spec(queen): finalize.t27 -- a cancelled or failed run keeps its observations, its first error and what it still owes (Closes #10424)

.trinity/seals/queen_QueenFinalize.json

@@ -0,0 +1,53 @@
+{
+  "built_by": "t27c-bootstrap@0.5.2+9448ab4be",
+  "closure": [
+    {
+      "spec": "specs/queen/control.t27",
+      "spec_hash": "sha256:2c9c2698bca4bd146d21cf8ee002935e8c80cb88a0fc29cc5bb75e90eb496e58"
+    },
+    {
+      "spec": "specs/queen/dispatch.t27",
+      "spec_hash": "sha256:e71997bc00e5e3c6fffc1d92d0ea1e73c09326215533f7bdec6a27ed9fffd7fe"
+    },
+    {
+      "spec": "specs/queen/dispatch_exit.t27",
+      "spec_hash": "sha256:2a1b5a1bb24298ee05319cd9458d0a31c4e596a4d87fe9e360fdc93dc9a56037"
+    },
+    {
+      "spec": "specs/queen/drain.t27",
+      "spec_hash": "sha256:ae64ef302f98301a0594298d43848603ad55177e387f72d7edcf0b863f5a081a"
+    },
+    {
+      "spec": "specs/queen/turn_stop.t27",
+      "spec_hash": "sha256:092bab0e5725ec0fbc8acded0c268a4df145f3b66fec8833865126eb541d6ba2"
+    },
+    {
+      "spec": "specs/queen/worktree_keep.t27",
+      "spec_hash": "sha256:2684225bc4e87d79b5d5846324d6171620f681c44ed9ccf8cf3957abdbbeae89"
+    }
+  ],
+  "config": "gen=c,rust,verilog,zig;tests=zig",
+  "gen_hash_c": "sha256:88877b12a394b717089e75cdff27f3265f4797fe41375643707ffd7f381881db",
+  "gen_hash_rust": "sha256:1c72d330b327f1ad4b21447ef3252e5f8bcdcad4666017a608b437c16cb4ba0e",
+  "gen_hash_verilog": "sha256:8d38d8d6ae00c88b15800576a2a0c7ca3965ce698a45d7ef37a9a78a238597d9",
+  "gen_hash_zig": "sha256:65b455c89f698f4e5f5c1307c27958d0d7e259ff6124d04d25dea4d7ea5156ad",
+  "module": "QueenFinalize",
+  "ring": 12,
+  "seal_schema": 2,
+  "sealed_at": "2026-10-18T11:26:18Z",
+  "sealed_by": "t27c-bootstrap@0.5.2",
+  "spec_hash": "sha256:11ee7b14394465bd9a760876293edb38745e6fee878bd99a3e7aa62d1e1ba194",
+  "spec_path": "specs/queen/finalize.t27",
+  "tests": {
+    "failed": 0,
+    "failing": [],
+    "forced": false,
+    "passed": 26,
+    "total": 26,
+    "vacuous": 0
+  },
+  "toolchain": {
+    "test_runner": "sha256:1982ae948e24d00c5d221ccec12dfce1eb01204b53e37b0f857f93b16d64629d",
+    "zig": "0.16.0"
+  }
+}

docs/now/2026-10-18-queen-finalize.md

@@ -0,0 +1,14 @@
+# NOW -- finalize.t27: how a bee run or a claw check run ends, what it keeps and what it owes (2026-10-18)
+
+## specs/queen/finalize.t27 (Closes #10424)
+
+- Idea from REA (morluto/rea, rea-agents 6.6 and 6.7, MIT); no code taken.
+- Measured 2026-10-06, the bee-runner disk jam: both 49 GB runner volumes had about 6.8 GB free, bee merges were 0 for about eight hours.
+  - Leaked finished worktrees (about 294 MB each; 80 and 98 removed by hand) were listed nowhere as owed.
+- Rule 1: `observations_kept()` and `entry_after_finalize()` keep every completed observation for every outcome; `report_kind()` says partial, never empty.
+- Rule 2: `takes_primary()` keeps the first failure or cancel; a cleanup error is never primary. `cancel_reason_after()` keeps the exact reason that dispatch_exit's DX_SHUTDOWN folds away.
+- Rule 3: `entry_after_check()` keeps an entry pending, with its owner, until a read says the resource is gone; `reap_action()` releases a worktree only on worktree_keep's `reap_verdict` proof. This list is what ai.t27.runner-reap should read.
+- Rule 4: `shutdown_step()` waits for in-flight probes up to PROBE_GRACE_MS (3000 ms) on top of drain's `exit_step()`; a cut probe answers unknown.
+- Rule 5: `final_status()` keeps unknown unknown; `check_conclusion()` posts neutral for cancelled or unknown, never failure.
+- Railway lab, `t27c test-report`: 26 tests, 0 FAIL, 0 vacuous, 6 invariants; `spec-status` IMPLEMENTED; seal verify all MATCH.
+- Not here: wiring into gHashTag/trios and the reaper itself.