# Memory Index **Memory holds no fact you can look up.** Version, roadmap, protocols, architecture, what shipped, test counts — read `APP_VERSION`, `QUEUE.md`, `CLAUDE.md` + `rules/`, `library/CODE_MAP.md`, `CHANGELOG.md` + git log. Memory holds preferences, decisions and their reasoning, rejections, incidents, and hard constraints. Rewritten 2026-07-20 after an accuracy audit; superseded versions are in git history on purpose. ## ⭐ Read first - **[⏳ Pending: re-ask about L on next "continue"](pending-L-eyeball-reask.md) — the owner (away) hasn't eyeballed the phone queue view; surface L before advancing on his next "continue". Delete once confirmed.** - [RobCo U.O.S. project](robco-uos-project.md) — what it is, where it lives, standing instructions, recurring gotchas. - [Folder & repo naming](folder-and-repo-naming.md) — ⛔ the app repo is NEVER renamed (installed PWAs bound to its Pages origin; the folder/repo mismatch is deliberate) · ⚠⚠ `_RobCo-Archive` (private forever) and `_RobCo-Exhibit` (public) must never converge. - [Dispatch autonomy — auto-advance, don't ask](dispatch-autonomy-preference.md) — reports are not permission requests. - [Dispatch session-launch discipline](dispatch-session-launch-discipline.md) — fresh session per task; on a launch timeout, stop and wait for "resend". - **[Task kind = network isolation](dispatch-task-kind-network-isolation.md) — ⭐ git/push/gate/deploy MUST be a CODE task (start_code_task, on-machine). start_task (cowork) is a walled sandbox: no GitHub, no gh, no Chromium — it can't even commit. Cost a wasted session.** - **[Mid-flight messages don't reach running sessions](dispatch-midflight-messages-unreliable.md) — ⭐⭐ "Message delivered" ≠ read. A BRIEF MUST BE COMPLETE AT LAUNCH; correct after idle, never assume a mid-run send landed.** - [Always acknowledge actions](always-acknowledge-actions.md) — never act silently; the 4 W's, every time. - **[Plain text is invisible on mobile](plain-text-invisible-on-mobile.md) — ⭐⭐ ONLY SendUserMessage reaches the owner. Prose written outside it is never seen. Flagged 4+ times. Every substantive statement goes through SendUserMessage.** - **[Session reporting standard](session-reporting-standard.md) — ⭐⭐ 3rd correction. POLL until idle after every launch; "yo" must never be what triggers a report. And report IN FULL — every finding, the reasoning, what the session REFUSED and why, what it left alone, where it corrected the brief. Under-reporting costs the owner the ability to steer.** - [Time & work attribution](time-and-work-attribution.md) — never assert the project's age; derive it from git log. Inflating timescales biases advice. ## Governance & workflow - **[G review — adopted practices](g-review-adopted-practices.md) — ⭐ the blind 3-model workflow audit (2026-07-22): the CONTROL PLANE is the weakness, not the build loop. #1 fix = scope the push-gate lint off `eslint .`. Spare laptop DEFERRED (software-fix-first). Spec receipt + release receipt + proposals-are-hypotheses adopted.** - [The four-stage trim plan](trim-stages-plan.md) — staged trim over amputation; stage 2 waits for evidence of use. - [The audit gap + workflow fixes](workflow-audit-gap-and-fixes.md) — audits start from the DIFF, never a summary. ⚠ amendment still owed. - [Engineering metrics log](engineering-metrics-log.md) — measure workflow, not gates; never relay a session's dramatic finding as fact. - [Self-improving code](self-improving-code.md) — fix a CLASS, then add a guard so it can't recur. - [Opus plans, Sonnet implements](opus-plan-sonnet-implement-workflow.md) — Fable is the design/creative model. - [Session-management rule](session-management-rule.md) — serialize pushes, no overlapping edits, no swarm. - [Dev-branch workflow](dev-branch-workflow.md) — the `--no-ff` lesson; why staging is a separate origin. - [Changelog rules](changelog-rules.md) — env-aware viewer, hotfix bucket, final-state consolidation. - [Parked-task workflow](parked-task-workflow.md) — "park" means never start until the owner says go. - [Delete leftovers rule](delete-leftovers-rule.md) — sweep junk at the end of every task. - [UTF-8 write safety](utf8-write-safety.md) — never PowerShell-write non-ASCII on this machine. - [CI/Node parity](ci-node-parity.md) — the owner develops on Windows; a Linux-only CI validated a platform nobody uses. - [External-AI prompt delivery](external-ai-prompt-delivery.md) — always a fenced copy-paste block; attach referenced files. - [Planning + prompt library](prompt-library-campaign-framings.md) — adopted campaign framings and where the library lives. ## Hard constraints & principles - [Project must stay free](project-must-stay-free.md) — no paid infra; AI stays BYO-key. *"this project must remain free."* - [Player-control principle](player-control-principle.md) — the player has full authority; AI is opt-in, never authoritative. - [Primary users + devices](primary-users-devices.md) — the real test surface is Android Chrome / installed PWA; brother is max priority. - [Owner context beyond RobCo](owner-context-beyond-robco.md) — he also builds NCLEX nursing study guides (the "studying" Claude use); a SECOND Claude-heavy workload sharing his usage cap with RobCo. - [Owner templates](owner-templates.md) — keep #updateBanner + the FO3 warning disabled, never delete. Overrides audit findings. - [Melee-scope rule](melee-scope-rule.md) — melee guards gate on weapon OR mode, never mode alone. - [Popup card style standard](popup-card-style-standard.md) — transient pop-ups become compact toasts; persistent stay big. ## Design direction & architecture - [RobCo OS architecture](robco-os-architecture.md) — OS-first philosophy, the taxonomy, the core subsystems, and the rulings that constrain them. - [Design overhaul direction](design-overhaul-direction.md) — the standing design charter; the densest collection of owner quotes in the archive. - [Per-game device form](per-game-device-form.md) — NV terminal / FO3 Pip-Boy / FO4 Mark IV, and the schematic-view direction. - [FO3 Pip-Boy program](fo3-pipboy-program.md) — the locked FO3 design; three Fable passes rejected and why. - [Bezel fidelity pass](bezel-fidelity-pass.md) — reskin, don't rearrange; the owner is almost exclusively on mobile. ⚠ governs the APP's device shells ONLY. - **[The museum has its OWN visual identity](museum-has-its-own-visual-identity.md) — ⭐ NO BEZEL, not fully in-theme. The terminal look belongs to the exhibits, not the walls. Dispatch briefed the opposite once already.** - **[Museum core thesis — the self-maintaining system](museum-core-thesis-the-self-maintaining-system.md) — ⭐⭐ the museum's SUBJECT is the workflow + every measure that maintains it, told as failure→lesson→measure→improvement arcs. "EVERYTHING." Raw materials already exist (bugs/, QUEUE_LOG, graveyard, protocols, audits, memory). This is why the owner keeps raising the Atlas + protocols + a connection-visual.** - [Docked ideas](docked-ideas.md) — the dock, the Diagnostic Shell, and why cheats never reach production. - [CLI terminal ideas](cli-terminal-ideas.md) — the command-set design intent; SELF-DESTRUCT was cut, don't re-add. - [Idea: secret figure variant](idea-secret-figure-variant.md) — the alternate STATUS figure and its six invariants. - [Free Firebase ideas](free-firebase-ideas.md) — the no-server verdict; co-op is affordable, conflict handling is the hard part. ## External reviews & verdicts - [Architecture review](architecture-review-2026-07-13.md) — the UI↔services cycles; a framework would not have fixed them. - [Deep systems review](deep-systems-review-2026-07-13.md) — the gate verifies shape, not behavior. Make green mean the user can see it. - [Reframing review](reframing-review-2026-07-13.md) — a protocol caused the blind gate. "Green is scoped evidence." - [Audit-completeness gaps](audit-completeness-gaps-2026-07-14.md) — it worked because the gaps were CHECKABLE. - [Local AI (Qwen) — REJECTED, do not reopen](local-ai-delegation-verdict.md) — ripgrep beat it on speed and accuracy. Includes the Dell rejection. ## Programs closed (do not re-run) - [Data-provenance program](data-provenance-program-complete.md) — guards pin numbers, not citation labels. - [FO3 karma rebuild](fo3-karma-rebuild.md) — the invented "Enclave hit squad"; correct-then-cite. - [RobCo audit program](robco-audit-program.md) — the origin program; the densest set of incident lessons in the store. - [Main-revert cloud-save compat](parked-mainrevert-cloudsave-compat.md) — closed; in a rollback the danger is the save-back, not the load. - [Post-2.8.0 brain dump](post-2-8-0-brain-dump.md) — written; the five locked prompt rules survive as standing doctrine. - [Synthesis masterplan](parked-synthesis-masterplan.md) — closed, its inputs never all existed. QUEUE.md does this job. - [Roadmap parked placement](roadmap-parked-placement.md) — placement REASONING only; QUEUE.md owns the mapping. - [Roadmap re-baseline](parked-roadmap-rebaseline.md) — ⚠ Web Workers ruling and the Dusk/Dawn re-add are still open. ## Closed briefs — kept for the doctrine inside them - [Diegetic audit](parked-diegetic-audit.md) — ⭐ the CLARITY VETO is live: *"in theme but not confusing."* Do not rename inventory to manifest. - [NV overhaul design audit](parked-nv-overhaul-design-audit.md) — ⭐ per-instrument variation is UNIQUE ON PURPOSE. Do NOT homogenize. - [UI consistency audit](parked-ui-consistency-audit.md) — ⛔ OVERRULED by the ruling above. Only density/contrast/tap-targets survive. - [AI feature eval](parked-ai-feature-eval.md) — the six-axis AI cost model; AI is a typist, not an authority. - [Capability survey](parked-capability-survey.md) — the RobCo design lens and the ideation rubric. - [Test-strength audit](parked-test-strength-audit.md) — passing ≠ catching; presence guards pass after the behavior breaks. - [Code-quality audit](parked-code-quality-audit.md) — the no-build load-time-vs-build-time tradeoff. - [Cloud audit](parked-cloud-audit.md) — a silently unsynced field is the top-severity failure mode. - [Mobile audit](parked-mobile-audit.md) — a trickle of one-off mobile bugs is the tell for a systematic pass. - [Accessibility audit](parked-accessibility-audit.md) — WCAG 2.1 AA is the standing target, without sacrificing the CRT aesthetic. - [Performance audit](parked-performance-audit.md) — the bar is how fast the terminal FEELS, not synthetic scores. - [Content audit](parked-content-audit.md) — fallout.wiki is the only source; never pin UNVERIFIED as canonical. - [VATS/offline audit](parked-vats-offline-audit.md) — ran clean; native features must work with zero network. - [Token-usage audit](parked-token-usage-audit.md) — ⛔ diegetic code naming was rejected outright; don't re-propose. - [Queued fixes](queued-fixes.md) — closed as history; two owner requests survive (cloud version history, small-feature audit). ## Parked — never start without an explicit go - [Fallout 4 support](parked-fallout4-support.md) — the only genuinely-still-parked program. Full data parity, not a stripped port. - [Deferred-work program](parked-deferred-work-program.md) — the Hardware Life no-persistence invariant + the metaphor-lane rule. - [2.9.0 feature update](parked-2-9-0-update.md) — the merge/consolidation rulings; QUEUE.md owns the slate. - [WASTELAND UPLINK engine](parked-wasteland-uplink-engine.md) — the hard no-persistence invariant. - [Phone access to the archive](parked-phone-archive-access.md) — ⚠ blocked on a keyboard. Route A written out and ready. - [Coverage mind-map / SYSTEM ATLAS](coverage-mind-map-deliverable.md) — the split decision; Gource ruled out, don't re-propose. - [AI→native + oversight audit](parked-ai-to-native-feature-audit.md) — *"why is use an ai command?"* Part 2 may still be owed. - [Command-list cleanup](parked-command-list-cleanup.md) — remove obsolete commands; genuinely still parked. - [Mistify salvage audit](parked-mistify-salvage-audit.md) — only DECODE survived. ⚠ still waits for an explicit go. - [NV test save fixture](nv-test-save-fixture.md) — the two-gate split (staging vs minigame) is a real constraint. - [App Check](appcheck-enforce-reminder.md) — enforcement is DONE, nothing pending. Kept for the enforce-early reasoning and the localhost-only debug-guard rule. - [Phase 6 backlog](robco-phase6-backlog.md) — ⛔ un-actionable: the referenced spec path no longer resolves. Do not reconstruct. - [Recreation fun prompt](parked-recreation-funprompt.md) — the six-model collaborative remake. Dead last.