ukmesh/local-vs-github-pr-plan.md

Local vs GitHub Change Review

Repository: meshcore-analytics (GitHub remote origin → https://github.com/gadgethd/ukmesh.git) Local branch: main · Remote default branch: main Generated: 2026-06-18 · Source of truth: local working tree on disk

Summary

The local checkout sits on main at the same commit as origin/main (69acf7a Add beta path neighbor affinity scoring, 0 ahead / 0 behind). Every difference between disk and GitHub is therefore uncommitted working-tree change, not unpushed commits. The delta is large and spans the whole stack:

• ~73 tracked files modified, 11 tracked files deleted, and ~24 new untracked files/directories (≈ 4,100 insertions / 7,800 deletions in tracked files alone).
• Headline functional work:
  1. Lazy path resolver rewritten from greedy per-position selection to a global Viterbi (max-product) decode, plus a new shared scoring module, an accuracy harness, and beta-resolver multibyte/ML/anchor scoring upgrades.
  2. Message-spam detection ("Spam Watch") — a brand-new backend/src/spam/ subsystem (clustering, sanitization, origin estimation, tests) with a public /spam page; plus the older advert-spoof spam-suspect detector finally landing in the tree.
  3. Region-scope / transport-code decoding for TransportFlood/TransportDirect packets (new DB columns, MQTT decode, stats breakdowns, a one-shot backfill, decoder bump 0.2.x → 0.3.0).
  4. Network-scope unification (ukmesh now means ukmesh + northeast) and the "disappearing nodes" fix via node_network_sightings, plus several query-performance bounds.
  5. Stats dashboard overhaul (route types, transport codes, observer diversity, signal summary, path-decode trend; removed repeaters-per-day metrics).
  6. Map rendering: terrain-aware 3-D path elevation, a multibyte-path overlay, clash path lines, staleness tweaks.
  7. Frontend restructure: shared LoadingIndicator, CSS consolidated into styles/globals.css, dead pages/hooks/utils deleted, new Docs page, nav/SEO updates, an (unwired) Regions page.
  8. Deployment/ops hardening: docker-compose restart: always + log limits + healthchecks + infra port exposure + Redis password + disabled/removed services; nginx cache header; Anubis terrain-tile bypass; viewshed/ML worker changes.

Most of this is coherent, in-progress work that has been running on the VM but never committed. It needs to be split into reviewable PRs and committed, with particular care around a handful of files (db/index.ts, mqtt/client.ts, api/routes.ts, docker-compose.yml) that are touched by several unrelated concerns at once.

Assumptions

• Comparison baseline: I compared the working tree against HEAD, which equals origin/main after git fetch. Because local is 0/0 vs the remote default branch, git diff HEAD + untracked files is the local-vs-GitHub delta. There was no need to diff against a divergent remote.
• Default branch confirmed via git remote show origin → HEAD branch: main. A second local branch docs/harden-codebase-roadmap exists on the remote too but is unrelated to this review.
• Source of truth = disk. Deletions on disk (e.g. frontend/src/globals.css, AboutPage.tsx) are treated as intentional removals to be reflected in PRs, not as accidental loss.
• Files intentionally ignored (generated / noise / secrets — excluded from PR scope):
  • vacuum-compressed-chunks.log (untracked runtime log; should be git-ignored, not committed).
  • backend/package-lock.json / frontend/package-lock.json — reviewed only for the dependency changes they encode; lockfile line-noise is not itemised.
  • Per .gitignore, project context files (CLAUDE.md, AI_MEMORY.md, knowledge.md, push.md, multipath.md), .env, mosquitto/passwd|acl, and the teesside/ frontend tree are not part of this repo's committable surface.
• .env is managed on the VM and never committed; only .env.example (the documented template) is in scope.
• I did not build, typecheck, or run tests as part of this review — risk/test notes are based on reading the diffs.

Commands used

git remote -v
git fetch origin
git remote show origin | grep -i 'HEAD branch'        # -> main
git rev-list --left-right --count origin/main...HEAD   # -> 0  0
git status --porcelain=v1 | sort
git diff --stat HEAD
git diff HEAD -- <path>           # per-file inspection
git diff HEAD --unified=0/1/2 -- <path>
git check-ignore vacuum-compressed-chunks.log          # -> not ignored

---

Proposed PR Breakdown

Several backend files (db/index.ts, mqtt/client.ts, api/routes.ts) contain hunks belonging to multiple PRs below. The plan assumes hunk-level staging (git add -p) when these are split. Where that is impractical, the affected PRs should be merged in the listed order to avoid conflicts.

PR 1 — Network-scope unification + "disappearing nodes" & query-perf fix (backend foundation)

• Goal / rationale: Make ukmesh a logical mesh of ukmesh + northeast, stop nodes flip-flopping off the map because nodes.network is last-writer-wins, and bound several unbounded hypertable scans. Foundational: the new networks.ts is imported by many later PRs.
• Files changed:
  • backend/src/networks.ts (new) — UKMESH_NETWORKS, MESH_RESOLVER_NETWORKS, networkMatchesScope, expandResolverScope.
  • backend/src/api/utils/networkFilters.ts
  • backend/src/db/index.ts (scope-clause hunks only: buildScopePlaceholders / buildNodeScopeClause / buildPacketScopeClause)
  • backend/src/ws/server.ts
  • backend/src/http/requestScope.ts
  • backend/src/health/status.ts (3-day bound on latest_rx / active_rx)
• Summary: Scope params become arrays (= ANY($n)) when ukmesh is requested; node scope now also matches node_network_sightings within 30 days (instead of teesside-only); observer scope rewritten as an uncorrelated IN (...) subquery with a 7-day window; app.ukmesh.com host now resolves to ukmesh scope (was all); WS scope matching uses networkMatchesScope; health query bounded to 3 days.
• Risk: Medium — pure SQL/scope-semantics change with real query-plan implications; a regression silently changes which nodes/packets appear.
• Test plan: Compare node/packet counts for ?network=ukmesh, =northeast, =teesside, =all, and ?observer= before/after; confirm previously-disappearing boundary nodes stay visible; EXPLAIN ANALYZE the observer/node-scope queries to confirm the 7-day/30-day bounds keep plans inside recent chunks; verify health endpoint latency dropped.
• Dependencies: none (do first).

PR 2 — Lazy path resolver: greedy → global Viterbi decode (+ beta resolver scoring upgrades)

• Goal / rationale: Replace greedy per-position relay selection with a global per-group Viterbi decode (the long-standing "make the lazy resolver the most accurate" goal), share scoring with the beta resolver, and add a measurable accuracy harness.
• Files changed:
  • backend/src/path-shared/scoring.ts (new) — single source of truth for prior key-formats and Viterbi weights.
  • backend/src/path-lazy/lazyResolver.ts — rewritten to Viterbi with emission/transition scoring, forward/backward marginals → ambiguous flag, synthetic "unresolved" candidate, ABLATE_LEAKY_PRIORS diagnostic.
  • backend/src/path-lazy/lazyResolverLegacy.ts (new) — verbatim greedy copy, harness-only.
  • backend/src/path-lazy/evaluate.ts (new) — gold-packet accuracy harness (degrades multibyte→1-byte, runs legacy vs Viterbi, stratifies by prior support).
  • backend/src/path-beta/resolver.ts — trusted-path pairs, ML 1-byte-prefix scores, direct-observer anchors, multibyte observation counts, trellis candidate cap; removed dead clashPressure/corridorCheck.
  • backend/src/path-beta/fallback.ts, backend/src/path-beta/types.ts — multibyte boost + MlPrefixScore/trusted-path context types.
  • backend/src/path-learning/rebuild.ts — drop unused destructured var (strictness).
  • backend/src/workers/path-history.ts — window 72h→168h, max hashes 5k→12k.
• Summary: Major resolver-accuracy work; shared weights prevent the two resolvers from drifting.
• Risk: High — core inference path; changes which relays are drawn on every path overlay and the path-history cache.
• Test plan: Run npx tsx backend/src/path-lazy/evaluate.ts ukmesh and confirm the "supported" (leakage-resistant) bucket accuracy ≥ legacy; spot-check /api/path-lazy/resolve and /api/path-beta/* for known packets; confirm path-history-worker still populates the cache within the larger window without timeouts.
• Dependencies: PR 1 (networks.ts / expandResolverScope).

PR 3 — Region-scope / transport-code decoding (ingest + schema + decoder bump)

• Goal / rationale: Decode the 4-byte transport codes on TransportFlood (routeType 0) / TransportDirect (3) packets, tag each packet with its matched region (#Europe, …), and persist it.
• Files changed:
  • backend/src/db/migrations/004_region_scope.sql (new) — ALTER TABLE packets ADD transport_codes TEXT, region_scope TEXT.
  • backend/src/mqtt/client.ts (region-probe + transportCodes hunks) — REGION_PROBE_LIST from MESHCORE_REGION_NAMES, transportCodeMatchesRegion probe, plumb transportCodes/regionScope into insert + WS payload; handleMessage now has a top-level .catch.
  • backend/src/types/index.ts — transportCodes, regionScope; corrected routeType enum comment.
  • backend/src/db/index.ts (insertPacket transport_codes/region_scope columns)
  • backend/backfill-transport-codes.mjs (new) — one-shot backfill for existing route 0/3 rows.
  • backend/package.json — @michaelhart/meshcore-decoder ^0.2.7 → ^0.3.0 (provides calcRegionKey / transportCodeMatchesRegion).
• Summary: New optional packet metadata; everything degrades to defaults (Europe,Global) if env unset.
• Risk: Medium — decoder major-ish version bump and a hot-path change in handleMessage.
• Test plan: Apply migration 004 (additive); confirm new route 0/3 packets get transport_codes/region_scope; run the backfill on a copy and verify counts; smoke-test the decoder upgrade against a sample of stored raw_hex.
• Dependencies: independent, but must land before PR 4's transport-codes stats chart has data.

PR 4 — Stats dashboard overhaul (backend + frontend)

• Goal / rationale: Replace low-value repeaters-per-day metrics with route-type, transport-code, observer-diversity, signal-quality, and path-decode-trend breakdowns.
• Files changed:
  • backend/src/stats/statsRepository.ts, backend/src/stats/statsService.ts
  • frontend/src/pages/StatsPage.tsx (≈ rewritten), frontend/src/components/LiveStatsSection.tsx
• Summary: New ROUTE_LABELS, decodeTransportCodes, and aggregate queries; removes repeatersPerDay / activeRepeaters / staleRepeaters.
• Risk: Medium — large frontend churn; any consumer of the removed summary fields breaks.
• Test plan: Load /stats for each scope, verify all new cards render and totals are sane; confirm nothing else reads the removed fields.
• Dependencies: PR 3 (transport codes), PR 9 (LoadingIndicator).

PR 5 — Spam detection backend: Spam Watch (messages) + advert-spoof suspects

• Goal / rationale: Land the two spam subsystems. They share routes/spam.ts, db/index.ts spam functions, and mqtt/client.ts, so they're grouped (could be split into 5a advert-spoof / 5b message-incidents if the shared files are hunk-split).
• Files changed:
  • Message-spam (new): backend/src/spam/ (analyzer.ts, cluster.ts, config.ts, normalize.ts, origin.ts, repository.ts, sanitize.ts, similarity.ts, spamResolver.ts, types.ts + *.test.ts), backend/src/db/migrations/005_spam_message_incidents.sql, backend/src/tools/recomputeSpamMessages.ts, backend/src/tools/reresolveSpamOrigins.ts, docs/spam-detection.md, .env.example (Spam Watch block).
  • Advert-spoof (new, older system): backend/src/mqtt/spamDetector.ts, backend/src/db/migrations/003_spam_suspects.sql, backend/src/tools/recomputeSpamSuspects.ts, backend/src/db/index.ts (insertOrUpdateSpamSuspect, getSpamSuspects, getSpamSuspectSummary, getSpamPacketObservers, getSpamAllObservers), advert-eval hunk in backend/src/mqtt/client.ts (suppress upsert/emit for verdict==='spam').
  • Shared wiring: backend/src/api/routes/spam.ts (new — serves /spam/messages/* and /spam/suspects|observers|packet), backend/src/api/routes.ts (registerSpamRoutes), backend/src/index.ts (initSpamMessageAnalyzer, await startMqttClient()), backend/package.json (spam:* + test scripts), backend/tsconfig.json (noUnusedLocals/Parameters, exclude *.test.ts).
• Summary: Public API serves only pre-sanitized incident JSON; raw text/coords/sender names stay local-only in the DB.
• Risk: High — adds work to the MQTT hot path, can suppress legitimate node upserts if a verdict is wrong, runs a periodic heavy analyzer, and makes startMqttClient awaited (startup-order change).
• Test plan: npm test (spam unit tests); apply migrations 003 + 005 (note: existing deploys used DATABASE_SKIP_SCHEMA_INIT + manual ALTERs); verify /api/spam/messages/* returns sanitized data only; confirm a legitimate advert still upserts its node (false-positive guard); confirm the analyzer's first pass doesn't block startup.
• Dependencies: PR 1 (networks.ts); coordinate with PR 3 on the shared mqtt/client.ts file.

PR 6 — Private-node redaction

• Goal / rationale: Honour the 🚫 opt-out — replace name with "Private Node", deterministically fuzz coords (~500 m), strip identifying fields in API responses.
• Files changed: backend/src/api/utils/privateNode.ts (new), backend/src/api/routes/nodes.ts.
• Risk: Low–Medium (privacy-positive; verify fuzz is deterministic and applied everywhere a node leaks).
• Test plan: Hit /api/nodes and the peer endpoint with a 🚫 node present; confirm name/coords/keys are redacted on every path.
• Dependencies: none.

PR 7 — Owner portal: 24 h packet counts + LoRa distance sanity filter

• Files changed: backend/src/owner/ownerRepository.ts, backend/src/owner/ownerService.ts, frontend/src/pages/OwnerPortalPage.tsx.
• Summary: Adds packetsSent24h / packetsReceived24h; filters incoming peers with known coords beyond 150 km (LoRa can't reach that far → spoofed/bad coords).
• Risk: Low.
• Test plan: Owner dashboard shows correct 24 h counts; a peer >150 km away is dropped, peers without coords are kept.
• Dependencies: PR 9 (LoadingIndicator).

PR 8 — Map: terrain-aware 3-D path elevation + multibyte-path overlay + clash lines

• Goal / rationale: Render path/link lines at true terrain elevation when terrain is on, add a live "multibyte paths" overlay, and surface clash path lines.
• Files changed:
  • Frontend: frontend/src/components/Map/DeckGLOverlay.tsx, MapLibreMap.tsx, LiveOverlayController.tsx, Map/types.ts (ClashPathLine, LatLonPosition), geojsonBuilders.ts (client vs repeater staleness), mapConfig.ts (terrain exaggeration 2→3, DEM minzoom: 5), store/overlayStore.ts (clashPathLines), hooks/usePacketPathOverlay.ts (drive overlay off latest message, not packet), frontend/src/App.tsx (poll /api/path-beta/multibyte-paths, loading indicators, links:false sanitize).
  • Backend: backend/src/api/routes/pathing.ts (GET /path-beta/multibyte-paths), backend/src/db/index.ts (getMultibytePathSegments), backend/src/api/routes.ts (wire dep).
• Risk: Medium — deck.gl Z-coordinate math and a new backend aggregation query.
• Test plan: Toggle terrain and confirm lines lift onto the DEM with no flicker; verify /api/path-beta/multibyte-paths segments render and respect scope; confirm clash lines appear in clash mode.
• Dependencies: PR 9 (LoadingIndicator); overlaps api/routes.ts with PR 5.

PR 9 — Shared LoadingIndicator component + loading states (frontend foundation)

• Goal / rationale: One spinner component replacing ad-hoc "Loading…" strings across the app. Imported by ~11 files, so land early.
• Files changed: frontend/src/components/LoadingIndicator.tsx (new), frontend/src/styles/globals.css (loading-indicator + app-refresh-status styles), and call-site updates in Map/NodePopupContent.tsx, UKCompanionPage.tsx, PacketDetailPanel.tsx, UKFeedPage.tsx, UKRepeaterSearchPage.tsx (the App/StatsPage/SpamPage/OwnerPortal call-sites travel with their feature PRs).
• Risk: Low.
• Test plan: Visual check that spinners render in block / overlay / inline variants.
• Dependencies: shares styles/globals.css with PR 11 (sequence them).

PR 10 — Navigation / routing / SEO restructure + dead-page cleanup

• Goal / rationale: Add /docs & /spam routes, redirect retired routes (/about, /mqtt, /regions), refresh nav/footer + SEO/sitemap, and delete now-dead pages/hooks/utils/data.
• Files changed:
  • Modified: frontend/src/main.tsx, frontend/src/config/seo.ts, frontend/src/pages/shared/SiteLayout.tsx, frontend/src/pages/ukmesh/UKLayout.tsx, UKHomePage.tsx, UKInstallPage.tsx, frontend/src/pages/OpenSourcePage.tsx.
  • Deleted: frontend/src/pages/AboutPage.tsx, HealthPage.tsx, MqttPage.tsx, pages/dev/DevMqttPage.tsx, pages/ukmesh/UKMqttPage.tsx, hooks/useLinks.ts, hooks/usePathLearningModel.ts, utils/betaLinks.ts, utils/betaPathing.ts, data/ni-ring.ts, data/uk-mainland.ts.
• Summary: UK nav now surfaces Docs, drops Stats/Flasher from the top nav (Stats moved to footer), /regions redirects home (page intentionally not enabled — see PR 11).
• Risk: Medium — deletions; confirm no remaining imports of the removed modules.
• Test plan: grep for imports of each deleted file (expect none); click every nav/footer link; verify retired routes redirect; check generated sitemap.
• Dependencies: PR 5 (SpamPage) and PR 11 (UKBestPracticePage) provide the components these routes point at — land those page components first or together.

PR 11 — New content pages: Docs (Best Practice) + Regions (Regions currently disabled)

• Files changed: frontend/src/pages/ukmesh/UKBestPracticePage.tsx (new, wired at /docs), frontend/src/pages/ukmesh/UKRegionsPage.tsx (new, not wired — /regions redirects home and the component is unimported), frontend/src/pages/ukmesh/regionData.ts (new, only used by UKRegionsPage).
• Risk: Low (Docs) / Low but incomplete (Regions is dead code until routed).
• Test plan: /docs renders; decide whether to ship Regions now (wire it) or hold it out of this PR until ready.
• Dependencies: PR 10 (routing/SEO/nav), PR 9 (LoadingIndicator if used).

PR 12 — Frontend deps / build / CSS consolidation

• Goal / rationale: Remove unused @deck.gl/geo-layers, consolidate styling into styles/globals.css, delete the old root stylesheet.
• Files changed: Deleted frontend/src/globals.css (3,788 lines); frontend/src/styles/globals.css (+927, gains the consolidated rules); frontend/package.json (drop @deck.gl/geo-layers); frontend/vite.config.ts (manualChunks.deck without geo-layers); frontend/package-lock.json.
• Risk: Medium — a CSS regression is easy to miss; confirm nothing imports the deleted globals.css and no code imports @deck.gl/geo-layers (it left with utils/betaPathing.ts in PR 10).
• Test plan: Full visual pass across pages; grep for geo-layers and from './globals.css'; production build succeeds and chunking is intact.
• Dependencies: PR 9 / PR 11 (shared styles/globals.css) — merge after them or fold the loading/page styles in here.

PR 13 — Deployment & ops hardening (compose / nginx / anubis / workers)

• Goal / rationale: Production-grade compose (auto-restart, bounded logs, healthchecks, ordered startup), expose infra ports on the VM, password Redis, retire dead services, harden workers.
• Files changed:
  • docker-compose.yml — restart: always + json-log-limits everywhere; healthchecks + condition: service_healthy deps; infra ports bound via INFRA_BIND_ADDRESS (loopback by default); max_parallel_workers_per_gather=0; Redis now requires REDIS_PASSWORD; commented out ml-path-learner; removed website-teesside, app-dev, website-dev, anubis-website-teesside; cloudflared extra_hosts + external meshcore-beacon-ukmesh network; consolidated Anubis cookie domains to ukmesh.com; link-worker radio-bot URL now opt-in (LINK_WORKER_RADIO_BOT_URL, empty default).
  • nginx.website.conf — Cache-Control: no-cache… on the SPA fallback.
  • anubis/botPolicy.yaml — allow ^/terrain-tiles/ unchallenged (so Cloudflare can cache tiles).
  • viewshed-worker/Dockerfile — pin GDAL to ubuntu-full-3.12.2@sha256:….
  • viewshed-worker/worker.py — radio-bot sync optional, already_calculated gated on elevation_m, neuter backfill_elevations (reverse-radius estimate deemed unsafe), enqueue nodes missing elevation, requests import.
  • ml-path-learner/worker.py — cleanup_training_artifacts (generation-retention pruning of variant results + model artifacts).
• Risk: High — most operationally dangerous PR. Redis password + un-passworded REDIS_URL default is a live footgun (see Concerns); removing services and changing healthcheck gating can stall the whole stack.
• Test plan: Deploy to a staging stack: every service reaches healthy; backend/workers connect to Redis with the password; confirm intended services are gone and nothing references them; Cloudflare serves cached terrain tiles; viewshed worker re-queues elevation-less nodes; ML cleanup deletes only old non-active generations. Note the ML worker is disabled in compose even though its code changed — confirm that's intended.
• Dependencies: none code-wise, but should land after the app-level PRs are committed so a rollback of ops config doesn't strand new features.

PR 14 — Repo hygiene: ignore runtime logs

• Files changed: .gitignore (add vacuum-compressed-chunks.log / *.log); do not commit vacuum-compressed-chunks.log.
• Risk: Low.
• Dependencies: none.

---

Detailed File-by-File Differences

PR column references the breakdown above. "M" = modified, "A" = added/untracked, "D" = deleted.

Backend — core / infra

File	Type	What changed	Why it matters	PR
backend/src/networks.ts	A	Scope constants + expandResolverScope / networkMatchesScope	Single definition of "what is the UK mesh"; imported widely	1
backend/src/api/utils/networkFilters.ts	M	ukmesh → ANY([ukmesh,northeast]), node_network_sightings 30 d, observer IN(...) 7 d	Fixes disappearing nodes + query perf	1
backend/src/db/index.ts	M	Scope clauses (PR1); insertPacket transport cols (PR3); spam suspect fns (PR5); getMultibytePathSegments (PR8); coord-recalc helpers/threshold	Touched by 4 PRs — hunk-split required	1,3,5,8
backend/src/ws/server.ts	M	networkMatchesScope in packet/message/node scope checks	WS respects combined ukmesh scope	1
backend/src/http/requestScope.ts	M	app.ukmesh.com → ukmesh (was all); ForcedScope type	Default site scope semantics	1
backend/src/health/status.ts	M	3-day bound on rx scans	20s+ query → fast	1
backend/src/index.ts	M	await startMqttClient(); init spam analyzer; re-cover nodes missing elevation	Startup ordering + spam + viewshed	5 (with 3,8)
backend/src/api/routes.ts	M	Register spam + multibyte-paths deps; drop unused import	Route wiring	5,8

Backend — pathing

File	Type	What changed	Why	PR
backend/src/path-shared/scoring.ts	A	Shared weights + key formats	De-dupes resolver scoring	2
backend/src/path-lazy/lazyResolver.ts	M	Greedy → Viterbi decode, marginals, ablation flag	Accuracy goal	2
backend/src/path-lazy/lazyResolverLegacy.ts	A	Frozen greedy copy	Harness comparison only	2
backend/src/path-lazy/evaluate.ts	A	Gold-packet accuracy harness	Measures Viterbi vs greedy	2
backend/src/path-beta/resolver.ts	M	Trusted-path pairs, ML prefix, direct anchors, multibyte; removed dead helpers	Beta path accuracy	2
backend/src/path-beta/fallback.ts	M	Multibyte observed boost	Scoring parity	2
backend/src/path-beta/types.ts	M	MlPrefixScore, trusted-path context	Types for above	2
backend/src/path-learning/rebuild.ts	M	Drop unused destructure	Strictness	2
backend/src/workers/path-history.ts	M	72h→168h, 5k→12k hashes	Wider cache window	2
backend/src/api/routes/pathing.ts	M	GET /path-beta/multibyte-paths	Map overlay backend	8

Backend — region/transport

File	Type	What	Why	PR
backend/src/db/migrations/004_region_scope.sql	A	Add transport_codes, region_scope	Persist new metadata	3
backend/src/mqtt/client.ts	M	Region probe + transportCodes (PR3); advert-spoof eval gating upsert (PR5); top-level catch	Touched by 2 PRs	3,5
backend/src/types/index.ts	M	transportCodes/regionScope; fixed routeType comment	Type surface	3
backend/backfill-transport-codes.mjs	A	One-shot backfill of route 0/3 rows	Migrate existing data	3

Backend — stats / owner / spam / privacy

File	Type	What	Why	PR
backend/src/stats/statsRepository.ts	M	New aggregate queries; removed repeaters-per-day	Stats overhaul	4
backend/src/stats/statsService.ts	M	ROUTE_LABELS, decodeTransportCodes, new payload fields	Stats overhaul	4
backend/src/owner/ownerRepository.ts	M	24 h sent/received counts	Owner metrics	7
backend/src/owner/ownerService.ts	M	LoRa >150 km peer filter; expose counts	Data sanity	7
backend/src/spam/* (16 files incl. tests)	A	Message-spam detection subsystem	Spam Watch	5
backend/src/mqtt/spamDetector.ts	A	Advert-spoof signal engine	Suspect detection	5
backend/src/db/migrations/003_spam_suspects.sql	A	spam_suspects table	Suspect storage	5
backend/src/db/migrations/005_spam_message_incidents.sql	A	Incidents + members tables	Spam Watch storage	5
backend/src/tools/{recomputeSpamSuspects,recomputeSpamMessages,reresolveSpamOrigins}.ts	A	Operational recompute CLIs	Backfill/recompute	5
backend/src/api/routes/spam.ts	A	/spam/messages/* + `/spam/suspects	observers	packet`
docs/spam-detection.md	A	Tuning/operator docs	Documentation	5
backend/src/api/utils/privateNode.ts	A	isPrivateNode / redactPrivateNode	Privacy opt-out	6
backend/src/api/routes/nodes.ts	M	Apply redaction to node + peer responses	Privacy	6

Backend — build/config

File	Type	What	Why	PR
backend/package.json	M	decoder ^0.3.0 (PR3); spam:* + test/typecheck scripts (PR5)	Deps + scripts	3,5
backend/tsconfig.json	M	noUnusedLocals/Parameters, exclude *.test.ts	Strictness + tests	5
backend/package-lock.json	M	Lockfile for decoder bump	Deps	3
.env.example	M	Spam Watch env block	Config docs	5

Frontend

File	Type	What	Why	PR
frontend/src/components/LoadingIndicator.tsx	A	Shared spinner	DRY loading UI	9
frontend/src/components/Map/NodePopupContent.tsx	M	Use LoadingIndicator	Loading UI	9
frontend/src/pages/ukmesh/{UKCompanionPage,PacketDetailPanel,UKFeedPage,UKRepeaterSearchPage}.tsx	M	Use LoadingIndicator (+ minor)	Loading UI	9
frontend/src/App.tsx	M	Poll /path-beta/multibyte-paths; loading overlays; links:false sanitize	Overlay + UX	8 (with 10)
frontend/src/components/Map/DeckGLOverlay.tsx	M	Terrain Z-elevation; clash lines	3-D paths	8
frontend/src/components/Map/MapLibreMap.tsx	M	Terrain source/zoom wiring	Terrain	8
frontend/src/components/Map/LiveOverlayController.tsx	M	Position-elevation map; clash lines; terrain flag	3-D paths	8
frontend/src/components/Map/types.ts	M	ClashPathLine, LatLonPosition	Types	8
frontend/src/components/Map/geojsonBuilders.ts	M	Client vs repeater staleness window	Map accuracy	8
frontend/src/components/Map/mapConfig.ts	M	Terrain exaggeration 2→3, DEM minzoom 5	Terrain look	8
frontend/src/store/overlayStore.ts	M	clashPathLines state	Overlay	8
frontend/src/hooks/usePacketPathOverlay.ts	M	Drive off latest message	Overlay correctness	8
frontend/src/pages/StatsPage.tsx	M	New stats cards	Stats overhaul	4
frontend/src/components/LiveStatsSection.tsx	M	Loading + stats tweaks	Stats overhaul	4
frontend/src/pages/SpamTransparencyPage.tsx	A	/spam Spam Watch page	Spam Watch UI	5/10
frontend/src/pages/spam-page.css	A	Spam page styles	Spam Watch UI	5/10
frontend/src/pages/ukmesh/UKBestPracticePage.tsx	A	/docs page	New content	11
frontend/src/pages/ukmesh/UKRegionsPage.tsx	A	Regions page (unwired)	Incomplete	11
frontend/src/pages/ukmesh/regionData.ts	A	Region dataset (only used by above)	Incomplete	11
frontend/src/pages/OwnerPortalPage.tsx	M	24 h counts + loading	Owner UI	7
frontend/src/main.tsx	M	Route table (spam/docs/regions redirects)	Routing	10
frontend/src/config/seo.ts	M	SEO + sitemap for new routes	SEO	10
frontend/src/pages/shared/SiteLayout.tsx	M	Nav/footer; drop Flasher; add Docs/Regions/Stats	Nav	10
frontend/src/pages/ukmesh/UKLayout.tsx	M	Toggle nav flags	Nav	10
frontend/src/pages/ukmesh/{UKHomePage,UKInstallPage}.tsx	M	Minor nav/content	Nav	10
frontend/src/pages/OpenSourcePage.tsx	M	Remove hero section	Cleanup	10
frontend/src/pages/{AboutPage,HealthPage,MqttPage}.tsx, pages/dev/DevMqttPage.tsx, pages/ukmesh/UKMqttPage.tsx	D	Deleted dead pages	Cleanup	10
frontend/src/hooks/{useLinks,usePathLearningModel}.ts, utils/{betaLinks,betaPathing}.ts, data/{ni-ring,uk-mainland}.ts	D	Deleted dead hooks/utils/data	Cleanup	10
frontend/src/globals.css	D	Old 3,788-line stylesheet removed	CSS consolidation	12
frontend/src/styles/globals.css	M	+927 lines (loading, spam, regions, etc.)	CSS consolidation	9/11/12
frontend/package.json	M	Remove @deck.gl/geo-layers	Dep cleanup	12
frontend/vite.config.ts	M	manualChunks.deck without geo-layers	Build	12
frontend/package-lock.json	M	Lockfile	Deps	12

Ops / workers

File	Type	What	Why	PR
docker-compose.yml	M	Restart/log/healthcheck/ports/Redis-pw; remove/disable services; cloudflared hosts	Ops hardening	13
nginx.website.conf	M	SPA no-cache header (also missing trailing newline)	Cache control	13
anubis/botPolicy.yaml	M	Allow /terrain-tiles/	CDN caching	13
viewshed-worker/Dockerfile	M	Pin GDAL by digest	Reproducible builds	13
viewshed-worker/worker.py	M	Optional radio bot; neuter radius-based elevation backfill; re-queue elevation-less	Correct elevation	13
ml-path-learner/worker.py	M	cleanup_training_artifacts retention pruning	DB bloat control	13
vacuum-compressed-chunks.log	A	Runtime log (should be ignored, not committed)	Hygiene	14

---

Potential Concerns

Breaking / behavioural

• Redis password vs default URL (PR 13): compose now starts Redis with --requirepass ${REDIS_PASSWORD}, but the backend/worker default REDIS_URL is still redis://redis:6379 (no auth). Unless REDIS_URL is set on the VM with the password embedded, every Redis client will fail auth. Verify .env carries a passworded REDIS_URL before deploying.
• ukmesh scope semantics (PR 1): every node/packet/WS query that received ukmesh now returns ukmesh + northeast. This is intended, but any external consumer expecting the old narrow scope changes silently. app.ukmesh.com host scope flipped from all → ukmesh.
• MQTT startup now awaited (PR 5): await startMqttClient() + await initSpamDetector() mean a slow spam-detector init delays broker connect. Confirm it can't hang ingestion.
• Advert-spoof can suppress node upserts (PR 5): when verdict === 'spam' the node is not upserted/emitted. A false positive makes a legitimate node vanish. The detector is uncommitted/under-exercised — review thresholds.
• Removed stats fields (PR 4): repeatersPerDay, activeRepeaters, staleRepeaters are gone from the API payload; ensure no other client/tool consumes them.

Migrations

• Migrations 003, 004, 005 are additive but must be applied in order. Existing deploys run with DATABASE_SKIP_SCHEMA_INIT and have historically been ALTERed by hand (003 even documents the manual ALTER TABLE … ADD first_seen). Confirm the live DB already matches these schemas or apply them deliberately; backfill-transport-codes.mjs should run after 004.

Incomplete / unrelated

• Regions page is dead code (PR 11): UKRegionsPage.tsx + regionData.ts (≈ 1,770 lines) are added but /regions redirects home and the component is never imported. Decide: wire it up, or hold it back until it's ready.
• ML worker contradiction (PR 13): ml-path-learner/worker.py gains real cleanup logic, yet the ml-path-learner service is commented out in compose. The code change is dead until the service is re-enabled — confirm intent.
• Spam systems coupling: the new /spam page only consumes /spam/messages/*; the advert-spoof /spam/suspects|observers|packet endpoints are served but unused by the new UI. Decide whether the advert-spoof system is still wanted or is legacy to retire.
• nginx.website.conf / several .tsx / seo.ts / main.tsx end without a trailing newline (\ No newline at end of file) — cosmetic, but worth normalising.

Missing tests

• Only the spam subsystem ships tests (backend/src/spam/*.test.ts). The Viterbi resolver rewrite (PR 2) — the highest-risk change — has only the read-only evaluate.ts harness, no committed assertions. Consider capturing a small golden-output regression test. Frontend changes (map elevation, stats) have no automated coverage.

Security / privacy

• Infra ports exposed (PR 13): Postgres 5432, Redis 6379, and Mosquitto 9001 are published on INFRA_BIND_ADDRESS, which defaults to loopback. If a LAN interface is configured, confirm it is not internet-reachable and is firewalled to the LAN/VM mesh.
• Privacy redaction (PR 6) and the spam public-JSON sanitization (PR 5) are positive, but both must be verified to cover every code path that could leak a 🚫 node or raw spam text/coords.

Dependency / version risk

• @michaelhart/meshcore-decoder ^0.2.7 → ^0.3.0 (PR 3) is a minor bump that the region-decode path depends on; regression-test decoding against stored raw_hex.
• Removing @deck.gl/geo-layers (PR 12) is only safe because its sole consumer (utils/betaPathing.ts) is deleted in PR 10 — PR 12 must not merge before PR 10.

Files touched by multiple PRs (need git add -p)

• backend/src/db/index.ts → PRs 1, 3, 5, 8.
• backend/src/mqtt/client.ts → PRs 3, 5.
• backend/src/api/routes.ts → PRs 5, 8.
• frontend/src/styles/globals.css → PRs 9, 11, 12.
• backend/package.json → PRs 3, 5. If hunk-splitting is too fiddly, collapse the sharing PRs (e.g. land 1→3→5→8 in that order on the same file).

---

Recommended Review Order

1. PR 14 — ignore runtime logs (trivial; stops vacuum-compressed-chunks.log sneaking into a later commit).
2. PR 1 — network-scope foundation (networks.ts is imported by 2 and 5; review the SQL carefully).
3. PR 6 — private-node redaction (small, self-contained, privacy win).
4. PR 3 — region/transport decoding (schema + decoder bump; unblocks stats data).
5. PR 2 — lazy/beta resolver Viterbi (highest-risk backend; review after the scope foundation it depends on).
6. PR 5 — spam detection backend (large; depends on PR 1, shares files with PR 3).
7. PR 9 — LoadingIndicator (frontend foundation for 4, 7, 8, 10).
8. PR 8 — map overlay/elevation (depends on PR 9; shares routes.ts/db.ts with 5/8 — land after 5).
9. PR 4 — stats overhaul (depends on PR 3 + PR 9).
10. PR 7 — owner portal (small; depends on PR 9).
11. PR 10 — routing/SEO/dead-page cleanup (needs the new page components from 5 + 11 to exist).
12. PR 11 — Docs + Regions pages (decide Regions' fate here; depends on PR 10).
13. PR 12 — deps/CSS consolidation (must follow PR 10 for the geo-layers removal; follow 9/11 for styles/globals.css).
14. PR 13 — deployment/ops hardening (last: highest blast radius; resolve the Redis-password and ML-worker questions before deploying).