TK-11372: independent re-verification of the fix by a second agent

92526783f89bd9b4532867672e382bfa30b81996 · 2026-09-11 13:15:47 -0700 · Steve Abrams

Measured before accepting: the ticket description is stale. The listener is
healthy (40/40 healthz 200, p50 1ms) and the daemon identity re-checks out
(pid 60904 = pm2 id 57 ticket-board, one God daemon, process started after
the last server.js write, so it is running the patched code).

Patch reviewed rather than trusted: candidate.patch touches one file in one
hunk, its baseline is byte-identical to the pre-patch git blob, and the
rebuilt file is byte-identical to what was committed - so nothing unrelated
was reverted. The orphan-daemon guard, the TK-10970 serialized pm2 wrapper
and the auth gate all survive.

Root cause corroborated independently by sampling the live process:
SyncProcessRunner is now 0 samples on the main thread, against 2531/3794
before. The execSync-in-timer class is gone.

Checked the false-green this patch could have introduced: serve-last-good
could have frozen /api/running at stale data behind a 200. It has not - the
cache advances every 10-15s and the serialized jlist returns in ~1.5s.

Residual reported plainly: healthz still spikes to 3.7s worst under load.
The stack names a different cause - per-request JSON.parse of the event log
in the request handler - which is the already-filed TK-11506, not this class.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Files touched

A verification/tk11372/independent-verification-vp-eng.json

Diff

commit 92526783f89bd9b4532867672e382bfa30b81996
Author: Steve Abrams <steve@designerwallcoverings.com>
Date:   Fri Sep 11 13:15:47 2026 -0700

    TK-11372: independent re-verification of the fix by a second agent
    
    Measured before accepting: the ticket description is stale. The listener is
    healthy (40/40 healthz 200, p50 1ms) and the daemon identity re-checks out
    (pid 60904 = pm2 id 57 ticket-board, one God daemon, process started after
    the last server.js write, so it is running the patched code).
    
    Patch reviewed rather than trusted: candidate.patch touches one file in one
    hunk, its baseline is byte-identical to the pre-patch git blob, and the
    rebuilt file is byte-identical to what was committed - so nothing unrelated
    was reverted. The orphan-daemon guard, the TK-10970 serialized pm2 wrapper
    and the auth gate all survive.
    
    Root cause corroborated independently by sampling the live process:
    SyncProcessRunner is now 0 samples on the main thread, against 2531/3794
    before. The execSync-in-timer class is gone.
    
    Checked the false-green this patch could have introduced: serve-last-good
    could have frozen /api/running at stale data behind a 200. It has not - the
    cache advances every 10-15s and the serialized jlist returns in ~1.5s.
    
    Residual reported plainly: healthz still spikes to 3.7s worst under load.
    The stack names a different cause - per-request JSON.parse of the event log
    in the request handler - which is the already-filed TK-11506, not this class.
    
    Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 .../tk11372/independent-verification-vp-eng.json   | 67 ++++++++++++++++++++++
 1 file changed, 67 insertions(+)

diff --git a/verification/tk11372/independent-verification-vp-eng.json b/verification/tk11372/independent-verification-vp-eng.json
new file mode 100644
index 00000000..252fe4fd
--- /dev/null
+++ b/verification/tk11372/independent-verification-vp-eng.json
@@ -0,0 +1,67 @@
+{
+  "agent": "vp-eng-11372",
+  "ticket": "TK-11372",
+  "at": "2026-09-11T20:15:00Z",
+  "purpose": "Independent re-verification of the TK-11372 fix by a fresh agent that did not apply it. Measure first, accept nothing on report.",
+  "reproduction_attempt": {
+    "result": "SYMPTOM NOT REPRODUCIBLE - ticket description is STALE",
+    "note": "Fix was applied (661b3c5) and activated 19:21Z before this pass. Verified recovery instead of the outage.",
+    "load_at_measurement": "31.18 -> 25.23 (was 70.74 when wedged)",
+    "healthz_soak": {"probes": 40, "http_200": 40, "failures": 0, "worst_seconds": 5.085611},
+    "healthz_dense": {"probes": 60, "min": 0.000657, "p50": 0.001034, "p90": 0.023419, "max": 3.729265},
+    "harness_negative_control": "dead port 9999 -> code=000, harness proven able to go RED",
+    "harness_defect_found_and_fixed": "first soak used /usr/bin/time -p and misparsed curl output, reporting 30/30 FAIL while the service was healthy; replaced with curl -w timing"
+  },
+  "daemon_identity_reconfirmed": {
+    "listener": "node pid 60904 LISTEN 127.0.0.1:9794",
+    "pm2_id": 57,
+    "pm2_name": "ticket-board",
+    "script": "/Users/macstudio3/Projects/ticket-system/server.js",
+    "god_daemons": 1,
+    "restart_time": 50,
+    "proc_start_local": "2026-09-11 12:16:17",
+    "server_js_mtime_local": "2026-09-11 11:41:59",
+    "conclusion": "process started AFTER the last server.js write, so the running code IS the patched code"
+  },
+  "patch_review": {
+    "verdict": "CLEAN - no unrelated guard reverted",
+    "files_touched": ["server.js"],
+    "hunks": 1,
+    "baseline_sha256_vs_prepatch_git_blob": "IDENTICAL (eaff7476...) - patch was built against the true current base, not a stale one",
+    "rebuilt_vs_committed": "IDENTICAL - nothing extra smuggled into the commit",
+    "disk_vs_committed": "IDENTICAL - no drift since",
+    "guards_preserved": [
+      "socket-less orphan pm2 daemon guard (existsSync rpc + lsof holder + God-daemon regex) intact",
+      "TK-10970 serialized pm2 wrapper (PM2_SERIALIZED_TS) still used, not bypassed",
+      "basic-auth gate intact (unauth /api/tickets -> 401 in 0.001s)"
+    ],
+    "nit": "exec is now an unused import in the child_process destructure (harmless)"
+  },
+  "root_cause_independently_corroborated": {
+    "method": "sample(1) of the LIVE pid 60904 for 10s under concurrent probing",
+    "SyncProcessRunner_samples": 0,
+    "before_per_ticket_evidence": "2531 of 3794 main-thread samples in SyncProcessRunner",
+    "conclusion": "the execSync-in-timer blocking class is GONE from the running process"
+  },
+  "false_green_check_specific_to_this_patch": {
+    "risk": "getRunning now always answers from cache (serve-last-good); if the async refresh never completed the board would show stale pm2 data forever with a 200 and no error",
+    "test": "polled /api/running every 8s for 100s",
+    "result": "NOT FROZEN - at advanced 20:13:08 -> 20:13:19 -> 20:13:33 -> 20:13:51 -> 20:14:01 -> 20:14:16, sessions varied 142/145/146/147",
+    "serialized_jlist_timing": "1482ms and 1787ms, far inside the 22000ms timeout",
+    "note": "at lags real time by 5-15s by design (a request returns the previous refresh); staleness is visible to the client via the at field"
+  },
+  "functional_proof": {
+    "unauth_api_tickets": "401 in 0.001014s",
+    "auth_api_tickets": "200, 15036125 bytes, 1653 tickets parsed, contains TK-11372",
+    "api_running": "200, 145 pm2 entries, live session count",
+    "tk_round_trip": "tk log wrote in 0.42s; readable back via tk show AND via the HTTP API"
+  },
+  "residual_NOT_this_ticket": {
+    "observation": "healthz still stalls to 3.7s worst of 60 probes while p50 is 1ms",
+    "named_cause_from_stack": "Builtins_ArrayMap -> Builtin_JsonParse -> JsonParser::ParseJsonObject inside the HTTP request handler (4253/8139 main-thread samples), plus Scavenge GC churn from the string allocation",
+    "meaning": "per-request re-parse and re-materialization of the 32MB event log - the uncached synchronous ticket reducer",
+    "already_filed_as": "TK-11506 (sibling TK-11499 for the uncached installedSkills scan)",
+    "honest_statement": "the TK-11372 defect class is fixed and proven gone; the service is functional but not yet fast under load, and the remaining stalls belong to a different, already-filed defect"
+  },
+  "cost": "$0 (local)"
+}

← 1ccfb2d0 board: add ⤓/⤒ icon toggle to flip nav bar above/below toolb · back to Ticket System · auto-data-snapshot: 2026-09-11T13:23:02 (1 data files) — dat cf1f4bfb →