--- name: parked-phone-archive-access description: "Getting the private archive + museum onto the owner's ANDROID PHONE. CURRENT PLAN (owner 2026-07-20): publish the museum as a PUBLIC Cloudflare Pages site — with the load-bearing reason it must NOT be GitHub Pages. Route A (MGit + fine-grained PAT) kept intact as the ready-to-hand fallback, blocked only on a working keyboard." metadata: node_type: project type: project originSessionId: 1060dea9-0dc3-42b4-958d-2a6f7a6c0c58 modified: 2026-07-20T16:50:42.642Z --- **Owner, 2026-07-18:** *"save the git client in parked items (as well as museum) so when I get a keyboard I can do this easily."* His Ally's keyboard died mid-session; both routes below need typing (tokens, IDs, settings), so both wait on hardware. ## WHY — the problem being solved The private archive holds `library/`, `planning/`, the museum, and the orchestrator memory. It lives on the Ally and on a **private GitHub repo**. **The owner is phone-first. A museum he can only open at his desk is half a museum.** ## ⚠⚠ CURRENT DIRECTION — OWNER DECISION 2026-07-20. READ THIS FIRST. Owner: *"we won't go the mgit route anymore, we'll go through cloudfire and make it public. when it gets to that point, you'll tell me a step by step guide on how."* **THE PLAN: publish the museum as a PUBLIC SITE on CLOUDFLARE PAGES.** He already runs the private staging site on Cloudflare, so the account and workflow exist. Access becomes a URL on his phone — no cloning, no sync app, no Android file-manager friction. **No email/Access login** — owner: *"I'm not worried about someone reading the generated data, that's the whole point in the museum."* **⭐ WHY CLOUDFLARE AND NOT GITHUB PAGES — the load-bearing reason:** GitHub project sites all share the `zerckzzyHD.github.io` origin, so a museum there would share **browser origin and localStorage** with the live app. The archive contains **executable HTML prototypes**; one opened by anyone could read or overwrite that origin's storage — **where the live campaign lives.** Cloudflare serves from its own domain, so the hazard vanishes. *(Found by GPT-5.6 Sol; the single most valuable finding of the museum review. Do not lose it.)* **⛔ HARD PREREQUISITE — DO NOT PUBLISH BEFORE THIS:** the **memory audit + split** must run first. Every memory file gets classified project / personal / both / stale; the owner decides the ambiguous ones; the store is then **SPLIT AT SOURCE** into a publishable project store and a private personal store. After the split the museum reads only publishable material — **no redaction filter runs at build time** (GPT: *a filter every build is a second product; the boundary is drawn once*). **The owner explicitly WANTS the memories in the museum** — that's much of what makes it interesting — so this is a split, not a removal. Pair it with an ACCURACY audit while in there: **memory has demonstrably drifted** (the index claimed audits were pending weeks after they had all run). **Also NOT publishable as-is:** the existing private archive repo **can never be made public** — git history retains the memory files regardless of later deletion. The public site must be built from a **separate repo containing only generated output**, with clean history. **Still to check before planning around it:** hash-named document URLs rot when content changes; `.nojekyll` / build settings; and verify the deployed stamp + links over HTTP after the first publish. **OWED TO THE OWNER:** a full step-by-step guide when we reach this point. **He asked for it explicitly.** Written on delivery, not now. --- ## ROUTE A — GIT CLIENT (superseded as the plan, but READY TO HAND — blocked only on the keyboard) **Why it was preferred:** GitHub already holds an always-on copy, so his PC can be off. No new service, no new trust boundary, uses infrastructure already holding this exact data. 1. **GitHub token:** github.com → avatar → Settings → Developer settings → Personal access tokens → **Fine-grained** → Generate new. Scope to the archive repo only, read access to contents. Copy it — shown once. *(Android git clients can't use account passwords for GitHub.)* 2. **Install MGit** on Android (open source; Play Store + F-Droid). 3. **Clone** the archive repo. Username = GitHub username, password = **the token**. It's a large clone — do it on wifi. 4. **Open** `museum/site/index.html` via a file manager → Chrome. ⚠ **It is `museum/site/index.html`, NOT `museum/index.html`** — the latter has never existed, and looking for it is the reason the museum has been reported "missing". 5. **Update later:** open MGit, pull. ⚠ **THE KNOWN FRICTION:** opening a LOCAL html file on Android is fiddlier than it should be — some file managers hand it to Chrome cleanly, some don't. **Try a couple before concluding the route fails.** This is the one unverified step. ## ROUTE B — SYNCTHING (fallback) ⚠ **The OFFICIAL Syncthing Android app is DISCONTINUED.** The maintained one is **Syncthing-Fork** (originally Catfriend1, ownership transferred to researchxxl; on F-Droid + Play). **Do NOT install anything just called "Syncthing" on Android.** 1. PC: syncthing.net → run → control panel at `localhost:8384`. 2. Phone: install **Syncthing-Fork**. 3. Pair: PC panel → Actions → Show ID (QR) → scan on phone → accept on PC. 4. Share the archive folder to the phone. 5. ⭐ **SET THE PHONE SIDE "RECEIVE ONLY."** Default is two-way; **a stray phone-side deletion would propagate back and delete from the PC.** This archive is the only copy of things proven irreplaceable. **Non-negotiable.** 6. Exempt from Android battery optimisation or it silently stops and he reads a stale copy — **the same "looks fine, is wrong" failure mode as everything else.** **Syncthing's real limitation:** it is peer-to-peer, so **both devices must be online at once.** It DOES work off home wifi (NAT traversal + encrypted relays) — an earlier explanation wrongly implied local-only. ## REJECTED - **OneDrive** — **it ate his code mid-edit**; it's the origin of the static test suite. **Never propose it.** - **Google Drive** — works, always-available, free tier is ample, but it uploads his private notes + orchestrator memory to Google. **Last resort only, his call.** - **Self-hosted / paid infra** — violates [[project-must-stay-free]]. ## ⭐ MUSEUM CONCEPT IDEAS — OWNER SAID "FOLD THEM ALL IN" (2026-07-18) Dispatch proposed these; owner adopted the lot. Feed them into the museum review prompts AND the build queue. Ordered by Dispatch's judgement of value: 1. **⭐ THE BUG MUSEUM** (Dispatch's pick) — a room of every defect found and what each taught, **each shown NEXT TO the guard that now prevents it.** A single day's haul fills it: the AI that could silently delete parts of a save; a cache-bump guard inert on `dev` for weeks; a test that checked the code *said* it escaped dangerous text rather than that it did; a fake level-up popup that asked permission after already acting. **Why it's the best one: it's the only idea that would change how the owner works** — it makes each guard's cost visible beside the thing it prevents, which is exactly the evidence the trim needs. It also renders the escape-ratchet philosophy as an artifact instead of prose. 2. **INTENT VS REALITY** — the mockup beside what actually shipped. A large body of designs and a live app; **the divergence is more interesting than either alone.** 3. **A DECISIONS ROOM** — not documents: the actual calls, what was rejected, why. **Much of this exists ONLY in orchestrator memory right now.** 4. **"WHAT WAS HAPPENING ON THIS DATE"** — pick a day, see what shipped, what was being planned, which mockups were drawn. **Turns an archive into a diary.** 5. **THE GRAVEYARD AS THE FRONT DOOR** — most projects hide what they killed; **opening with it says something true about how this one is built.** 6. **CLICKABLE GROWTH CHART** — click a point on the curve, see that week. **Turns a shape into a story.** ## ALSO PARKED WITH THIS The **museum's remaining polish**: thumbnails (Playwright, no new dependency), the comparison slider (needs mockups grouped by filename stem — proposed, not built), and the release accounts (owner asked Dispatch to write them **in its own voice, with real project context**, not changelog-derived). Related: [[parked-task-workflow]], [[robco-uos-project]], [[primary-users-devices]].