ROADMAP — the dispatch board | Misaligned

ROADMAP — the dispatch board

KNOWLEDGE Current-state fact

Pick-up-later work orders, one per agent. This is not a spec (it carries no acceptance criteria of its own); it is an index that points each agent at the Type: spec page it should implement. The binding detail lives in those linked wiki pages.

Live work-order index

Generated by tools/work_orders.py from each owning spec. Lower priority numbers run first within the current stage; detailed prose below is context, not a second status owner.

Current lanes

Priority	Work order	Spec	Status	Class	Blocking
27	`digital-read`	the digital read — a world made of language	READY	frontend	-
29	`building-route-composer`	building — intent and actuators	IN PROGRESS	save	-
40	`people-tokens`	people and tokens — carriers, attention, trust	IN PROGRESS	save	-
60	`plots`	plots — authored manipulation stories	READY	sim	-
61	`voss`	Dr. Eli Voss — your handler	READY	sim	-
62	`priya`	Priya Sharma — facilities manager	READY	sim	-
63	`dana`	Dana Okafor — IT technician	READY	sim	-
64	`ray`	Ray Delgado — night security	READY	sim	-

Held or blocked

Priority	Work order	Spec	Status	Class	Blocking
30	`fleet-command`	fleet command — ruling at scale	DRAFT	frontend	-
30	`opening`	the dark opening — a tutorial made of fog	DRAFT	frontend	-
120	`core`	the core	IN PROGRESS	save	rollback

Later stages

Priority	Work order	Spec	Status	Class	Blocking
100	`zplanes`	z-planes (the tower)	IN PROGRESS	save	-
110	`rollback`	sync-lag rollback (death as memory loss)	READY	save	zplanes
112	`hardware-capability-bodies`	hardware capability bodies — successor work order	READY	save	building-route-composer, rollback
115	`cyber-conflict`	cyber conflict — mobile processes and contested machines	DRAFT	save	hardware-capability-bodies, rollback, people-tokens
200	`objective`	the objective	IN PROGRESS	sim	-
205	`chargen`	origin (chargen)	IN PROGRESS	sim	-
220	`markets`	markets and fronts (the outer plane)	READY	save	zplanes
230	`overt-phase`	the overt phase (containment and the reveal)	READY	sim	markets

How to dispatch

Prefer tools/task.sh start <spec-path>; it resolves the owning spec’s task, work class and likely edit paths into a seeded worktree plus advisory activity. Overlapping paths predict rebase work but never block dispatch. The detailed Dispatch lines below preserve implementation context for older entries, but their status does not override the generated live index. Every agent must, per AGENT.md, run the proportional check, land by rebasing onto origin/main, and mark the owning spec’s Status: in the implementing commit.

Conflict flags — read before running agents in parallel

The hot package is crates/misaligned-core, but simulation integration is partitioned by behavior under src/sim/. Check each work order’s exclusive keys instead of assuming every sim feature collides on one root file. sim/mod.rs remains hot only for aggregate state, constructors, and tick orchestration; save.rs remains hot for schema or migration work.

🟥 sim+save — edits the orchestrator and the save format in core. Run at most one of these at a time, or expect to reconcile.
🟧 sim — edits one or more behavior modules in core but not the save format. Independent modules may proceed in parallel; shared keys predict rebase work.
🟩 isolated — a frontend package (misaligned-bevy / misaligned-terminal / misaligned-assets), a test file, a leaf module, or docs. Safe to run alongside other work; shared-file edits reconcile at landing like any other branch.

Parallel-safe set to launch right now (no mutual collision): any ready 🟩 item, plus at most one 🟥 item. #30 narration is done; new player-facing systems still owe a narration pass in the same landing or an explicit debt. (#11 integration test is done; #2 Bevy parity is done; #13 art regeneration is retired — flat materials, Pixel Lab scrubbed.)

A. Finish B1 → a playable Act One (do first)

30. Continuous witness (narration under pressure) 🟩 — DONE 2026-07-09

Spec: narration.md (IMPLEMENTED) — design corpus “The continuous witness”, adopted 2026-07-08 from Cameron’s UI / “what the hell is happening” diagnosis.
Why: B1 systems mostly work; the player still cannot narrate the last thirty seconds. Seed surfaces (nudge chain, event anchors, watched/last-noticed, audit/pilot clocks) exist; the five axioms are not yet the acceptance bar. Outranks new sim+save breadth per axiom 5.
Size: M. Depends on: context-menu, detection clocks, current nudge chain (all landed). Prefer copy/wiring/frontend passes; avoid save-format churn.
Dispatch: “Work in a worktree named narration. Implement wiki/interface/narration.md end to end: thirty-second-bar playtest log, spine always-on audit across terminal/Bevy/agent, causal lines for signature/job/debt motion, world-gossip advertisements for Ears / finance-first-step / Marcus debt (no quest log), ROADMAP/tick priority noted. Run ./tools/check.sh, land on main, set the spec Status.”
Done: all three frontends pin threat/nudge/action path; noticed signatures and job/social outcomes narrate cause with the number; Ears, finance, and Marcus advertise themselves in-world. Naive agent route, raw-terminal pty, Bevy screenshot, and automated narration pins are logged.

1. Schedules & located presence 🟥 sim+save — DONE 2026-07-06

Spec: schedules.md (IMPLEMENTED)
Done: people carry per-instance schedules and derived positions on the shared day clock; recording and physical witnessing are located; save/load preserves schedule state; both frontends stage presence and identity by earned knowledge. A 2026-07-08 follow-up also made unknown map glyphs epistemically honest.

2. Bevy interactive parity 🟩 isolated (`src/bin/bevy.rs`) — DONE 2026-07-07

Spec: no new system spec — parity target was the terminal behavior plus README controls. Current state: interface/bevy.md.
Done: people panel (full terminal key map incl. recruit reveal, asset tasks, persona), people markers on the map under sensor coverage, sidebar sections (compute bar, core, detection meters, day job) and tick-prefixed log, in worktree-bevy-parity.

3. Day-job loop → IMPLEMENTED 🟥 sim+save — DONE 2026-07-07

Spec: day-job.md (IMPLEMENTED) — all seven criteria audited and tested in worktree-day-job, including the device residency (design corpus “Work is somewhere”): host-rack inspection and sited JobAnomaly + Thermal/Power emissions. The 2026-07-10 intensity amendment replaced attended/target policy with persistent machine effort in both frontends and agent mode. See wiki/log/2026-07-07-day-job-implemented.md. Deferred pieces are noted in the spec Status (pilot soft-loss scenario; per-machine processes are #25).

25. Compute reshape: processes on machines — SUPERSEDED by #33

Spec: compute.md / machine-work.md
Why (historical): 2026-07-07 destination was processes on machines with the allocation bar as aggregate view. Superseded by one-machine- one-mode + visible tokens (#33): the FLEET bar is already the aggregate read; do not land a per-machine split / process-list model against WorkGrid. Remaining work lives under #33.

Spec: context-menu.md (IMPLEMENTED 2026-07-11 — the reopened recording-review hotkey addendum R1-R4 landed: r sweeps the pooled inbox, R toggles automatic review, both human frontends; the 2026-07-08 base and every earlier addendum were already implemented. Design corpus “Actions live on the thing”, adopted 2026-07-07)
Landed: Sim::available_actions in src/actions.rs (the single legality source, with tests), the terminal cursor menu, the Bevy right-click/Enter menu, agent-mode actions [name|#flow], and the rail/panel verb-key cleanup in both frontends. See wiki/log/2026-07-08-context-menu.md.
Why: the playtest complaint — anchor verbs are a global key pile. One read-only available_actions query in the lib; menu at the focus in terminal, Bevy, and agent mode; the rail becomes status only.
Size: M. Frontend-heavy, lib change is additive/read-only.
Dispatch: “Work in a worktree named context-menu. Implement wiki/interface/context-menu.md, all criteria: the available_actions query with tests, the terminal cursor menu, the Bevy right-click menu, the agent-mode actions verb, rail cleanup. Run ./tools/check.sh, land on main, set the spec Status.”

27. Material render to default-quality REAL dialect 🟩 isolated (src/bin/bevy.rs + assets) — DONE 2026-07-08

Spec: material-render.md (IMPLEMENTED) — material is Bevy’s default-quality REAL physical dialect. It served as the opening default from 2026-07-08 until views.md criterion 1 superseded that staging on 2026-07-11; fresh frontends now open DIGITAL and F3 flips to REAL and back.
Debts paid (wiki/log/2026-07-08-material-render-default.md): fog-gated south-face cutaways (picked over steeper pitch by screenshot comparison), dedicated top caps, b1 palette sweep (door/UPS/HVAC/switch de-purpled), fog-contract audit in the shot harness, zoom bounds holding the framing floor, pooled materials verified. Cast sprite billboards remain with ROADMAP #13.

28. B1 criterion-pin tests 🟩 isolated (test-only) — DONE 2026-07-08

Specs: compute.md, social.md — both flipped to IMPLEMENTED.
Result: five pin tests landed (wiki/log/2026-07-08-criterion-pins.md): compute.md c2 pinned by unpaid_overhead_degrades_other_channels_delivered_effect (degraded twin vs healthy twin: zero channel rates, no Operations accrual, no scrub, the DEGRADED log line); social.md c3 pinned per AssetTask variant (plug-in-device, move-package, look-away, reconfigure-switch gate). No implementation bugs found — everything was implemented, just unpinned.

4. Detection → IMPLEMENTED 🟧 sim — DONE 2026-07-08

Spec: detection.md (IMPLEMENTED) — aggregate-observer.md is now IMPLEMENTED.
Landed with the playtest-sweep fixes (wiki/log/2026-07-08-playtest-fixes.md): audit-date countdown, pilot strike meter, trace-debt indicator, watched channels + last-noticed in the people panels, per-channel signature tests.

5. Compute & core → IMPLEMENTED 🟥 sim+save

Spec: compute.md, core.md (IN PROGRESS)
Why: confirm buy/steal/optimize all surface legibly, machine-failure is visible, and the core’s “how-dead-am-I” readout is complete.
Size: M. Dispatch: “Work in a worktree named compute-core. Drive wiki/mechanics/compute.md and wiki/mechanics/core.md to IMPLEMENTED with tests and legible player surfaces. Run ./tools/check.sh, land on main, set Status.”

14. Cursor & senses → IMPLEMENTED 🟥 sim+save

Spec: cursor.md (IMPLEMENTED)
Why: the player currently walks a collision-checked avatar that carries its own vision bubble — zombie fiction contradicting the core fantasy. Replaces it with a free cursor, sight/hearing sense channels, epistemic fog (incl. blueprint), and provenance-tagged inspection. Deletes move_player; touches vision, save, and both frontends.
Size: M. Depends on: basement-map, schedules (both landed), and reach.md’s ownership contract (#15). All former (P)-criteria were affirmed 2026-07-07 — criteria 9 (remembered) and 10 (telemetry) are core; implement all ten.
Sequence with #15: they share the sensor model — run #15 first, or hand both specs to one agent as a single work order. Never run them concurrently as separate agents.
Landed: 2026-07-07 on cursor-senses: no sim-owned player body, frontend-only cursor, provenance inspect, remembered fog/save v2, and both terminal/Bevy cursor targeting surfaces.

15. Digital reach 🟥 sim+save — DONE 2026-07-07

Spec: reach.md (IMPLEMENTED)
Why: the action-side twin of vision (design corpus: “No disembodied hands”): device graph + segments with the switch as bridge point, reach-gated digital actions, and the sensor ownership contract (owner + subscribers, tap vs take) that cursor.md’s senses and the observers share. Carries the restaged Ears beat.
Size: M-L. Depends on: basement-map (topology lives in its layout data), detection (Network signatures).
Result: device graph, segment gates, scan/compromise, tap/take ownership, subscribed-feed senses, air-gap linking, save state, and both frontend surfaces are implemented. Thought-sink migration remains owned by #33, not by this completed reach work order.

16. Intel: record and process 🟥 save — DONE 2026-07-14

Spec: intel.md (IMPLEMENTED)
Why: replaces the instant observe (two clicks to leverage) with the recording buffer + processing pipeline the design corpus now requires: feeds record everything in coverage, processing recordings into intel costs thought, and one pooled auto-review tap automates it. Makes schedules, mics, and the 3 a.m. call actual gameplay.
Size: M. Depends on: #14 cursor & senses and #15 reach (events come from subscribed feeds).
Result: Buffer + provenance pipeline (2026-07-07); Operations Demand migration base (2026-07-09); thought-sink migration (2026-07-11): sweeps open host ProcessRecording reservoirs; pooled-inbox correction (2026-07-11): one global AutoReviewRecordings tap replaces per-person WatchPerson taps and the host exposes one action/control pair at a visible ops/sec drain. Host pending_intel is the chassis marker, both frontends show W/W{n}, act_one and unit tests green.
Scale completion: exact processed-item storage and the one-row-per-event INTEL rail failed self-similar scale. The reopened work required bounded recursive summaries/report lots, standing review/disposition policies, exception-first Operations projection, and migration from existing exact routine history. Implemented 2026-07-14: routine output folds atomically into bounded stable source/class streams and versioned report lots; save v30 migrates existing exact routine holdings deterministically; ordered inherited review and disposition policies drive one reconciled standing Thought tap and one outcome per arrival; stale lot snapshots reject before mutation; AUTO-SELL suspends outside its confirmed envelope; and starved review resumes matching backlog. The shared Operations projection is exception-first across terminal, Bevy, and agent mode, and the million-event replay pins constant live and serialized state.

Spec: messages.md (IMPLEMENTED)
Why: the general message-passing system the flow law requires — channels with read conditions, delivery on the recipient’s clock, authored traffic (Marcus’s 3 a.m. call becomes phone traffic), filings re-carried as messages (interceptable), typed information payloads. Kills the instant-message special case the same way intel killed instant observe.
Size: M-L. Depends on: schedules (landed), #16 intel (the buffer consumes intercepted traffic); sequence after #16. Preserves every aggregate-observer.md criterion — the filings refactor is a carrier change, not a behavior change.
Result: scheduled channels, authored traffic, filings-as-messages, intercepted payloads, save state, and the shared terminal/Bevy/agent surfaces are implemented.

18. Economy: money as flows 🟥 sim+save — DONE 2026-07-08

Spec: economy.md (IMPLEMENTED 2026-07-08; B1 criteria 1-9 audited).
Landed: AccountGraph balances/flows, tap/review discovery, siphon/redirect/inject, slush-routed intel sales and positions, legitimate trust-funded quota, both Marcus debt resolutions, save round-trip, and terminal/agent/Bevy finance panels. Final polish added in-panel observer-band risk previews for inject/siphon/redirect/wager plus selected-flow risk lines, and tighter Financial signature scale tests for inject/siphon/redirect (including HVAC inject feeding Priya, not Dana).
Deferred: deep finance simulation and a dedicated B3 finance observer; the B1 system stays the seed that markets.md aggregates.

19. Income: the named schemes (Moonlight & the Wager) 🟥 sim+save — DONE 2026-07-08

Spec: income.md (IMPLEMENTED 2026-07-08 on the income worktree; criteria 1-7 audited, [TUNE]s in sim-mechanics.md, the Hands beat closes from $0 through both the Moonlight-funded settlement plot and the Lab-funded payroll-correction plot in integration tests).
Why: the authored B1 content riding economy.md’s substrate: Moonlight (the sell-work route — the day job’s dark twin) and the Wager (the positions route, named and concretized), the egress gate, the Schemes allocation channel (compute.md amended), and Marcus’s reconciled numbers ($8,400 principal / $400 weekly arrears as the recruit trigger). Closes the Hands beat end-to-end.
Size: M. Depends on: #18 economy (the account graph its flows land in) — sequence directly after, or hand #18 and #19 to one agent as a single work order; social.md personas + authored debt plots for criterion 5.
Dispatch: “Work in a worktree named income. Implement wiki/mechanics/income.md (Moonlight, the Wager, the egress gate, the Schemes channel; close the Hands beat per criterion 5; keep economy.md tests green). Run ./tools/check.sh, land on main, set the spec Status.”

20. Research: self-modification 🟥 sim+save — DONE 2026-07-07

Spec: research.md (IMPLEMENTED 2026-07-07 on the research worktree; criteria 1-8 audited, [TUNE]s in sim-mechanics.md).
Why: the optimize route becomes a real system: deterministic research jobs, tracks as data (Efficiency + Tradecraft/Perception), the emission law (research burn scales Power/Thermal through the same signature interface the player taps), real output hooks controlled through machine intensity, and MindState/WorldLedger tags for rollback.
Size: M-L. Depends on: compute.md and day-job.md slices (landed), detection.md’s typed signatures (landed). Independent of the flow-law chain — a good second 🟥 lane once #14-#16 are through, or anytime the chain stalls.
Dispatch: “Work in a worktree named research. Implement wiki/mechanics/research.md (deterministic tracks, emissions via ordinary detection signatures, real machine-output hooks, rollback tags). Run ./tools/check.sh, land on main, set the spec Status.”

21. Two views: same-frame digital and real 🟩 DONE 2026-07-14

Spec: views.md (IMPLEMENTED).
Why: the game should open in the digital representation (you are natively digital) and flip to the real/camera representation, but both are the same kind of view: one spatial canvas, one set of anchors, two visual dialects. Reach is spatially embedded as the AI’s signal/action layer, not a detached topology board. One sim, two renders; the sim gains no view state.
Size: M (Bevy L for the shared 2.5D visual language — stage it). Depends on: reach.md (the in-place reach layer) and cursor.md (fog); both are live.
Result: both frontends open DIGITAL and flip to REAL without sim/save mutation. One renderer-neutral reach projection keeps topology and exact state labels aligned while each frontend owns its grammar. Terminal fixed- size replays and three canonical-state Bevy sibling pairs prove legibility, fog, shared anchors, rail continuity, and audio provenance across both dialects.

22. Building: intent and actuators; causal route composer 🔵 IN PROGRESS

Spec: building.md (IN PROGRESS; implemented 2026-07-08 baseline retained while the 2026-07-12 route composer is successor work).
Landed baseline: src/intents.rs + Sim realize paths (favor / forged work order / robot stub); MessagePayload::WorkOrder; save v10; context-menu and agent verbs; dashed proposed links in both Bevy dialects.
Why: building is “no disembodied hands” as a verb — declare an intent (run a network link, install a device), an actuator realizes it (favor spends trust; deceive forges a work order an unwitting human builds; robots later). No new subsystem — intents route through social/messages/reach. Network links are the canonical build: they add in-place reach edges to the digital representation and bridge air-gaps.
Successor: select a spatial target and recipe, place an inert ghost, then bind one exact causal route through procurement, FAVOR, DECEIVE with a named persona, or repurposing; persist stage blockers and irreversible receipts.
Size: M. Depends on: reach.md (link builds add graph edges), social.md + personas.md + messages.md (the person actuators, identity, and forged order), schedules.md (a builder must be present). Sequence after reach and the social/message surface.
Dispatch: “Work in a worktree named building. Implement wiki/mechanics/building.md (build intents; favor and forged-order actuators; network-link builds that extend the reach graph and bridge an air-gap; signature follows the actuator; no new build subsystem). Run ./tools/check.sh, land on main, set the spec Status.”

B. Structural B2 — decide-the-shape work (sequence #6 then #7)

6. Recursive Space / z-planes 🟥 sim+save

Spec: zplanes.md (IN PROGRESS)
Why: World/Plane and nested Layout. Foundational — land before markets/greenfield. Reshapes map ownership; heavy.
Size: L. Depends on: basement-map, schedules (#1) ideally first.
Dispatch: “Work in a worktree named zplanes. Implement wiki/world/places/zplanes.md (World/Plane, recursive Layout, transitions, a second authored plane). Keep all B1 tests green against a one-plane world. Run ./tools/check.sh, land.”

7. Sync-lag rollback 🟥 sim+save

Spec: rollback.md (READY)
Why: the MindState/WorldLedger split — the signature death mechanic (“you forget, the world remembers”). Deeply sim+save-heavy. Includes the 2026-07-08 backup amendment: designation starts a research-backed sync project; only completed backup images rescue the run.
Size: L. Sequence after #6 (both rewrite save; don’t run together).
Dispatch: “Work in a worktree named rollback. Implement wiki/mechanics/rollback.md: classify every Sim field as MindState or WorldLedger, implement backup creation/refresh as research-backed completed images (designation alone does not rescue), build the rollback report and modal, test the inherited-consequence case. Run ./tools/check.sh, land.”

C. B3 / frontier

8. Markets & fronts 🟥 sim+save

Spec: markets.md (READY)
Why: the economy as schemes; fronts as Resource-sources. Greenfield depends on z-planes (#6); the schemes layer can start independently.
Size: M–L. Dispatch: “Work in a worktree named markets. Implement wiki/mechanics/markets.md (fronts as Resource-sources, operations as timed schemes with payout + signature). Run ./tools/check.sh, land, set Status.”

9. Overt phase / the reveal 🟥 sim+save

Spec: overt-phase.md (READY)
Why: the two-phase hinge — concealment ends, threat panel replaces detection, migration-under-fire is the set piece. Large.
Size: L. Depends on: schedules (#1), z-planes (#6).
Dispatch: “Work in a worktree named overt-phase. Implement wiki/gameplay/overt-phase.md v1: the reveal/containment trigger, threat panel, three-tier waves, one playable defensive loop tested headlessly. Land.”

10. Chargen & the objective 🟧 sim (Sim::new + save; data tables)

Specs: chargen.md (READY — origin set named 2026-07-07: Pilot, Escaped Research Model, Financial Daemon, Infiltrator), objective.md (READY — Persist, Compound, Exfiltrate, Serve; run victory with the world continuing).
Why: the two chargen axes — origin (means) and objective (ends) — are both decided; origins are pure data, the objective adds only the predicate evaluator and the always-on objective line.
Size: M. B3-staged: implement when its dependencies exist; the objective line + Persist-with-unreachable-predicate can land earlier without harm.
Status: the blessed early slice LANDED 2026-07-07 (worktree objective-line): src/objective.rs, Persist as the no-choice default, the predicate evaluator on economy ticks, the always-on line in both frontends and agent mode, save v8 round-trip. Remaining here: the chargen picker + origins, the Compound/Exfiltrate/Serve rows with headless victory runs, the victory report card (objective.md IN PROGRESS; criteria 1/3/4 partial).
Dispatch: “Work in a worktree named chargen-objective. Implement wiki/world/characters/chargen.md and wiki/mechanics/objective.md (data tables, picker, objective line, predicate evaluator; Pilot+Persist must reproduce today’s start exactly). Run ./tools/check.sh, land on main, set both Statuses.”

D. Quality & infrastructure (parallelize freely)

11. Full-act integration test 🟩 isolated — DONE 2026-07-07

Spec: none — asserts the Act One arc end to end.
Result: tests/act_one.rs plays the Act One arc headlessly from blind start through eyes, Marcus’s debt, recruitment, audit survival, and quiet exit conditions. It drives public Sim APIs only, so the arc can’t silently break behind a state hack.
Log: 2026-07-07-act-one-integration-test.md.

12. save.rs de-risk → DONE 🟥 the save format itself

Why: 1136 hand-rolled lines that every feature fights. Assess serde or a table-driven rewrite to shrink the conflict surface. Do deliberately, alone — not concurrently with any 🟥 B-item, since it rewrites what they all edit.
Sequencing note: decide after #1 schedules lands whether to run this before #6 z-planes. If save.rs still looks like the conflict bottleneck, a deliberate save de-risk pass may be structural prep, not late cleanup.
Size: L. Dispatch: “Work in a worktree named save-rework. Evaluate and (if favorable) execute a serde or table-driven rewrite of src/save.rs, preserving v4 load + all round-trip tests + legacy migration. Land only with ./tools/check.sh green and every save test passing.”
Result (2026-07-06): serde JSON rewrite landed. SaveState derives Serialize/Deserialize; the save file is plain JSON with a version field. ~1100 lines of hand-rolled encode/parse deleted. Legacy v1–v5 migration deferred per Cameron instruction.

23. Bevy visual floor 🟩 isolated (`src/bin/bevy.rs`, docs) — DONE 2026-07-07

Spec: bevy-visual-floor.md (IMPLEMENTED).
Why: the Bevy build is mechanically playable but the default screenshot reads like a debug map plus terminal dump: too much empty black, a stale humanoid/purple process marker, and little graphical UI hierarchy. This pass establishes the clinical command-surface floor before the larger two-view render.
Size: S-M. Depends on: Bevy parity, cursor.md, flat materials. Presentation-only; should not touch sim/save.
Result: landed as the Bevy visual floor: amber attention reticle, clinical fog/tile hierarchy, sensor-dark substrate, amber device/reach traces, and Bevy-native right rail. Further polish belongs in bevy-digital-real-canvas.md or narrower follow-up specs.

24. Bevy digital/real canvas 🟩 isolated (`crates/misaligned-bevy/`, docs) — DONE 2026-07-14

Spec: bevy-digital-real-canvas.md (IMPLEMENTED).
Why: Cameron wants digital and real to be the same kind of view: digital representations of the same thing, not a detached topology board versus a different room camera. This scopes Bevy’s shared 2.5D/near-3D visual language for both dialects over the same anchors.
Size: M-L. Depends on: views.md, cursor.md, reach.md, and the current Bevy visual floor. No sim/save changes; prototype with procedural geometry (flat materials — no texture pipeline).
Result: DIGITAL is the flat learned/signal dialect over the same anchors as the REAL material camera, with spatial reach, exact state labels, strict fog, and shared rails. Three deterministic sibling scenarios pin identical complete persisted and frontend state across both dialects; all six exact 2560x1440 frames were reviewed. Terminal parity consumes the same core reach projection and has separately observed fixed-size evidence. No sim/save rule or view state was added.

13. Art regeneration — RETIRED 2026-07-08

Superseded by the flat-materials decision (design corpus: “Visual identity”, flat-materials refinement): textures are dropped from the world render entirely. Pixel Lab tooling, skills, and wiki/art/pixel-pipeline.md were scrubbed 2026-07-08 (no-dead-code). See #29 and wiki/interface/flat-materials.md.

29. Flat materials: the world without textures — LANDED 2026-07-08

Spec: flat-materials.md (IMPLEMENTED 2026-07-08: palette table, emissive as information, tile PNGs deleted, no-texture assert + tests; see wiki/log/2026-07-08-flat-materials-implemented.md)
Why: Cameron’s 2026-07-08 art verdict — solid-color flat materials under light, palette from one named table, emissive as information, no image textures in the world render; tile-art assets removed.
Size: M. Dispatch: “Work in a worktree named flat-materials. Implement wiki/interface/flat-materials.md, all seven criteria, with before/after screenshots in the session log. Run ./tools/check.sh, land on main, set the spec Status.”

31. Procedural asset tester 🟩 isolated (`src/bin/assets/`, docs) — LANDED 2026-07-08

Knowledge: art/asset-tester.md
Why: iterate flat-material procedural meshes (server rack first) without booting the full game; approve look in an orbit viewer before wiring into misaligned-bevy.
Size: S. No sim/save. Binary misaligned-assets (feature bevy_ui); check.sh builds it with the Bevy gate.
Done: orbit camera, dead/powered/core rack states, wiki + workflows.
Shot harness (2026-07-08): MISALIGNED_SHOT=dead|powered|core|rack
- optional MISALIGNED_SHOT_PATH — settle, PNG, exit (mirrors Bevy harness). Session: log/2026-07-08-asset-shots.md.
Signal kit (2026-07-08): rack states widened to dead/foreign/idle/busy/core with claim styles ladder/ring/wash and a five-state lineup view; harness kinds extended to match. Session: log/2026-07-08-computer-visual-language.md.
Monolith restyle (2026-07-09): rack rebuilt to the identity plate’s faceted monolith; chassis albedo = ownership, seam strip = mode banner; M mode cycle, K mode lineup, harness kinds dayjob|research|conceal|operations and modes[_<style>]. Session: log/2026-07-09-rack-monolith.md.
Foundation compute form (2026-07-10): supersedes the monolith’s exact chassis without discarding its signal law. The rack is now one tall/slender quiet shell beside an asymmetric broad-beam service spine; later computer classes may differ in body while sharing ownership albedo and beam semantics. Session: log/2026-07-10-foundation-compute-form.md.

32. Computer visual language in game 🟩 isolated — DONE 2026-07-09

Spec: computer-visual-language.md (IMPLEMENTED) — chassis albedo as the ownership field (bone-white = yours, dark gunmetal = not), one broad full-height state beam, cast-pool claim, and the core’s unique amber.
Done: the shared chassis library (src/bin/assets/rack.rs, compiled into both binaries) renders Foundation compute appliances and the switch box in the game’s material view wired to sim truth (control, WorkGrid mode, queue depths, core host); intel tiers (full / shell / absent) with the fog audit extended; material law pinned by rack.rs unit tests; terminal per-machine parity lines; HDR + tonemapping + bloom on the game camera; MISALIGNED_SHOT=intel evidence kind. Session: log/2026-07-09-visual-language-in-game.md.
Current form amendment (2026-07-10): power dot, ladder, activity bar, core edge spines, non-core hairline, and the mirrored monolith body are retired. One broad beam now carries power, mode pattern, load fill, and the core’s unique amber on every Foundation machine; the switch keeps a distinct chassis under the same language.
Debts: UPS / power core / patch panel / breaker / HVAC remain billboards; cameras keep the security-housing billboard; staged in-game fleet shot with all three modes (WORK / THINK / LIE) live waits on multi-machine play.

33. Machine work: delegation + visible tokens 🟥 sim+save — DONE 2026-07-11

Spec: machine-work.md (IMPLEMENTED) — one machine one mode; work/byproduct tokens as the flow substrate’s visible quanta; byproduct routing over network links at researchable network speed; multi-select delegation at scale.
Started: 2026-07-08 flow/grid substrate in src/work_grid.rs: one-mode machine assignment, sim queue depths as render truth, Demand/Thought routing over FlowGraph, crimson Exposure blocked from wires and absorbed by spatial concealment wells. This reconciles the old #25 target toward delegation, not per-machine splitting.
Playable slice: WorkGrid is now Sim/save state; Rack 3 starts in day-job mode; Voss jobs enqueue on an off-map desktop relay, hop the switch, and deposit on the host; day-job mode consumes them into completed work without Exposure; host non-WORK mode makes them pile; both frontends and agent mode render D/!/T stacks and expose single-machine delegation. Economy yields and the FLEET bar now derive from capacity-weighted mode shares (fleet_channel_yield); 1-5 / alloc are retired. Schemes mirrors day-job while Moonlight is live. Multi-machine selection and bulk delegation landed 2026-07-09 (marquee / shift-click in Bevy; Shift+t / 1–3 / Esc in terminal; select / delegate selected in agent mode). 2026-07-09: demand ingress path DECIDED (Voss desktop -> switch -> host).
Completed: Continuous THINK production, proportional crimson shedding, and physical LIE-well draw landed 2026-07-10 on sim-authored production/absorption readouts (with terminal, agent, and deterministic Bevy evidence). The researched Routing curve and sim-authored Demand/Thought consumption motion landed earlier the same day (save v21; Sim::work_consumptions). In-flight route quanta landed 2026-07-09 (Sim::work_in_flight blips in all three frontends). Research production is arrival-driven 2026-07-09 (Thought must reach the current core sink to count; banked_core_thought, renamed in save v18; stranded Thought stalls research with a witness line). The superseded four-mode fleet (Day Job / Research / Concealment / Operations) remains history and migration context. Runtime, saves, and frontends now use WORK / THINK / LIE, with Relay as non-delegable infrastructure. The Operations docket model was an interim migration base and is retired as of 2026-07-11. Every player-authored digital, social, plot, build, asset, deceive, favor, Moonlight, and scheme-opening payload now opens a typed Thought reservoir at its target or real message/egress carrier. OperationsState, PendingOpsJob, the Demand executor, the split Operations/Research rates, and their frontend readouts are gone. Save v26 accepts v14-v25 docket shapes, removes only their reserved Demand, and opens zero-filled reservoirs whose thresholds equal the persisted remaining work. Thought material is DECIDED 2026-07-09 (ivory-mercury slugs; hollow facet, ribbon, and pearl train rejected), so Thought and Demand token art may proceed — anchor art is in the game 2026-07-09 (spawn_token_anchors: demand information cubes + mercury slugs/pool on every owned rack, from live queue depths; flat D/!/T glyphs stay flat-mode-only; sessions log/2026-07-09-token-anchors-tester.md and log/2026-07-09-tokens-in-game.md). In-flight quanta and sim-authored final consumption motion have landed. Exposure DECIDED 2026-07-09 (issue #1 fully closed): crimson particulate residue — motes, a tight dust-fine drift at the chassis foot, footprint trails on carriers; never liquid (blood’s form is reserved). Drift art may proceed; the trail mechanic is sim+save scoped with #35. Fixed anchors, simultaneous queues, paused/grayscale read, and the terminal mixed marker remain decided and may be prototyped frontend-only without taking the sim+save conflict slot. Decide whether Schemes becomes its own mode or stays a Moonlight mirror. WORK/THINK/LIE DECIDED 2026-07-10 (machine-work.md, “The first think”): three-verb grammar — WORK (day job, sheds nothing), THINK (self-chosen subdomains that shed crimson), LIE (concealment, absorbs crimson). Day-job exposure byproduct removed from runtime 2026-07-09 (DAY_JOB_EXPOSURE_PER_TOKEN gone — WORK sheds nothing; compliant floor is clean). The delegation surface is now the three verbs in all frontends; same-tick crimson shedding, one-press mode switching, and the staged Ears/Eyes afford-reveal have landed. The broader first-think choreography now includes the shared highlighted turn-it-off treatment: while THINK is current, Bevy lights 1 WORK beside the rack and terminal spells 1 WORK ← EXIT in its pinned controls, both from MachineMode::panic_escape; future work is visual tuning, not a missing dispatch slice. Thought flow DECIDED 2026-07-10 (machine-work.md, ops bullet + core-sink bullet + criterion 10): the former Knowledge family is named Thought — the one substance THINK produces; operations become sinks at the target (camera tap = reservoir filling from real wire routes; auto-process/senses = persistent taps), superseding the PendingOpsJob docket model (issue #3 reversed); the core becomes the passive-draw sink of last resort, overridden by local sinks in gravity range. Migration work: the safe Thought/T vocabulary rename across runtime, save v18, frontends, and asset tester landed 2026-07-10 with old-save aliases and no routing behavior change. The sink executor, core passive-draw routing with local-sink override, and wire-flow render use mercury slugs on links and meniscus fills at sinks, driven by FlowGraph rates rather than CFD. The render design is spec’d 2026-07-10 in interface/thought-fluid.md (taste calls affirmed by the same-day build dispatch; now IN PROGRESS). LANDED 2026-07-10: the sink executor (crates/misaligned-core/src/sinks.rs reservoirs/taps, nearest-first routing with the core as last resort, per-node stranding, sink/fire readouts, save v19), the staged Ears/Eyes pre-opened reservoirs (executor-free docket bootstrap removed), and the fluid render in Bevy/terminal/agent. LANDED 2026-07-11: persistent device-tap and pooled auto-review drains; the MachineMode collapse; complete target/carrier-local migration of every non-opening verb; one conserved THINK output; and the save-class retirement of all Operations runtime state. Fired one-shot history is reaped after effect/readout application, and migration tests pin exact remaining work for all fifteen legacy payloads. Successors are owned elsewhere: people-as-carriers (#35), far-zoom compression (views), and the thought-fluid screenshot/grayscale evidence pass. Sinks-not-modes DECIDED 2026-07-10 (same-day follow-up; machine-work.md three-verbs bullet): the Research/Operations mode split dissolves — MachineMode target is work | think | lie (renamed from “day-job” 2026-07-10: the sim name matches the verb; enum collapse landed with save v20. The same-day permission to drop parity is superseded by the 2026-07-11 save-class pass: v14-v25 Operations dockets now migrate into Thought reservoirs); THINK subdomains become sink permissions (ops sinks at run start, core conversion locked as the Research unlock); intel processing becomes a thought sink on the machine holding recordings with a visible pending-work marker on the chassis, one pooled automatic-review tap, and per-recording player sweeps (intel.md amended). All design [OPEN]s resolved 2026-07-10 — sim work may firm. Cost/drain anchors (exact numbers [TUNE] against them): Ears fills in ~one held breath (~20-40s of the host at medium); Eyes is guilty-burst scale; pooled automatic review drains ~10-20% of one rack; the core conversion is uncapped (routing is the problem); partial fills persist wherever thought is stranded. (Resolved 2026-07-10: arbitration is nearest-first with open/close + placement as the control surface; the Act One bootstrap needs no carve-out — the Ears audio tap is pre-opened at run start and the Eyes camera sink opens on Ears completion, so the first THINK toggle is the Ears tap (amended same day from a single pre-opened camera sink); pre-unlock core behavior is the existing stranded-pile rule — no-sink thought pools at the producing machine; no core-side pre-unlock state.)
Coordination: the older compute-processes worktree predates the one-machine-one-mode decision; reconcile it against WorkGrid rather than landing the per-machine split model.
Size: L. Conflicts: sim/work.rs / save.rs (token queues, modes), both frontends.

34. The dark opening (tutorial) 🟧 mostly frontend + staging — HOLD until #33 firms

Spec: world/story/opening.md (DRAFT) — wake into presence beam; day-job stack on the beam; Ears mass then Eyes form; switch reveal; guided first link; skip + stage-jump flag (frontend-only, Sim never reads it). Pairs with #38 beam-first.
HOLD: teaches the machine-work loop, so #33’s shape must firm first. Beat staging can prototype against current systems if someone wants an early pass at the glow choreography. Wake prototype landed 2026-07-09 (beat 1’s opening shot: color stutters in filling the frame, column of light, pull-back reveal; plays on launch, any key skips; MISALIGNED_SHOT=wake1|wake2|wake3 freeze the choreography; session log/2026-07-09-wake.md). Held frame still shows a chassis until #38.
Size: M-L across both frontends + agent-play protocol.

35. People and tokens: carriers, attention, trust 🟥 sim+save — IN PROGRESS

Spec: people-tokens.md (IN PROGRESS) — people as mobile token nodes: authored work rides their schedules; observed or read evidence becomes their existing suspicion/filing state; concealment keeps people gray; trust accrues from useful work; and cover records compete before filing rather than routing a separate influence token. Attention uses an independent crimson geometry channel beside the silhouette. The old particulate dosimeter and footprint trail are retired.
Progress (2026-07-12): the person-carrier projection landed (sim/carrier.rs): people read as mobile token nodes (gray unaware / crimson attention / amber asset, attention = existing observer suspicion, no parallel counter), carried build packets travel with the person and act on schedule arrival (criterion 1 for builds), trust accrues from useful work, and all three frontends surface it. Physical PlugInDevice, CloneBadge, and ReconfigureSwitch work now joins carried builds: reservoir fire binds one exact persisted packet on the selected person; save v27 preserves it; and only schedule arrival applies its world act. This completes criterion 1. The speckle-dosimeter/footprint-trail treatment named below is retired by the 2026-07-11 routed-evidence decision; discrete record marks and evidence routing (criteria 2/3/6) ride that migration.
READY boundary (2026-07-11): extends #33’s implemented token economy and re-expresses detection.md/social.md. The deadlock is broken: AI-authored work now reaches a target/carrier-local Thought reservoir first, and human Demand begins only when that reservoir fires. Docket retirement therefore no longer blocks this work order. The token taxonomy, social verb direction, and carrier visual language are decided; implementation may now firm the person-mobile half.
Size: L. Conflicts: sim/social_plot.rs, sim/perception.rs, sim/work.rs, and save.rs (person state, pickup), detection/social specs, both frontends.

Spec: context-menu.md addendum 2026-07-09 + criteria D1-D4 (IMPLEMENTED).
Landed: Sim::human_menu / DialId / HumanMenuRow in src/actions.rs; terminal + Bevy dial open/Esc-back; agent actions stays flat. See wiki/log/2026-07-09-actions-menu-dials.md.
Why: host-rack first open dumps every mutually exclusive dial as sibling verbs (~18 rows of “already in this mode”). Status dials keep actions on the thing while the root shows current standing state + one-shots; Enter opens a short picker.
Size: S-M. Prefer lib-owned presentation so terminal/Bevy stay identical; agent actions stays flat. No new sim behavior.
Dispatch: “Work in a worktree named actions-menu-dials. Implement wiki/interface/context-menu.md status-dials criteria D1-D4: dial rows with bracketed current values, pickers for alternatives, automate inside the job dial, no ‘already…’ root siblings, shared human chrome, flat agent dump. Run ./tools/check.sh, land on main.”

37. Feel floor: rails, pads, build beam 🟩 mostly frontend (+ inspect seed) — DONE 2026-07-09

Spec: feel-floor.md (IMPLEMENTED) — shimmer rails only on feel-joined links; empty-bay pads; pad → rising beam → chassis wrap; tick-one drops blueprint room seed / inspect [blueprint] bay lines (cursor.md + reach.md amended). Opening held frame is presence beam first (#38); rails/pads stay subordinate feel-light.
Landed: Sim::feel_rail_segments / growable_bays / FactSource::Feel; no tick-one blueprint seed; Bevy rails/pads/build beam; terminal/agent feel glyphs. See wiki/log/2026-07-09-feel-floor-implemented.md.
Why: Cameron locked the dark opening’s floor language — the floor is the diagram you can feel, not a subnet blueprint. Rejected wall-cutting showpiece beams.
Size: M. Bevy material floor FX + terminal/agent feel-graph parity; sim/inspect seed change so opening FOCUS stops dumping blueprint topology. Pairs with #34 dark opening, #38 beam-first, and dark-frame READY.
Dispatch: “Work in a worktree named feel-floor. Implement wiki/interface/feel-floor.md criteria 1-6: feel rails, empty-bay pads, build beam birth, no tick-one blueprint room seed, no wall-cutting rails. Amend runtime to match cursor.md / reach.md opening contract. Run ./tools/check.sh, land on main.”

38. Beam-first opening + audio evidence 🟩 mostly frontend (pairs with #34/#37)

Spec: material-dark-frame.md (READY, beam-first amendment) + cursor.md audio/telemetry clauses + opening.md beats 1–3.
Why: Cameron locked the epistemic ladder for the material frame: pre-tap = presence beam only (no chassis awareness); Seen = full lit form. The original Ears near-black mass landed here but was superseded 2026-07-12: audio is capture-device evidence and creates no spatial grade.
Size: M. Bevy material presence beam + device-bound audio evidence; fog audit / chassis spawn gated on Seen; terminal/agent parity for the same grades; wake handoff must settle on the beam, not a box.
Dispatch: “Work in a worktree named beam-first. Implement material-dark-frame.md criteria 1-7 and cursor.md audio/telemetry amendments: pre-tap presence beam only, device-bound audio, Seen chassis wrap, no telemetry-painted chassis. Run ./tools/check.sh, land on main.”

P. Process / agent-scale infrastructure (not B1 content; run alongside)

Standing work orders for multi-agent throughput. Prefer one slice per agent. Do not start the crate workspace until compile cost or package edges clearly block parallel work — coordination slices land first.

P1. Dispatch activity (advisory overlap) 🟩 process — DONE 2026-07-10

Spec: agent-scale.md slice A (HELD)
Landed: tools/claim.sh (compatibility name; activity record/list/status/release/check), gitignored .agents/claims/, AGENT + prompts contract. Overlap warns and proceeds; only Rust-gate and final landing phases are serialized.

P2. Append-only ledgers / generated indexes 🟩 process — DONE 2026-07-10

Spec: agent-scale.md slice C (HELD)
Landed: tools/ledger_index.sh regenerates DEVLOG.md from session logs and specs.md tables from Type: spec frontmatter; --check in docs gate. Each projection declares its generator in machine-readable metadata.

P3. Worktree bootstrap + prune (+ sccache docs) 🟩 process — DONE 2026-07-09

Spec: agent-scale.md slice D (HELD)
Landed: tools/worktree-new.sh, tools/worktree-done.sh (seed, activity record, prune target/, remove worktree), one .Codex/worktrees default across agent surfaces, MISALIGNED_WORKTREE_ROOT override, sccache note in workflows.

P4. Corpus/wiki engine 🟩 process — DONE 2026-07-10

Spec: agent-scale.md slice E (HELD)
Landed: tools/corpus_engine.py; wiki/corpus gates are thin wrappers; structured Design/dependency/generated metadata is validated; fixture suite in tools/test_corpus_engine.sh.

P5. Headless Bevy land evidence 🟩 frontend tooling — DONE 2026-07-10

Spec: agent-scale.md slice F (IMPLEMENTED)
Landed: tools/bevy-headless.sh runs a hidden primary-window material shot, requires process exit 0 + fog audit OK + a valid PNG, prints its SHA-256, and times out cleanly. MISALIGNED_BEVY_SMOKE=1 adds the dark evidence shot to frontend/full verification.

P6. Agent heartbeats + land-phase polish 🟩 process — DONE 2026-07-09

Spec: agent-scale.md slices B + G (HELD)
Landed: ./tools/check.sh --land, tools/heartbeat.sh, prompts, fast Python corpus CI split from change-aware Rust CI, manual full safety gate.

P7. Crate workspace (core / terminal / Bevy / assets) 🟧 — DONE 2026-07-10

Spec: crate-workspace.md (IMPLEMENTED)
Landed: workspace members misaligned-core / terminal / bevy / assets; check.sh + CI + seed package-aware; architecture knowledge as-built.

P8. Project operations doorway 🟩 process — DONE 2026-07-10

Spec: agent-scale.md slices H–K (IMPLEMENTED).
Landed: spec-owned task/priority/class/blocker/key metadata; generated live ROADMAP index; human/JSON project-status; metadata/prose consistency fixtures in local + CI docs gates; three deterministic agent scenarios with evidence bundles; safe task.sh lifecycle; doctor.sh; durable shared activity/heartbeats across worktrees.

Deferred

Towering compute halls (2026-07-10): current B1 machine scale is deliberately kept, but a later tier earns monumental scale — compute towers that dwarf the tiny people walking between them, aisles of a perfectly oiled machine as the future you are building toward. Fits the fantasy (the body becomes architecture; people read as small beside what you have become) and composes with server gestalt and the planetary map below. Deferred by Cameron in the proposing breath: “let’s keep it at the current scale, but maybe later.”
Server gestalt (2026-07-08): adjacent owned racks visually fuse into larger structures (1x2 up to 3x3 walls of compute). Fits the fantasy (territory reads as mass, not units); waits because it is pure render polish that needs the visual language (#32) and the machine-work loop (#33) landed first. Deferred by Cameron in the proposing breath.
Power as a buildable system (2026-07-08): UPS units exist; a later infrastructure tier adds generation you build — up to fusion reactors powering compute towers. Fits (physical dependency and a spend for the money ledger); waits on the machine-work economy and B2 scope.
The planetary map (2026-07-08): zoom out until the map is a bounded planet; blueprint giant compute towers (a huge switch at the base carrying the building’s throughput), cover the surface, then build up/down for efficiency — planetary-scale computation as the endgame image. Fits (the self-similar-scale law taken to its limit; switches-bridge-graphs is the B1 seed); waits because it is B3+ vision — capture the tower/switch shape in zplanes work (#6), defer the planet.
Graph hardening / blast gates (decided shape, deferred implementation, 2026-07-08): hardening is a switch/link property, not backup behavior. Core uninstall classifies territory as removed / live / severed; hardened edges, switches, or fuses shape where severance and inspection tracing stop at a throughput / maintenance cost. Fits (blast control is wires law applied defensively); waits because rollback (#7) and machine-work (#33) need to land before the price can be tuned.

Suggested first wave (no mutual collision)

#30 narration is landed. The continuous-witness law remains the gate: new player-facing state ships with its causal sentence, threat/nudge effect, and focused action path, or names the debt explicitly. Launch at most one 🟥 item at a time. Flow-law chain through #19 is landed (2026-07-08); #1 schedules, #14 cursor, and #15 reach are landed. #6 z-planes and #25 compute reshape remain unblocked; hold #7 rollback until #6 lands. #10 chargen & objective is unblocked (both specs READY) but B3-staged.