Backlog: P0 mirror push side DONE (repo live, deploy key, launchd daily job); size math corrected to full-dump 2.58MB/day @180d retention (Option 2)

BACKLOG.md

@@ -11,31 +11,36 @@ Status: TODO / IN PROGRESS / DONE / BLOCKED.
 
 ## P0 — Next build
 
-- [ ] **GitHub price-history mirror (start now — warm the data for Trends)** — the
-  relay does the GOV.UK heavy lifting (OAuth token stays in relay `.env`) and
-  pushes a daily snapshot to a PUBLIC repo; the app reads from GitHub, so this
-  doubles as the production data source (App Review can't reach the LAN relay) AND
-  the history archive the Trends graph (P1) reads. Shape: relay writes
-  `latest.json` (tiny pointer: date, station_count, data_updated, available range)
-  + `history/YYYY-MM-DD.json` (full ~200 KB dump) and pushes once/day; the app
-  fetches `raw.githubusercontent.com/aptonline/fuelboard-data/main/latest.json`,
-  then only the missing day files for starred (station, fuel) pairs (deterministic
-  URLs — 404 = no data that day, leave a graph gap, never error). LAN relay stays
-  the fallback source. Telemetry implication: raw GitHub fetches are invisible to
-  repo analytics (page views/clones only) — keep the widget-diag beacon firing
+- [ ] **GitHub price-history mirror (warm the data for Trends)** — the relay does
+  the GOV.UK heavy lifting (OAuth token stays in relay `.env`) and pushes a daily
+  snapshot to a PUBLIC repo; the app reads from GitHub, so this doubles as the
+  production data source (App Review can't reach the LAN relay) AND the history
+  archive the Trends graph (P1) reads. Shape: relay writes `latest.json` (tiny
+  pointer: date, station_count, data_updated, available range) +
+  `history/YYYY-MM-DD.json` (FULL dump — 2.58 MB/day, decision Option 2) and
+  pushes once/day; the app fetches
+  `raw.githubusercontent.com/aptonline/fuelboard-data/main/latest.json`, then only
+  the missing day files for starred (station, fuel) pairs (deterministic URLs —
+  404 = no data that day, leave a graph gap, never error). LAN relay stays the
+  fallback source. Telemetry implication: raw GitHub fetches are invisible to repo
+  analytics (page views/clones only) — keep the widget-diag beacon firing
   regardless of data source (build/stale-build checks survive) and add an
   opportunistic fire-and-forget X-Client app beacon on fetch (LAN = attribution +
   cadence; off-LAN skip is itself a reachability datum). `/stats` keeps the
   intrusion canary and gains a new signal: app-hit spikes = GitHub path failing →
-  fallback active. Setup: create the PUBLIC repo `aptonline/fuelboard-data`
-  (gh installed); GitHub PAT (aptonline) lives in relay `.env` only — never in the
-  repo or IPA. App Review: public repo required (reviewer fetches the data
+  fallback active. App Review: public repo required (reviewer fetches the data
   themselves); HTTPS/ATS fine; plain JSON is not "remote code" (2.5.2-safe); bundle
-  a small fallback dump so the app never demos empty. Growth ~200 KB/day (~73 MB/yr
-  — no pruning for years); git history doubles as an audit trail. Optional
-  first-weeks tweak: hourly pushes to build history faster, then settle to daily.
-  Accumulating TODAY means the Trends graph launches with weeks of real history
-  instead of an empty chart.
+  a small fallback dump so the app never demos empty. Retention: 180 days (~470 MB,
+  pruning enforced by the push script — full dump, NOT the earlier ~200 KB
+  price-only estimate). *PUSH SIDE DONE 2026-08-15 — repo live
+  (https://github.com/aptonline/fuelboard-data); auth = SSH deploy key
+  ~/.ssh/fuelboard_deploy via alias github-fuelboard-data (no token anywhere);
+  clone ~/workspace/fuelboard-data; launchd com.apt.fuelboard-mirror daily 10:00
+  (script ~/.hermes/scripts/mirror_push.py, log
+  ~/Library/Logs/fuelboard-mirror.log); manual trigger `launchctl start
+  com.apt.fuelboard-mirror`; first snapshot 2026-08-15 landed + verified raw 200.
+  REMAINING: app-side read (provider chain GitHub → relay → bundled dump) + the
+  telemetry beacon.*
 
 ## P1 — Soon