Gate overlap zones on node world

[jehanazad]

Pairs with offworldlabs/retina-analytics#25 — this PR is the pin bump plus the
status field. Merge the library PR first.

What

• libs/retina-analytics pinned to 4439c2c (branch feat/world-gate-overlap-zones),
  which stops register_node/rebuild_zones_for building an overlap zone
  between two nodes in known, different worlds, and drops one that already
  exists.
• /api/radar/association/status gains assoc_world_skipped_pairs next to
  overlap_zones.

Why

PR #267 gated ADS-B claiming and seeding on node_world_provider, but
bottom-up pairing was untouched. This fleet is where that bites: 50 synthetic
nodes and 8 receivers (ret*, radar3*-retnode) share one footprint over
Greenville, so the associator holds a zone for every sim/real pair, and any
pairing drawn from one of those grids can only ever match a simulated tracklet
against a real echo. A read-only audit found real node ids (retf1144fcd,
ret7dd2cb0d) in 48 of 156 synthetic dark solver records over a 7-minute
window.

Nothing new is wired: state.node_world is already the single authority for
the question and is already injected as node_world_provider.

Numbers, measured live on retina-test (read-only /api/radar/association/status)

	pairs
registered nodes	58 (50 synth-*, 8 real)
overlap zones today	1653 — every pair, C(58,2)
cross-world (sim × real) pairs	400
...with a non-empty grid, i.e. a live adjacency edge	39
zones after the gate	1253 (1225 sim/sim, 28 real/real)

So the gate removes 24% of the zone table and 39 real neighbour edges, all of
them edges no true pairing can live on.

assoc_world_skipped_pairs is what makes that visible: it counts pairs
considered and refused, so it climbs with re-registrations, and zero means the
fleet is single-world (or untagged) rather than the gate being off.

Verified

• libs/retina-analytics suite: 420 passed (9 new in tests/test_world_overlap_gate.py).
• tests/test_analytics_routes.py tests/test_adsb_seed_backend.py: 54 passed
  — new: the status field reflects the counter; a synth-* node and a hardware
  node registered on overlapping coverage get no zone and no adjacency; two
  synthetic nodes on the same coverage still pair.
• Full backend suite: 2777 passed, 2 skipped in 165.25s (-n 2 -m "not external").
• pre-commit run --all-files clean in both the worktree and the library.
• No deploy, no container restart.

🤖 Generated with Claude Code

[claude]

Claude finished @jehanazad's task in 1m 47s —— View job

---

Code Review

Todo list

• Gather context (PR diff, description, CLAUDE.md)
• Read changed files (backend/routes/analytics.py, test files, library pin)
• Check the paired library PR / pinned commit for the actual gating logic
• Review for correctness, security, style
• Post final review

Summary

This backend-side PR is exactly what it says it is: a submodule pin bump plus one status field. The diff is small (4 files, +50/-1) and low-risk.

Note on scope: libs/retina-analytics is a git submodule and isn't checked out in this environment (empty dir, network fetch unavailable to me here), so I could not inspect the actual world-gating logic in register_node/rebuild_zones_for — that lives in the paired library PR (offworldlabs/retina-analytics#25) and should be reviewed there. This review covers only what's in retina-server: the status field and its tests.

Findings

Nothing blocking. A few observations:

• backend/routes/analytics.py:135 — getattr(_a, "assoc_world_skipped_pairs", 0) is defensive against an associator instance that predates the pin, consistent with every other counter in this handler (track_pairs_gated, claim_rounds, etc. all use the same getattr(..., 0) pattern). Good consistency, no issue.
• The inline comment on assoc_world_skipped_pairs (routes/analytics.py:128-134) is long, but matches the house style already used for every other field in this same function (each counter gets a paragraph explaining what it means and why zero isn't necessarily "off"). Fits the existing pattern.
• Tests correctly rely on the existing is_synthetic_node prefix rule (synth-, e2e-, realnode-, test-) via state.node_world, so synth-GVL-9001/synth-GVL-9002 resolve to "sim" and hw-9001 resolves to "real" — the fixtures test what they claim to.
• State isolation between the new tests is handled by the existing autouse _reset_for_tests fixture in conftest.py, plus each test's own finally: state._reset_for_tests() — no leakage risk between tests.
• No contracts/nodes-v1.openapi.yaml update needed — /api/radar/association/status isn't under /v1/nodes.
• The counter naming (assoc_world_skipped_pairs) and its cumulative-since-boot semantics are clearly documented in both the route comment and the PR description, so the "zero doesn't mean the gate is off" footgun is well covered.

Overall: clean, well-tested, well-scoped change. The real correctness question (does the gate actually skip the right pairs and drop existing zones correctly) lives in the library PR, which the description says already has 9 new tests and a full green suite.