Simulation config: frac_adsb_outage knob + retina-simulation bump

[jehanazad]

Summary

Plumbs the simulator's new transponder-outage knob through the simulation config API, and bumps libs/retina-simulation to pick it up (offworldlabs/retina-simulation#13).

The simulator previously never stopped ADS-B: every transponder aircraft was tagged in every node frame and pushed every second for its whole life. That left the known-track hold (a node track claimed to a hex staying claimed once ADS-B stops) unverifiable against simulated traffic. With frac_adsb_outage an ADS-B aircraft goes silent mid-flight and comes back — has_adsb stays true, it just stops broadcasting.

Default 0.0, so nothing changes for a deployment that never sets it.

Changes

• core/state.py: simulation_config / _SIMULATION_CONFIG_DEFAULTS gain frac_adsb_outage: 0.0. Deliberately not added to _SIM_FRAC_FALLBACKS — it is not one of the spawn-type fractions and is not env-seeded with them.
• routes/test.py: PUT /api/simulation/config accepts frac_adsb_outage (0.0–1.0), validated by the existing frac_* range rule but excluded from the sum constraint — it is a fraction OF the ADS-B population, orthogonal to the type roll, so a dark-heavy scene at the 1.0 ceiling can still take its transponder aircraft silent. GET reports ground_truth_counts["adsb_silent"], counted alongside (not instead of) the type buckets, since a silent aircraft is still a commercial aircraft.
• routes/sim_ingest.py: the ground-truth push whitelists adsb_silent into ground_truth_meta, defaulting False for older fleet payloads. has_adsb is untouched, so the dark count does not move when a transponder goes quiet.
• services/state_snapshot.py: _SIM_CONFIG_RESTORE_KEYS gains the key, so an operator's setting survives a rebuild like the other fractions.
• Submodule pin bump only for libs/retina-simulation.

Test coverage

backend/tests/test_sim_ingest.py extended alongside the existing frac_dark / scene-key cases: default is off, PUT accepted and echoed, out-of-range rejected both ends, the knob accepted at 1.0 while the type fractions already sum to 1.0 (the sum-exclusion guard), counts reporting a silent aircraft without moving the dark count, and the ground-truth meta flag with its legacy-payload default. Existing exact-dict count assertion updated for the new key.

Review notes

• The knob only bites when the fleet polls the config; the orchestrator applies it alongside frac_dark and also re-rolls aircraft already in the air, so raising it is visible within a poll interval rather than a fleet turnover.
• The lib PR carries the world/orchestrator side and its own tests.

🤖 Generated with Claude Code

[claude]

Claude finished @jehanazad's task in 1m 20s —— View job

---

Review: PR #332 — frac_adsb_outage knob + retina-simulation bump (updated after synchronize)

• Gather context (PR description, new commit)
• Diff new commit against previous review
• Review submodule pin bump commit
• Re-check prior findings for whether they were addressed
• Post final review

This PR now has three commits. The first two (4afb1bf — the frac_adsb_outage knob/validation/snapshot-restore/submodule bump, and 03bfc82 — exposing adsb_silent on the v1 ground-truth endpoint) are unchanged from the last review and remain solid.

The new commit (67f1700, "Pin retina-simulation to main (transponder outages merged)") only re-points the libs/retina-simulation submodule:

-Subproject commit f03cae0a3fad80c6868125dde1c1ca5525508730
+Subproject commit f06d534709c45dac18d1dd24df99e27ec95138a1

This moves the pin off the feature-branch commit (f03cae0, set in the first commit of this PR) onto retina-simulation's main now that offworldlabs/retina-simulation#13 has merged. That's the right thing to do before this PR itself merges — it avoids retina-server's main ever depending on an unmerged branch of the submodule. No backend code changed in this commit, so nothing here alters the earlier analysis.

Note: I can't inspect the submodule's actual diff between those two commits in this sandbox — libs/retina-simulation isn't checked out locally and I don't have network access to fetch it here. Worth a quick sanity check that f06d534 is genuinely retina-simulation's main tip (not a stray commit) before merging.

Carried forward from the last review (still unaddressed, both minor/non-blocking):

1. No test coverage for adsb_silent on the v1 endpoint (backend/tests/test_output_routes.py). backend/routes/output.py:143 gained "adsb_silent": meta.get("adsb_silent", False) but no test asserts it, present or defaulted. Mirrors what test_sim_ingest.py already does for the ingest side.
   Fix this →
2. docs/simulation.md config table still missing frac_adsb_outage (around lines 69-75, unchanged). frac_anomalous, frac_drone, frac_dark are documented; the new knob and the adsb_silent ground-truth field aren't.
   Fix this →
3. Frontend Physics tab still not wired to the new knob (frontend/src/components/PhysicsSettings.tsx) — flagged again only for completeness; likely deferred deliberately.

I wasn't able to execute pytest, the pre-commit gate, or fetch the submodule in this sandboxed environment (no approved network/bash access here), so please confirm CI is green.