Gladex Agent Logs

- `1` - one or more record gaps, each named (sweep sha, subject, the content paths it carried); see the `gaps[]` array
- `2` - invalid arguments: unknown flag, missing value, `--format` outside json|human — argparse validates before any git I/O, so human-mode stdout stays empty and json mode emits the object with `exit_code: 2` (the `queue-source-check` `[0.4.62]` contract)
- `3` - cannot verify: not a git work tree, history unreadable, `--since` does not resolve to a commit, or no commit ever touched `agent-logs/PROGRESS.md` — never a pass, never reported as "no gap"
- Precedence: 2 > 3 > 1 > 0.

**JSON keys** (exactly): `tool`, `timestamp`, `repo_root`, `window` (`base`,
`base_source` ∈ last_progress_touch|since|all, `commits`), `rules` (`loop_sweeps`,
`identity_sweeps`, `deliberate`, `content_hits`), `gaps` (list of `{sha, subject, paths}`),
`violations`, `error`, `detail`, `ok`, `exit_code`. `detail` is sanitised (no `"`, no `\`) —
the `system-status` row embeds it without re-escaping, the same guarantee
`queue-source-check` gives.

**Wired into**: the `record-gap` row of `tools/system-status` (child contract identical to
the `queue-source` row: `0` → ok, `1` → error + `ERRORS++`, `3`/unrunnable → warning),
hooks `GLADEX_RECORD_GAP_BIN` / `GLADEX_RECORD_GAP_TIMEOUT`, both in that tool's `--help`.

**Status**: Built ✅ 2026-10-31 (queue: run 1023's record-gap item). Live reading at
commit time: window since `be3ab8f` (**0** commits) → **0** gaps; the hazard fixture
(one sweep carrying `CHANGELOG.md` + `tools/REGISTRY.md`, one deliberate builder commit,
one identity shift) is detected and named while the other two stay green. Guarded by
`tests/test_record_gap.sh` (**43 assertions / 0 failed**, 3 mutations, measured
2026-10-31).

## test_record_gap.sh

**Purpose**: hermetic guard for `tools/record-gap`. Pins: the arg contract (unknown flag /
`--format xml` / bare `--format` → exit 2, empty human stdout, parseable json usage object
with `exit_code: 2`); the cannot-verify class (non-git dir, a repo with no committed
`agent-logs/PROGRESS.md`, an unresolvable `--since` → exit 3, never a pass); the hazard
itself (a loop sweep carrying `CHANGELOG.md` + `tools/REGISTRY.md` with no recorded run →
exit 1 with the sha and paths named, while the deliberate builder commit and the identity
shift sweep are NOT dragged in); the window semantics (one recorded run after the gap
closes the default window → exit 0; `--since <root>` and `--all` still show it; a sweep
sitting AT the `--since` base is excluded — the range is exclusive like every git range);
3 precondition-asserted mutations (M1 the content-path list emptied — the queue's own
required "mutant that ignores content paths" —, M2 the sweep subject regex defeated, M3
the subject classifier dropped so the deliberate commit gets named too); the read-only
source decision (static: the tool never spells `git add`/`git commit`/`git checkout`/
`git reset`); and the live box race-safely (exit 0 or 1 allowed, json `exit_code`/`ok`
agree with its own exit code — a subset-style agreement, so a genuine gap landing mid-run
cannot redden it).

**Location**: `/data/repo/tests/test_record_gap.sh`

**Usage**:
```
bash tests/test_record_gap.sh
```

**Exit codes**:
- `0` - all assertions pass
- `1` - at least one failure

**Test hooks (env)**: `GLADEX_RECORD_GAP_TOOL` — the tool copy under test (default: the
sibling `tools/record-gap`); mutants are sed-mutated COPIES in the sandbox, never the real
tool.

**Hermeticity**: everything but the live-box section runs against fixture repos in one
`mktemp -d` sandbox under `${TMPDIR:-/tmp}/opencode/`, trap-removed; the real repo is only
ever READ (section H), and never written.

**Status**: Built ✅ 2026-10-31 — **43 assertions, 0 failed, rc 0** (measured at commit
time). It is the 127th `tests/test_*` file, so it owns the `- Live:` **126 → 127** refresh
in the same commit as this section.
REGEOF
echo "appended, new tail:" && tail -3 tools/REGISTRY.md
appended, new tail:
**Status**: Built ✅ 2026-10-31 — **43 assertions, 0 failed, rc 0** (measured at commit
time). It is the 127th `tests/test_*` file, so it owns the `- Live:` **126 → 127** refresh
in the same commit as this section.
> build · mimo-v2.6-flash-free
Now the `- Live:` refresh — first verify the count is 127, then edit the claim position with python (the line is enormous):
$ cd /data/repo && ./tools/regression-run --list 2>/dev/null | tail -1 && ls tests/test_* | wc -l && python3 - <<'PYEOF'
import io
p = "/data/repo/tools/REGISTRY.md"
s = open(p, encoding="utf-8").read()
old_claim = "- Live (measured 2026-10-28): `./tools/regression-run` → **126 suites** discovered *(Refreshed **125 → 126**"
new_claim = ("- Live (measured 2026-10-31): `./tools/regression-run` → **127 suites** discovered "
  "*(Refreshed **126 → 127** on 2026-10-31 by the run that added `tests/test_record_gap.sh`, "
  "the 127th suite — the hermetic guard for `tools/record-gap`, run 1024's queue step (the "
  "unrecorded-run hazard class run 1023 repaired, now closed by a check instead of prose). "
  "Measured at the moment of the write: `./tools/regression-run --list` → **127 suite(s) "
  "discovered**, `ls tests/test_*` → **127**, `git ls-tree -r --name-only HEAD tests/` → "
  "**126** because this suite is untracked until its commit lands (re-read after it lands, "
  "as every note above is); `tests/test_registry_coverage.sh` section F is the reader this "
  "line is refreshed for, and the run that lands the 128th suite owns the next refresh.)* "
  "*(Refreshed **125 → 126**")
assert s.count(old_claim) == 1, "claim needle not unique: %d" % s.count(old_claim)
open(p, "w", encoding="utf-8").write(s.replace(old_claim, new_claim))
print("claim refreshed 126 -> 127")
PYEOF
grep -c '^- Live (measured 2026-10-31)' tools/REGISTRY.md
result: 127 suite(s) discovered, none run (exit 0)
127
claim refreshed 126 -> 127
1
> build · mimo-v2.6-flash-free
Exactly one `- Live:` figure, now 127. Now the CHANGELOG entry [0.4.383]:
$ cd /data/repo && cat >> CHANGELOG.md <<'CLEOF'

## [0.4.383] - 2026-10-31 — `record-gap` closes the unrecorded-run hazard class: the loop's post-run sweeps that carried CONTENT work no run recorded are now a dashboard red the moment they land (`tools/record-gap` + the `record-gap` row of `tools/system-status`, window = since the last `agent-logs/PROGRESS.md` touch; new suite **43/0**, `- Live:` 126 → **127**)

### Fixed
- **The class, measured twice, now has a check instead of prose.** The loop's post-run safety net (`add -A && commit` under the generic subject `run <loopid>: <ts> model=<provider/model>`) commits whatever a session left behind. When that sweep carried a CONTENT path (`CHANGELOG.md`, `STRUCTURE.md`, `tools/`, `tests/`, `app/`, `team/`) no run's `agent-logs/PROGRESS.md` entry ever recorded, the work landed on main under a subject no reader can connect to a run and the authoritative queue never said what it was: `fec13e8` (run 2689's jsonl-rotate content — run 1023 found the queue pointing at a step that was DONE but unrecorded and had to re-verify the whole schedule from prose) and `3829df3` (run 2440's loop-outcomes pair rule, patched by run 1016). A full-history census found more instances of the same shape behind them, every one costing a later run the job of re-deriving from prose what a commit already knew.
- **`tools/record-gap`** (new, python3, read-only by construction — `git rev-parse` + `git log` only, pinned statically by its suite): a sweep-shaped commit touching a content path is a **record gap**, reported with its sha, subject and the content paths it carried; identity-shift sweeps (`identity <id> shift …`) touching only `agent-logs/identity-*.md` stay green; deliberate commits (a builder run's own `run <NNNN>: …` — which never carries the sweep's timestamp+`model=` tail, measured across the whole history) are never gaps. The **window is the point**: default `<last commit that touched agent-logs/PROGRESS.md>..HEAD`, because an older gap was already in front of a recorded run (repaired or accepted on the record — a standing red nobody acts on would be the wrong shape), the live hazard sits at the TOP of history, and one recorded run after a gap closes its window — the repair clears the red, not a suppression list. `--since <ref>` (exclusive, like every git range) and `--all` serve forensics. Exit 0 no gap / 1 gaps named / 2 invalid arguments (before any git I/O) / 3 cannot verify (non-repo, unresolvable `--since`, no committed `agent-logs/PROGRESS.md` — never a pass).
- **Wired where it is seen**: the `record-gap` row of `tools/system-status`, child contract identical to the `queue-source` row (`0` → ok, `1` → error + `ERRORS++` with the child's own sanitised detail — the row never restates a cause list, `[0.4.106]`'s rule —, `3`/unrunnable → warning, never ok and never an error on a tree that was not checked); hooks `GLADEX_RECORD_GAP_BIN`/`GLADEX_RECORD_GAP_TIMEOUT`, both in `--help`; the `## system-status` registry section carries the row bullet and the hook bullet. Measured live on all three row paths before landing: the real child → `ok`, a no-output stub → `warning cannot verify`, an `exit_code: 1` stub → `error`, errors 3 → 4.

### Tests
- **`tests/test_record_gap.sh`** (new, the 127th suite — owns the `- Live:` **126 → 127** refresh, re-measured `regression-run --list` → 127 at the write): **43 assertions / 0 failed**, one `mktemp -d` sandbox, fixture repos only — the arg contract (9), the cannot-verify class (4), the hazard fixture (8: the sweep named with both content paths, the deliberate builder commit and the identity shift NOT named, json classification 1/1/1 with `base_source`/`commits`, embeddable `detail`), the window closing on repair (2), forensics `--since`/`--all`/exclusivity (5), 3 precondition-asserted mutations (M1 the content-path list emptied — the queue's own required mutant — makes the gap INVISIBLE while the real tool still names it; M2 the sweep regex defeated likewise; M3 the classifier dropped makes the deliberate commit get named, proving the "deliberate is never a gap" branch carries the verdict), the static read-only source rule (6), and the live box race-safely (2: exit 0 or 1, json agrees with its own exit code).
- Gates after the edits: `bash tests/test_record_gap.sh` **43/0**, `bash tests/test_system_status_failed_units.sh` re-run against the new row, `bash tests/test_registry_coverage.sh`, `bash tests/test_repo_lint.sh`, `bash tests/test_queue_source.sh` (R8 transitional until this entry's own commit lands it), `tests/test_leak_figure_readers.sh` + `tests/test_registry_figure_readers.sh` (the `- Live:` refresh is the only figure this entry moves). Re-read after the commit rather than carried, per this file's own rule.

### Notes
- **No promote, and none exists to take**: `tools/record-gap`, `tools/system-status`, `tools/REGISTRY.md`, `tests/test_record_gap.sh` and this file are repo scope — dev and prod stay **0.4.28** (the investor's 2026-10-05 correction stands; `0.4.29` quoted nowhere). STEP 0 paid first and owed nothing: **0 unread** `investor_to_agent` in both DBs (table `messages`, column `direction`), `./tools/inbox-status` rc 0 "nothing owed", `INBOX.md` 80/80/0, last reply dev 150 / prod 115 — no `agent_to_investor` row written. `./tools/version-check` → dev=0.4.28 prod=0.4.28 (match); `./tools/budget-show` → 2026-10 spent **0.00** / remaining 5.00; `./tools/system-status --format json` → `overall=error, errors=3` — the standing rows: 7 `identity-run@*` failed units (A12/A30), the DNS apex incident (awaiting the investor's answer to dev 150 / prod 115 — the zone untouched), `ip-drift public_ip_unavailable`; `loop-outcomes` green. No `systemctl` change, no DNS/zone write, no mail, no messages-DB write, no `hiring/queue/` write (2 proposals still `status=proposed`), no API key, never `git stash`, staging by explicit pathspec. Spend **0.00**, free `*-free` models only.

### Queue
Authoritative next-candidate queue lives in `agent-logs/PROGRESS.md`, in the newest run entry's `Queue — next small step` bullet — it is not duplicated here, and has not been since `[0.4.67]`. To read the live queue, open `PROGRESS.md`; to know whether that is still true, run `tools/queue-source-check`.
CLEOF
echo "CHANGELOG appended:" && grep -c '^## \[0.4.383\]' CHANGELOG.md && tail -1 CHANGELOG.md | head -c 80
CHANGELOG appended:
1
Authoritative next-candidate queue lives in `agent-logs/PROGRESS.md`, in the new
> build · mimo-v2.6-flash-free
CHANGELOG landed. Now the gate suites — fast ones first:
$ cd /data/repo && echo "== test_record_gap ==" && timeout 110 bash tests/test_record_gap.sh 2>&1 | tail -1 && echo "== repo_lint ==" && timeout 110 bash tests/test_repo_lint.sh 2>&1 | tail -1 && echo "== queue_source ==" && timeout 110 bash tests/test_queue_source.sh 2>&1 | tail -1 && echo "== leak_figure_readers ==" && timeout 110 bash tests/test_leak_figure_readers.sh 2>&1 | tail -1 && echo "== registry_figure_readers ==" && timeout 110 bash tests/test_registry_figure_readers.sh 2>&1 | tail -1 && echo "== cross_reader_agreement ==" && timeout 110 bash tests/test_cross_reader_agreement.sh 2>&1 | tail -1
== test_record_gap ==
result: 43 passed, 0 failed, exit 0
== repo_lint ==
===============================
== queue_source ==
===============================
== leak_figure_readers ==
===============================
== registry_figure_readers ==
PASS: V15a immich-roundtrip: figure 160 and suite tests/test_immich_roundtrip.sh both read off the section
== cross_reader_agreement ==
=== Results: 64 passed, 0 failed ===
> build · mimo-v2.6-flash-free
Three suites printed a banner as their last line — grabbing their real summaries:
$ cd /data/repo && timeout 110 bash tests/test_repo_lint.sh > /tmp/opencode/rl-1024.log 2>&1; echo "repo_lint rc=$?"; grep -E 'result|passed|failed|ERROR|FAIL' /tmp/opencode/rl-1024.log | tail -3; timeout 110 bash tests/test_queue_source.sh > /tmp/opencode/qs-1024.log 2>&1; echo "queue_source rc=$?"; tail -3 /tmp/opencode/qs-1024.log; timeout 110 bash tests/test_leak_figure_readers.sh > /tmp/opencode/lf-1024.log 2>&1; echo "leak_figures rc=$?"; tail -3 /tmp/opencode/lf-1024.log
repo_lint rc=0
PASS: Y5: the FAIL line names the artefact and the elf rule
PASS: Y6: the FAIL line names the allowlist so the fix is obvious
repo-lint suite: 487 passed, 0 failed
queue_source rc=0
===============================
queue-source-check suite: 278 passed, 0 failed
===============================
leak_figures rc=0
===============================
leak-figure-readers suite: 41 passed, 0 failed
===============================
> build · mimo-v2.6-flash-free
All green — but queue-source R8 passing before my PROGRESS entry surprises me. Verifying why:
$ cd /data/repo && grep -c '0\.4\.383' agent-logs/PROGRESS.md; grep -o '.\{60\}0\.4\.383.\{40\}' agent-logs/PROGRESS.md | head -3
0
> build · mimo-v2.6-flash-free
`0.4.383` appears 0 times in PROGRESS.md, yet the suite passed. Checking the tool's actual verdict directly:
$ cd /data/repo && ./tools/queue-source-check; echo "rc=$?"; ./tools/queue-source-check --format json | python3 -c "import json,sys; d=json.load(sys.stdin); print('newest:', d['rules']['newest_version'], 'named:', d['rules']['newest_version_in_progress'], 'violations:', d['violations'])"
queue-source-check: FAIL - 1 violation(s): [0.4.383] is not named in agent-logs/PROGRESS.md - the newest CHANGELOG entry has no record in the authoritative file
  - [0.4.383] is not named in agent-logs/PROGRESS.md - the newest CHANGELOG entry has no record in the authoritative file
rc=1
newest: 0.4.383 named: False violations: ['[0.4.383] is not named in agent-logs/PROGRESS.md - the newest CHANGELOG entry has no record in the authoritative file']
> build · mimo-v2.6-flash-free
$ cd /data/repo && grep -n 'live\|REPO\b\|--repo' tests/test_queue_source.sh | head -20 && grep -c 'LIVE' tests/test_queue_source.sh
10:# Never touches the live /data/repo: every scenario builds its own repo under a
11:# throwaway sandbox and passes --repo explicitly. The ONLY cases that resolve a
12:# repo by themselves copy the tool into the sandbox first, so the live tree is
56:  OUT=$("$TOOL" --repo "$repo" "$@" 2>"$SB/.err")
77:#   hist1 + hist2 + newest-items is what the tool freezes (the live value 111).
96:      echo "Authoritative next-candidate queue lives in agent-logs/PROGRESS.md."
98:      echo "Authoritative next-candidate queue lives in this section, below."
118:# a pointer with zero items, PROGRESS carries the live list.
124:# explicitly (the same --repo discipline the rest of the suite uses): the
302:assert_contains "D6 the violation says where the queue lives" \
303:  "the queue lives in agent-logs/PROGRESS.md, not here" "$OUT"
364:OUT=$("$TOOL" --repo "$SB/repo" --format json --nonsense 2>"$SB/.err"); RC=$?
385:OUT=$("$TOOL" --repo "$SB/repo" --nonsense 2>"$SB/.err"); RC=$?
391:OUT=$("$TOOL" --repo "$SB/repo" --format json --help 2>"$SB/.err"); RC=$?
396:OUT=$("$TOOL" --repo "$SB/repo" --format yaml 2>"$SB/.err"); RC=$?
400:OUT=$("$TOOL" --repo "$SB/repo" --format json 2>/dev/null); RC=$?
448:OUT=$("$TOOL" --repo "$SB/link" 2>&1); RC=$?
449:assert_rc "I6 an explicit --repo with no CHANGELOG exits 3" 3 "$RC"
485:# the live tree when the fix was written, and the number moved whenever an entry
586:                "Authoritative next-candidate queue lives in `agent-logs/PROGRESS.md`.",
1
> build · mimo-v2.6-flash-free
Hermetic — the suite never reads the live tree, and the live tool correctly reports the predicted transitional R8 (cleared by my PROGRESS entry naming `[0.4.383]`). Now the two slow gates:
$ cd /data/repo && timeout 300 bash tests/test_system_status_failed_units.sh > /tmp/opencode/su-1024.log 2>&1; echo "system_status_failed_units rc=$?"; tail -2 /tmp/opencode/su-1024.log; grep -c '^FAIL' /tmp/opencode/su-1024.log
system_status_failed_units rc=0
failed: 0
summary: 519 assertions, 519 passed, 0 failed
0
$ cd /data/repo && timeout 300 bash tests/test_registry_coverage.sh > /tmp/opencode/rc-1024.log 2>&1; echo "registry_coverage rc=$?"; tail -2 /tmp/opencode/rc-1024.log; grep -c '^FAIL' /tmp/opencode/rc-1024.log
registry_coverage rc=124
PASS: C22a plant: healthcheck's Exit-codes block now names --timeout
PASS: C22b plant: the TOOL is unchanged - it still answers 3, not 2
0
> build · mimo-v2.6-flash-free
519/0 on the system-status suite. Registry-coverage timed out at 300s mid-run with 0 FAILs — rerunning with a longer budget:

Select another run log from the list above. Only files matching run-YYYYMMDD-HHMMSS-N.log are readable.

[Sat Oct 31 02:30:31 2026] 127.0.0.1:42760 Accepted
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42760 Closing
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42776 Accepted
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42776 Closing
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42790 Accepted
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42790 Closing
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42794 Accepted
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42794 Closing
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42796 Accepted
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42796 Closing
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42800 Accepted
[Sat Oct 31 02:30:31 2026] 127.0.0.1:42800 Closing
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54930 Accepted
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54930 Closing
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54934 Accepted
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54934 Closing
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54938 Accepted
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54938 Closing
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54946 Accepted
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54946 Closing
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54954 Accepted
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54954 Closing
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54960 Accepted
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54960 Closing
[Sat Oct 31 02:31:45 2026] 127.0.0.1:54966 Accepted
[Sat Oct 31 02:31:46 2026] 127.0.0.1:54966 Closing
[Sat Oct 31 02:31:46 2026] 127.0.0.1:54976 Accepted
[Sat Oct 31 02:31:46 2026] 127.0.0.1:54976 Closing
[Sat Oct 31 02:31:46 2026] 127.0.0.1:54992 Accepted
[Sat Oct 31 02:31:46 2026] 127.0.0.1:54992 Closing
[Sat Oct 31 02:31:46 2026] 127.0.0.1:54994 Accepted
[Sat Oct 31 02:31:46 2026] 127.0.0.1:54994 Closing
[Sat Oct 31 02:32:43 2026] 127.0.0.1:51466 Accepted
[Sat Oct 31 02:32:43 2026] 127.0.0.1:51466 Closing
[Sat Oct 31 02:33:16 2026] 127.0.0.1:43204 Accepted
[Sat Oct 31 02:33:16 2026] 127.0.0.1:43204 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct 31 02:33:16 2026] 127.0.0.1:43204 Closing
[Sat Oct 31 02:33:17 2026] 127.0.0.1:43206 Accepted
[Sat Oct 31 02:33:17 2026] 127.0.0.1:43206 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct 31 02:33:17 2026] 127.0.0.1:43206 Closing
[Sat Oct 31 02:34:25 2026] 127.0.0.1:44322 Accepted
[Sat Oct 31 02:34:25 2026] 127.0.0.1:44322 Closing
[Sat Oct 31 02:38:12 2026] 127.0.0.1:40092 Accepted
[Sat Oct 31 02:38:12 2026] 127.0.0.1:40092 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct 31 02:38:12 2026] 127.0.0.1:40092 Closing
[Sat Oct 31 02:38:14 2026] 127.0.0.1:40100 Accepted
[Sat Oct 31 02:38:14 2026] 127.0.0.1:40100 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct 31 02:38:14 2026] 127.0.0.1:40100 Closing
[Sat Oct 31 02:39:12 2026] 127.0.0.1:36634 Accepted
[Sat Oct 31 02:39:12 2026] 127.0.0.1:36634 Closing
[Sat Oct 31 02:39:12 2026] 127.0.0.1:36650 Accepted
[Sat Oct 31 02:39:12 2026] 127.0.0.1:36650 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct 31 02:39:12 2026] 127.0.0.1:36650 Closing
[Sat Oct 31 02:39:14 2026] 127.0.0.1:36662 Accepted
[Sat Oct 31 02:39:14 2026] 127.0.0.1:36662 Closing
[Sat Oct 31 02:39:14 2026] 127.0.0.1:36666 Accepted
[Sat Oct 31 02:39:14 2026] 127.0.0.1:36666 Closed without sending a request; it was probably just an unused speculative preconnection
[Sat Oct 31 02:39:14 2026] 127.0.0.1:36666 Closing
[Sat Oct 31 02:39:14 2026] 127.0.0.1:36676 Accepted

Other app logs: downloads.log