mobile_Game_v1

Imported from brchue-ux/mobile_Game_v1 (AGENTS.md). Install upstream with npx skills add brchue-ux/mobile_Game_v1. Copyright stays with the author.

Mobile Game (placeholder name)

Real-time mobile lane-battler with compound spellcasting. Concept design is underway; no code yet.

Status

Concept design in progress. The canonical artifact is the wayfinder map at .scratch/core-concept/map.md, with its 18 tickets in .scratch/core-concept/issues/. Read the map before doing anything here.

Session narrative lives in CHANGELOG.md — grep it for a past decision or why something was tried. It is not auto-loaded.

CONTEXT.md is the ubiquitous-language glossary — domain terms (lanes, creeps, target class, banking, etc.) with rejected synonyms flagged under _Avoid_. A first draft as of 2026-08-06; the map is still the source of truth if the two disagree.

Engine, stack, platform and name are deliberately out of scope until the concept settles.

Working method — read ## Working method in the map first

One ticket per session. The user word-dumps everything they think about it; that dump is used to rebuild the map, not just to close the ticket. Confirm understanding, then wait for the user to say proceed.

Sequential grilling was tried and dropped — this design is a graph, not a tree, and question-at-a-time forced forks faster than they could be answered.

One commit per rebuild. Every prior map version stays diffable. This is load-bearing, not hygiene.

/wayfinder is referenced in older notes but is not installed. /grilling and /prototype are.

Rules that govern every rebuild

• Rejections are append-only. Every rebuild carries forward every recorded "no" with its reason. A reversal is flagged as a reversal, in the user's words — never silently dropped. A rebuild is precisely what lost the tug-of-war rejection once already.
• After every rebuild, diff the deletions (git diff --cached), never trust the insertion count. This has caught silently dropped live content on five separate rebuilds across two sessions.
• Build from the user's words, not from a summary. A summary is lossy exactly where a correction lives.
• Locked constraints live in the map's ## Notes. Don't re-litigate without saying so explicitly.
• Ticket 18's "budget" governs what ships in a build, never what gets explored. It may not be invoked to discourage an idea. Define the term or don't use it.

Ticket state

• Resolved: 02 (combining = payload + modifiers, amended for small recipe sets), 13 (prototype).
• Closed by removal: 03 (gesture) — the subject was cut, not answered.
• Shelved: 09 (banking) — good shape, no payoff worth its cost. Not rejected; don't reintroduce unprompted. ⚠ See the open question below.
• Substantially answered: 01, 05 (protest lifted and narrowed 2026-07-29; the win condition is answered, the forward structure is what's open), 10, 15.
• Standing caveat, in the user's words (2026-07-29): "Everything that I say is always open to change." Nothing is a lock unless explicitly marked. This does not weaken the append-only rule — changes are reversals, flagged as such.

Where the design stands

All [provisional] unless marked.

• Core loop: accrue cards → combine into a compound spell → select a target. Casting is selection, not performance [committed]. Four target classes: help your hero, help your lane, hurt their lane, slow their hero.
• Camera: a pannable MOBA observer over a slightly angled overhead battlefield. The map exceeds the screen; you drag around it. The player is a commander, not an avatar.
• Screen: portrait 1080×2400. Warcraft 3 zoomed out — top ~75% map viewport, bottom ~25% command bar holding the cards, readouts on both heroes, and the hero's standing orders.
• Information: no fog of war (forced by casting-as-selection — you can't select what you can't see). The hand is hidden, and probably the opponent's field-manipulation capability. You always see what is happening, never quite what is coming.
• Hero: works the jungle automatically on route/behaviour modes; the player sets standing orders (lane, jungle camp) but never steers directly. Cards influence power, orders influence priorities. Differentiation is asymmetry, not power level. Death costs nothing; 1–2 priced buybacks per match.
• Match: the win condition is reinforcement exhaustion — one pool per side "as of right now", and "you only live as long as you have reinforcements." One win condition with three levers on it: your own creeps dying (baseline drain), terrain manipulation (a black-hole-class spell that eats their creeps for extra reinforcements), and hero intervention (a card putting your hero in the lane ~10s, farmed power-ups making the push harder or more survivable). The base survives as the creeps' source and may still be the thing that dies; what is rejected is the prototypical route — towers → barracks → towers → core. Concession/disconnect is a win by default. PvE and PvP share one shape.
• Jungle: playable space for spells, not a wall (the hero must traverse it — this is now a pricing baseline, since blocking traversal is only worth a card because passage is the default). Lane creeps leash back on lost aggro.
• Terrain manipulation — water, lava, holes impeding enemy traversal. No longer a candidate objective in its own right: it is lever 2 on the one win condition.
• Cards: ~100, bring ~8–12 (reversed from ~20 on 2026-08-26 so the Shape A rotation can cycle; the map and 11/16 carry the reversal). All obtainable by every player, none purchase-exclusive. All hero power variance is match-bound [committed] — no persistent power, no purchased power, no gacha.
• Card system shape: a composite, one price per arm [provisional] (2026-08-26/27, "written in pencil not stone") — lane-state cards on Shape A (cycle cost), the imbue arm on Prep + Sealed Beats (read risk, and the imbue now lives on the hero as an aura via a talent pick), the jungle arm on Claimed Grammar (affinity charge). A won/defended event pays a small on-scoreboard bonus, existence only — currency and size still parked. ⚠ The affinity charge is banking-as-a-slot; parked, and 09 stays shelved.
• Cards must be judgeable against visible battle state. The bottom bar is the decision substrate, not a HUD.
• No comeback mechanic may reward deliberate losing [committed]. Preventive beats restorative: a device that stops a snowball starting gives a thrower nothing; one that pays out in proportion to how badly you're doing is what a thrower farms.

Open questions that need the user

• ⚠ Is the accrual cost-gate the same system as banking (09)? "I did mean accrue" is confirmed, but its stated job is gating the cost of powerful effects, not 09's tempo-vs-investment, and it was held as a hypothetical. Either 09 returns, or there are two accrual systems and 18 must know it. A shelved ticket whose mechanic is in use under another name should not persist.
• ⚠ What the forward structure DOES. If creeps spawn where the outer towers would be rather than at the main base, what is the main base for? — open, not a defect. And the structure itself is unspecified: the user rejected his own two candidates (guards "just makes it a tower"; vicinity buffs are "just a defensive tower structure, just a different kind"). His test governs any replacement: it must "change how it makes your units interact with the game, not simply just make it take your units longer to get to the core." Four firstmate proposals are parked in 05 awaiting his reaction — none is live. Separately, the turret is still rejected as MOBA copying and the pseudo-tower brief stands: a lane breakpoint with a power curve that doesn't read as a building (pacing, untouched by the tower-chain rejection).
• Where execution skill lives, or whether the design accepts having none. Skill is now entirely cognitive. Is that the identity or a hole? Don't fill it reflexively — re-adding dexterity under another name would undo the removal.
• Is hero-only-after-exhaustion a loss, or just an endgame state? "then all you have is your hero. I don't know." Unanswered.
• Does the reinforcement pool go per-lane? One per side "as of right now"; per-lane is a live alternative conditional on hero position becoming more manually manipulable (15). A dependency, not a decision.
• Target resolution within a class — whole lane, or a point inside it? A per-lane pool would raise the stakes on this.
• Whether any gesture survives anywhere. Tap-only is live and unchosen.

Prototype

.scratch/core-concept/prototypes/ holds a playable feel prototype. Serve it:

cd .scratch/core-concept/prototypes && python3 -m http.server 8931

index.html symlinks the current version, so edits show on reload. Open it from a phone over LAN or Tailscale — it is a touch game and does not read on desktop.

The current one is the whole-as-envisioned build, corrected 2026-08-13 against his play-test and honed 2026-08-22. Read that directory's README.md first — it lists what to react to and what is deliberately crude, and it points at the commitment card, the delegated centre-screen decision, the finish-gate findings, and the hone pass's own findings (HONE-overgrowth.md) that sit beside it. Real-device frame measurement is still not met — see that file's own accounting.

Hard-won gotchas

• Creeps are units, not a meter. A tug-of-war / fill-bar abstraction has been rejected twice. The v1 prototype shipped one anyway; *.v1-tugofwar.html.bak is kept as the record. The word "tug of war" is not the ban — the user uses it for the strategic contest. The abstraction is what's rejected.
• Don't over-read rough artifacts. A napkin sketch answers only what it was drawn to answer. A super-rough layout sketch was once read as evidence the jungle didn't fit; the user had decided no such thing. Retracted.
• Don't over-extend a cascade. When a decision is removed, scope its consequences to what it actually touched. Removing player card skill shots did not shrink terrain's job — terrain is paid for by hero and creep movement, pathing and sightlines. Corrected by the user.
• Watch for reference-vs-proposal in quotes. The user describes other games to explain a shape. "Your selectable army" was Warcraft 3's UI, not a request for commandable units.
• Don't rebuild DOM inside the animation loop. v1 rebuilt the hand every frame, restarting CSS animations 60×/sec — cards flickered, taps missed, and the prototype was unusable. Board rendering belongs on canvas.
• A dark board breaks the finish gate's dead-block test, and re-tinting it won't fix that. When the jungle stopped being claim-tinted, its whole palette sat inside ~12 points of L*, so no colour change inside that band could raise the luminance variance the gate measures — one attempt made it strictly worse (9.5% → 26.2% dead, a hard fail). What worked was structure, not colour: size the growth so the floor shows between crowns, and draw the second grain scale big enough to survive the downscale. Measure the block, don't eyeball it.
• Board legibility is a value-tier budget, not a palette. The corrected build reads because it has four separated tiers — canopy L* 9, jungle opening 17, their bone 55, your bone 85. Adding a fifth territorial wash is what made the shipped build unreadable to him.
• A syntax check is not a test. Both prototype bugs were runtime-only. Say "unverified" when no browser is available.
• Prototypes need a guided first run and a passive-by-default AI. Without them, any reaction about feel is confounded — v2 read as "very unintuitive" purely because there was no tutorial and the AI crushed the player instantly. That was wrongly recorded as design evidence and had to be retracted. Label crude placeholders as crude on screen.

Maintaining this file

Keep this file for knowledge useful to almost every future agent session in this project. Do not repeat what the codebase already shows; point to the authoritative file or command instead. Prefer rewriting or pruning existing entries over appending new ones. When updating this file, preserve this bar for all agents and keep entries concise.