[clorth0]

Candidate #4 from the architecture review: deepen the contact-sighting module. Identity + dedup only — destination unchanged.

Inventory (pre-refactor, origin/main a673e07)

Source	Identity shape	Dedup key	Window / TTL	Consumer today
Live ISM (rtl_433 via _emit_event, app.py:247-252)	ident = str(event.get("id", event.get("channel", ""))); info picks channel/battery_ok/temperature_C/humidity/freq/rssi	ism:{model}:{ident}	Per-key 2.0s write throttle (_CONTACT_WRITE_DT, anchored at last write; suppressed attempts don't refresh); 3.0s global contacts emit throttle	store.record_contact → sqlite contacts PK → contacts socket event. decoded emitted every event
Live ADS-B (_emit_adsb, app.py:372-386)	hex (a["hex"], skipped if falsy); label = flight.strip() or registration or hex.upper(); info = registration/country/alt/gs (non-None)	adsb:{hex}	Same 2.0s per-key throttle	Same store + adsb socket event every update
Scan rtl_433 (_phase_rtl433, scan.py:252)	{model or '?'}/{id|hex|channel or '?'} (id > hex > channel)	same string, _key on device dicts	Whole phase (rtl433_seconds, default 20s), no TTL; first event wins, count++, last_time updates	phase.findings["devices"] → phase_completed; scan_decoded live-emitted for every event. Never the store
Scan ADS-B (_phase_adsb, scan.py:295-298)	hex alone	hex	Whole phase (adsb_seconds), no TTL; latest aircraft per hex wins	phase.findings["aircraft"] → phase_completed. Never the store

No other protocols construct sighting identity: Wi-Fi/BLE bands appear only as sweep band labels (scan.py BAND_MAP), with no sighting/dedup code. Frontend keys ADS-B markers by raw hex for map rendering only (waterfall.js:1489) — unchanged, not a sighting store.

Design

One deep module, backend/contacts.py (named in docs/architecture.md; architecture-names check passes):

• ContactSighting + ism_sighting(event, lat, lon) / adsb_sighting(aircraft, registration, country) — derivation of key/kind/label/ident/info/position in one place. Enrichment (readsb r, icao_to_registration/icao_country) stays in app.py; the module owns how it joins the sighting.
• Two keys named explicitly, not forced into one: durable inventory_key (ism:{model}:{ident}, adsb:{hex}) vs ephemeral scan_key ({model}/{id|hex|channel}, hex alone for ADS-B). Key-byte parity: both are byte-identical to the old strings — the inventory key is the sqlite contacts PRIMARY KEY and crosses to the frontend, so renaming would orphan stored history. No key changed; none flagged for change.
• Dedup: SightingThrottle(window_s) (register/seen; window anchored at last passed sighting) for live writes, ScanSightings (phase-lifetime, no TTL; first-wins+count for ISM, latest-wins for ADS-B) for scan findings. app.py and scan.py both consume these; no key string is built outside the module anymore (grep ism: in app/scan: none).

Parked question — untouched

"Should Auto-Scan sightings enter the persistent inventory or stay scan-only?" is NOT answered. Scan sightings go exactly where they went today (phase findings only); scan.py has no store reference. Inventory contents are unchanged: same keys, same rows, same throttle. This PR only gives sighting identity/dedup one home.

Parity evidence

• Characterization tests pin, per source: first sighting passes, repeat within window suppressed, at/after window passes, different model/ident not suppressed, suppressed attempt does not refresh the window, scan hex-alone collision (same hex, different flight → latest wins, one entry).
• App consumer tests: _emit_event records exactly ("ism:M:7", "ism", "M", "7", {...}) once for two events while decoded is emitted twice; _emit_adsb records adsb:abc123/UAL1 with position, no-hex aircraft skipped.
• Scan phase tests (fake decoder/receiver driving the real _phase_rtl433/_phase_adsb): devices ["M/7", "Other/7"] with _count=2, findings hold the deduped lists, scan_decoded emitted for every event.

Tests + mutant

• New tests/test_contact_sighting.py: 11/11 (written failing-first — ModuleNotFoundError: backend.contacts — then green).
• Mutant: dropping the model prefix from both ISM key builders in backend/contacts.py only → 6/11, reddening dedup across both consumers (test_app_emit_event_records_through_module, test_scan_ism_dedup, test_scan_phases_consume_module_and_stay_scan_only, + key-byte tests). Restored → 11/11.

CI snapshot (local, CI-way)

All tests/test_*.py script-style suites pass (24 files incl. new one), check_html_templates.py pass, check_architecture_names.py pass, node --check on all frontend JS pass. GitHub test check: pass (run 39144786324, 22s).

Candidates #5–#6, auth, and other parked questions untouched. Do not merge without review.