
  • LUCK on rolls — arm pips on the sheet (declared before the roll, per RAW) for a flat bonus on the next roll; spend is capped/decremented server-side; attack panel has its own LUCK selector
  • House rules panel (ADMIN → TTRPG_SYSTEM, staged APPLY/REVERT) — MELEE_DV TAKE-10 (10 + DEX + Evasion instead of 6 +) and LUCK BONUS ALSO NEGATES NAT-1 (also unlocks a dedicated 1-LUCK fumble shield); rules apply live via settingsUpdated
  • LUCK pips + admin reset — hexagonal pips on the sheet header; RESET_ALL_LUCK in the admin panel restores every player to max
  • Humanity → EMP — editing Humanity recomputes current EMP (= Humanity ÷ 10) on every write path (template-declared derived fields)
  • NPC library — create/delete NPC sheets, folders with MOVE control, ATTACH sheet to a token, OPEN full sheet editor; NPC sheets mirror their linked token's HP live
  • Leveled NPC generation — GENERATE_SHEET takes a per-system tier (CP:R: MOOK/SKILLED/PRO/ELITE) seeding stats, skills, armor, weapons, token HP and DVs; melee DV computed from the sheet (6/10 + DEX + Evasion), GM can override via EDIT_DV
  • Sheet import — IMPORT on every sheet window: fillable-PDF form extraction, JSON paste, or stat-block text; per-system alias mapping with preview before apply; linked fields (HP/cash) refused with explanation
  • Token defense per system — MELEE_AC/RANGED_AC labels become MELEE_DV/RANGED_DV under CP:R; CP:R hides the two-button melee/ranged flow behind one ATTACK button
  • OPEN_SHEET on token windows — players open their own sheet from their token; admins open any player's or NPC's sheet from any token
  • npm run dev (backend) — nodemon auto-restart so backend code changes apply without manual restarts

Changed

  • TTRPG_SYSTEM panel moved above CURRENCY_ICON in the admin panel; SHEETS list removed
  • Sheet UX polish — placeholders (ghost example text) on all free-form fields, upload hint bar attached under the portrait, weapon rows replace the free-text weapons area (notes field retained), CUR ≤ MAX clamping on paired fields (frontend + server)
  • CHECK_HEALTH window — now resolves NPC tokens and tracks live HP by token id (was frozen at open and player-only)

Fixed

  • NPC armor ignored in attacks — defender sheet lookup branched on token owner instead of token type, so enemy tokens (which carry an owner) never found their linked NPC sheet; SP always read 0
  • DV edits not reflected — EDIT_DV saved correctly but the token window showed the stale snapshot until reopen
  • Death save / FIRE appearing dead — stale backend process; mitigated permanently by the new npm run dev watcher

Tests

  • 377 backend / 574 frontend — new suites: cpr_attack (to-hit, armor, shield, crits, LUCK, death saves), sockets.deathsave (socket integration: death saves, NPC SP, import apply, tiered generation, fumble-shield gating), sheet_import (PDF/JSON/text extraction + mapping), npc_tiers, plus renderer/library/import-dialog coverage

[1.3.1] - 2026-07-08

Added

  • Health in broadcast info card — when the admin clicks a player/enemy/friendly token in broadcast mode, the spectator info card now shows the heart monitor (animated EKG, flatlines at 0 HP) and injury map (body silhouette with red zone overlays, BLIND/BLEEDING flags); injury map only appears when injuries are present
  • RETURN_TO_ORIGIN button — added to the top of the Quick Access menu; smoothly flies the camera back to world center

Fixed

  • Broadcast zoom direction — spectator camera was zooming toward screen center instead of the admin's cursor; fixed by deriving lookAt from the camera's actual forward direction rather than the orbit pivot, which dollyToCursor does not update
  • Dice roll broadcast delay — spectator overlay was showing roll results immediately; now waits the same 5-second animation delay as the roller's own DiceTray
  • AC hidden from non-admin players — the attack banner in the DiceMenu and the rhombus info window no longer show AC value or roll threshold to non-admin players; admins still see full detail

[1.3.0] - 2026-07-08

Added

  • PLAYERS panel in admin dashboard — primary admin can grant or revoke temporary admin access to online players directly from the admin panel; offline players are listed but cannot be granted access
  • Custom confirm modals — replaced all browser confirm() dialogs (PURGE_ALL_WATER, PURGE_CHAT_HISTORY, PURGE_ROLL_HISTORY) with consistent in-app modals
  • CUSTOM_STRUCTURE dashboard — renamed JOIN_STRUCTURE to CUSTOM_STRUCTURE with updated copy explaining prefab structure classification
  • Non-admin health view — non-admin players see only the heartbeat monitor and injury map when reviewing a token's health; HP numbers are hidden

Changed

  • UI label clarity pass — ACCESS_CODE → PASSWORD (all forms), JOIN_STRUCTURE → CUSTOM_STRUCTURE, GEOMETRY_PROTOCOLS → TOKEN_PROTOCOLS, all rhombus/beacon labels updated to TOKEN_*, INITIALIZE_RHOMBUS → PLACE_MY_TOKEN, and many more across Sidebar, AdminPanel, BankWindows, and UpdateModal
  • CUSTOM_SIGNS button — moved below all + placement buttons and prefixed with + for consistency
  • Admin panel section order — SIDEWALKS/SIGNAGE/SIGN_DENSITY moved above CURRENCY_ICON; BANK_SOUNDS grouped with CURRENCY section
  • CURRENT_SELECTION panel — repositioned above + ADD_NEW_STRUCTURE with equal padding for better visibility

Fixed

  • Stale elevatedUsers bug — revokeEditing and editingFinished now remove the user from elevatedUsers; previously a grant→revoke→re-request cycle gave full admin instead of an edit window
  • Stale pendingRequests bug — editingApproved and editingDenied now broadcast request removal to all clients; newly-promoted temp admins no longer inherit accumulated edit requests from earlier in the session

Tests

  • backend/__tests__/sockets.editing.test.js — 13 cases covering approveEditing, revokeEditing, editingFinished, grant/revoke elevated access, and regression scenarios
  • frontend/src/hooks/__tests__/useSocket.pendingRequests.test.ts — 9 cases covering pendingRequests state transitions and the stale-request regression

[1.2.4] - 2026-07-07

Added

  • Animated 3D login logo — LogoScene component renders a Three.js hex badge with wireframe skyline, spinning octahedron gem, and CITY_NET text label above the login panel
  • LogoScene.tsx — standalone Three.js scene (no react-three-fiber) with auto-rotation, gem bob animation, and green glow drop-shadow matching the app's CSS variables

Changed

  • Password field labels — ACCESS_CODE / CONFIRM_ACCESS_CODE renamed to PASSWORD / CONFIRM_PASSWORD on login, register, and password reset screens for clarity

[1.2.3] - 2026-07-07

Added

  • Silent update notification modal — admins see a draggable SYSTEM_UPDATE popup on login when a new version is available; supports UPDATE NOW, REMIND ME LATER (session), and SKIP VERSION (persistent)
  • Docker vs manual install detection — update modal shows one-click update for Docker installs; manual install users are directed to install instructions instead
  • GET /api/version endpoint — lightweight no-auth endpoint returning the running version; used for post-update polling without hitting Docker Hub

Fixed

  • Post-update page reload — frontend now polls /api/version after triggering an update and reloads only when the version changes, eliminating stuck "Update in progress" state
  • Nginx cache-busting — index.html served with Cache-Control: no-cache so JS bundles always reload after an update
  • Helper container path resolution — docker compose helper mounts hostWorkingDir:/project and uses --project-directory /project, fixing cross-OS path failures on Windows hosts
  • WATCHTOWER_API_TOKEN removed from required env vars — no longer triggers missing env var warning banner on admin login

Changed

  • Button hover/active states standardized — global button:hover applies filter: brightness(1.4) across all buttons; colored variants (danger, enemy, friendly, deploy, map save/load) use CSS classes instead of inline style overrides
  • Skip version and remind-later state is independent of the manual Check for Updates button in the nav panel — skipping the modal never blocks the sidebar update flow

[1.2.2] - 2026-07-06

Fixed

  • Update polling condition — poll now compares running version against originalCurrent (captured before update), not against Docker Hub latest; fixes stale comparison when multiple versions exist on Docker Hub

[1.2.1] - 2026-07-06

Fixed

  • One-click in-app update — admin panel now has a CLICK TO UPDATE button; no SSH required for future updates
  • APP_VERSION baked into Docker image — version is now embedded at build time so all users see accurate version info regardless of their local docker-compose.yml

[1.2.0] - 2026-07-06

Fixed

  • APP_VERSION sync in release workflow — Docker images now automatically deploy with APP_VERSION matching package.json version, eliminating false "update available" notifications

[1.1.9] - 2026-07-06

Added

  • Docker Hub version checking — admin panel button queries Docker Hub for new versions
  • GitHub Actions auto-tagging — release workflow now tags images with version numbers from package.json
  • PR template with checklists — ensures version bumps and env var updates before merging

Changed

  • Watchtower removed in favor of manual Docker Hub pulling and GitHub Actions workflow
  • README reorganized into Docker vs Manual setup paths with clear instructions
  • Added Windows PowerShell command variants for cross-platform support

[1.1.8] - 2026-07-05

Added

  • UPGRADE.md — comprehensive upgrade guide with step-by-step instructions for pulling new versions from Docker Hub
  • Env var status check on admin login — admins get an in-app warning if critical environment variables are missing
  • GET /api/admin/env-status endpoint — returns list of missing required env vars

Fixed

  • Backend startup now validates required env vars and logs helpful warnings if any are missing
  • Removed stale root-level .env.example to eliminate confusion (canonical location is backend/.env.example)

[1.1.7] - 2026-07-05

Added

  • Manual Check for update button in the CITY_NET system info panel (primary admin only) — triggers Watchtower on demand via its HTTP API
  • CHANGELOG.md added to the repo
  • Changelog link in the system info panel
  • DuckDNS support — optional duckdns service in docker-compose.yml keeps your subdomain pointed at your current IP automatically
  • Configurable host port via APP_PORT env var (default 80) — change to 8080 or any port your ISP allows; players connect to http://yourcity.duckdns.org:APP_PORT
  • IPv6 LAN direct connect documented — players on the same network can connect via http://[ipv6-address] with no port forwarding
  • Let's Encrypt / Certbot HTTPS instructions added to README for DuckDNS setups
  • DUCKDNS_SUBDOMAINS, DUCKDNS_TOKEN, TZ, and APP_PORT added to .env.example

Fixed

  • Build failure in Docker (tsc -b exit code 2) — viewSettingsUpdated socket type was missing renderSidewalks

[1.0.7] - 2026-07-05

Fixed

  • Custom sign position now saves correctly after using the Move gizmo — replaced unreliable dragging-changed event detection with an explicit UPDATE SIGN POSITION button
  • SAVE CHANGES on a sign now reads the mesh's actual position (not stale form values) so moving then saving no longer reverts on reload
  • Signs are now scoped to saved maps — they save with the map and restore/clear on load/clear
  • Signs no longer travel between maps on load
  • Custom signs and auto-generated building signage now render in front of traffic (renderOrder fix)
  • Cleared signMesh and selectedSignId on logout to prevent stale TransformControls state
  • Removed X/Y/Z coordinate fields from the custom signs panel — position is managed via the Move gizmo

Changed

  • Sign transform controls now activate explicitly via MOVE / ROTATE buttons rather than auto-attaching on selection
  • Added signs_data column to saved_maps table

[1.0.6] - 2026-06

Added

  • Custom signs: CRT/TV shader filter with per-sign intensity control
  • Custom signs: preset sign images and image URL rendering
  • Custom signs: multi-line support with per-line font size
  • Custom signs: free 3D transform gizmo — move and rotate selected sign in scene
  • Custom signs: camera-center placement and real-time rotation slider
  • Custom signs: font selector and remote font uploader
  • Auto-signage on buildings with neon, billboard, and poster variants
  • Sidewalk pavement rings around structures
  • Road sidewalks toggle
  • View settings (sidewalks, signage) broadcast to all connected clients

Fixed

  • Sign transform mode prop wiring
  • Font upload now surfaces actual server errors
  • Jitter sign positions along building face with overlap check

[1.0.5] - 2026-05

Added

  • Watchtower automatic update checks at 4 AM daily
  • Manual Docker Hub release workflow with image tags
  • NAV_OS version driven from package.json via Vite define
  • GitHub Actions CI pipeline

Fixed

  • Watchtower Docker socket path for Windows hosts (//var/run/docker.sock)
  • Watchtower schedule corrected from every-30s to 4 AM cron

[1.0.4] - 2026-04

Added

  • Ghost traffic drives overpasses with head/tail-light styling
  • Road eraser dashboard with segment, path, and purge-all tools (purge-roads)
  • Overpass split start/end ramp length controls
  • WASD/arrow key camera pan while drawing roads or overpasses
  • Continuous mitered ribbon road geometry with undo
  • Ko-fi widget in sidebar

Fixed

  • Overpass-to-overpass connectivity, snap joins, and undo support
  • Overpass ramp length persistence (ramp_length_start/ramp_length_end)
  • Traffic car rendering above road surface; ramp length max 160
  • Road segment chaining into full streets for smooth traffic fade
  • Camera pan speed and WASD direction lerping
  • rampLength=0 holds full height at both ends

[1.0.3] - 2026-03

Added

  • CRT scanline effects and per-theme gradients
  • Themes palette with 7 themes wired to canvas, grid, buildings, and dice tray
  • Water body persistence with map save/load/clear
  • Overpass persistence with map save/load/clear
  • Frontend hook extraction: useMapData, useSocket, useApi
  • 52 frontend tests (threeHelpers, useApi, themes); 35 music tests; 28 battle map tests

Fixed

  • Theme colors applied to 3D components
  • Camera dolly action for scroll-wheel zoom
  • Overpass renderer preview and geometry core

[1.0.2] - 2026-02

Added

  • Radio feed with volume mixer (separate BG and MUSIC sliders)
  • Attack animations — Phase 1–5: AC fields, attack UI, socket wiring, hit/miss animations
  • ARMOR_CLASS fields in GEOMETRY_PROTOCOLS menu
  • Streamer mode: director panel, broadcast overlay, spectator socket role, camera sync
  • Ping system with shadow outlines and Q-key shortcut
  • Measurement tool with real-time drawing and isFinal relay
  • Player name tags with hover/selection visibility and occlusion
  • Secure Mode: player account auth with admin-approval registration flow
  • Admin-approved password reset flow
  • NPC impersonation in global chat
  • Chat V2: context menus, PM tabs, NPC puppeting
  • Admin temporary delegation via global chat
  • CHECK_HEALTH button on player rhombus info window
  • High roller notification persisted to DB (fires once only)

Fixed

  • Radio feed: master mute, stale socket closure, double-click to load track
  • Attack: ARMOR_CLASS label rename, AC hidden from non-owners
  • Streamer: battle map camera sync, zoom broadcast, active user roster on join
  • Bank animations suppressed when closed; sounds suppressed over login chime
  • HP clamp when max is lowered below current
  • Socket reconnect with playerToken in Secure Mode
  • Player info window status for online players with no beacon

[1.0.1] - 2025-12

Added

  • CITY_NET // BANK and admin banking dashboards
  • Player rhombus path animation and battle map dedup
  • Injury tracking panel with body hit zones (major injuries)
  • Health review window for other players (read-only)
  • Battle map image upload (25 MB limit)
  • Battle map active indicator (pulsating cross-swords)
  • Friendly NPC rhombuses
  • Custom structure cross-map library (JOIN_STRUCTS)
  • District management UI and backend
  • Admin approval workflow for temporary editing access
  • Map Manager restricted to primary admin
  • Custom cinematic modals replacing all window.confirm/window.alert calls

Fixed

  • Battle map floor sync, camera orientation, rhombus scaling
  • Rhombus fade animation race condition
  • Bank -0.00 display and icon coloring
  • Map load preserving map_scale_multiplier
  • Structure deletion cleaning up associated battle maps and uploads
  • NPC race condition on startup

[1.0.0] - 2025-11

Added

  • Initial Docker containerized deployment (backend + frontend + Watchtower)
  • Procedural city generation with themed building layouts (Corpo, Urban, Industrial, Slums)
  • Interactive Draw a City mode
  • Road network with instanced geometry and spatial grid
  • Admin and player role system with JWT auth
  • Saved maps (save / load / clear)
  • Dice tray with physics
  • Battle map system with floor support
  • Player rhombuses with drag, overlap X-ray, and health tracking
  • Enemy and friendly rhombus deployment
  • Global and private chat
  • Measurement tool
  • Tree placer tool
  • Custom structure editor and premade structure picker
  • Water bodies with procedural shader
  • FPS counter