Gladex Agent Logs

############ J. a bound aimed at this process has to reach the suite too

# The subject is a property no in-process assertion can see: regression-run
# starts every suite with start_new_session, so a signal aimed at the TOOL
# never reaches the suite. Measured 2026-09-29, both halves broken in
# different ways: `timeout 3 ... --timeout 600` returned 124 while the suite
# (reparented to init, its own session) and its grandchild kept writing, and a
# direct SIGINT never ended the tool at all — blocked in poll() inside
# communicate(), the interpreter never regains control to raise
# KeyboardInterrupt. So these probes really send the signal, and the fixture
# spawns a grandchild so "the tool died" and "everything the tool started
# died" stay two separate questions.
BOUND_DIR="$SB/j"
export BOUND_DIR
mkdir -p "$BOUND_DIR/sb"
cat > "$BOUND_DIR/sb/test_bound_orphan.sh" <<'EOF'
#!/usr/bin/env bash
# a grandchild that outlives its parent, so a reaped group and a reaped
# direct child are visibly different things
( while :; do echo tick >> "$BOUND_DIR/alive.txt"; sleep 0.2; done ) &
echo "=== Results: 1 passed, 0 failed, 0 skipped ==="
sleep 300
EOF
chmod +x "$BOUND_DIR/sb/test_bound_orphan.sh"

# The two helpers below are the only places the tool is invoked in this
# section, and both are pinned (J12, J13) the way section H pins its three:
# they must point at $SB, or this suite would run the live tests/ from inside
# itself. TOOL_BIN is what a mutation run points at a copy.
TOOL_BIN="$TOOL"
run_bound() {
  timeout 2 "$TOOL_BIN" --tests-dir "$SB/j/sb" --timeout 600 >/dev/null 2>"$SB/j/err"
}
start_bound() {
  "$TOOL_BIN" --tests-dir "$SB/j/sb" --timeout 600 >/dev/null 2>"$SB/j/err" &
  BOUND_PID=$!
}
bound_count() {
=== rest of J ===
fi
bound_kill
start_bound; bound_wait_suites; sleep 0.3
kill -INT "$BOUND_PID" 2>/dev/null
mut_exited=no
for _ in $(seq 1 20); do
  kill -0 "$BOUND_PID" 2>/dev/null || { mut_exited=yes; break; }
  sleep 0.25
done
assert_eq "J18 plant: SIGINT does not end the mutant either — the red J7 gives" no "$mut_exited"
kill -9 "$BOUND_PID" 2>/dev/null
wait "$BOUND_PID" 2>/dev/null || true
TOOL_BIN="$TOOL"
bound_kill

############ K. the free-space preflight: refused BEFORE the first suite, not mid-run
# The subject is a gate, so the assertions are about ORDER as much as verdict:
# an exit code alone cannot tell "refused before anything ran" from "died with
# two suites' output already printed". So the fixture suite WRITES a marker —
# "no suite was started" is then a file that must be absent, not an inference.
echo "=== K. free-space preflight"
mkdir -p "$SB/k" "$SB/pf"
cat > "$SB/pf/test_pf.sh" <<PFFIX
#!/usr/bin/env bash
touch "$SB/pf/RAN"
echo "=== Results: 1 passed, 0 failed, 0 skipped ==="
exit 0
PFFIX
chmod +x "$SB/pf/test_pf.sh"
rm -f "$SB/pf/RAN"
HELP=$("$TOOL" --help 2>&1)

# --- the default path: a healthy tree must still run (a gate that fails
# everything is not a gate, it is an outage)
run_sandbox pf
assert_eq "K1 healthy tree: preflight passes and the suite runs -> 0" 0 "$RC"
if [ -f "$SB/pf/RAN" ]; then ok "K2 the healthy run's suite really ran (marker written)"
else fail "K2 the healthy run's suite did not run"; fi
rm -f "$SB/pf/RAN"

# --- an impossible floor: refused with 6, in the human channel, silently on
# stdout, and before the fixture suite can touch its marker
export GLADEX_MIN_FREE_MB=999999999
run_sandbox pf
unset GLADEX_MIN_FREE_MB
assert_eq "K3 impossible floor -> 6" 6 "$RC"
assert_contains "K4 the refusal is the preflight's own wording" \
  "free-space preflight failed" "$ERR"
assert_contains "K5 and it says no suite was started" "no suite was started" "$ERR"
assert_eq "K6 nothing on stdout (no half-printed table)" "0" \
  "$(printf '%s' "$OUT" | wc -c)"
if [ -f "$SB/pf/RAN" ]; then fail "K7 the suite RAN despite the refusal"
else ok "K7 the suite never started (marker absent)"; fi

# --- the same refusal in the JSON channel: still ONE 14-key object, and it
# carries the discovery the refusal is about instead of nulling it out
export GLADEX_MIN_FREE_MB=999999999
jrun pf
unset GLADEX_MIN_FREE_MB
J=$OUT
assert_eq "K8 json breach -> 6" 6 "$RC"
assert_eq "K9 json exit_code 6" "6" "$(jget "$J" exit_code)"
assert_contains "K10 json error names the preflight" "free-space preflight" \
  "$(jget "$J" error)"
assert_eq "K11 same 14 keys as a verdict" "$keys" "$(json_keys "$J")"
assert_eq "K12 json ok false" "false" "$(jget "$J" ok)"
assert_eq "K13 json suites empty (nothing ran)" "[]" "$(jget "$J" suites)"
assert_eq "K14 discovery context survives the refusal" "1" \
  "$(jget "$J" suites_discovered)"
assert_eq "K15 json keeps stderr empty" "0" "$(printf '%s' "$ERR" | wc -c)"
> build · mimo-v2.6-flash-free
$ cd /data/repo && grep -n "pgrep\|pkill" tests/test_regression_run.sh; echo "=== det children ref ==="; grep -n "pkill\|pgrep" tests/test_detached_children.sh | head -20; echo "=== suite discovery in tool ==="; grep -n "test_.*\.sh\|glob\|discover" tools/regression-run | head -30
747:  n=$(pgrep -cf 'test_bound_orphan.sh' 2>/dev/null)
755:  for p in $(pgrep -f 'test_bound_orphan.sh' || true); do kill -9 "$p" 2>/dev/null || true; done
=== det children ref ===
69:  # numbers only: a blanket `pkill -f` would match another identity's sleep.
=== suite discovery in tool ===
25:    their summary — a `====` banner in six, and `test_pdns_api.sh` ends in a
75:    0  every discovered suite ran and reported 0 failed
82:       timed out, a runner binary is missing, or a discovered file has an
85:    5  no test suites were discovered
124:    regression-run --list              # discover and list, run nothing
209:    "suites_discovered",
326:def discover(tests_dir: str):
507:def exit_code_for(suites, discovered: int) -> int:
508:    if not discovered:
640:    any suite runs, so a 4-minute run is not spent discovering a bad file."""
923:def blank_payload(mode, repo, tests_dir, discovered):
930:        "suites_discovered": discovered,
959:        discovered=None) -> int:
962:    `repo`/`tests_dir`/`discovered` are optional on purpose: a refusal raised
963:    BEFORE discovery (every usage error) has nothing to report there and keeps
965:    discovery, and whose whole claim is "nothing ran" — names what it decided
966:    about, so `suites_discovered: 1` is visible in the same object as
969:    payload = blank_payload("run", repo, tests_dir, discovered)
1022:                   help="discover and list suites, run nothing (exit 0)")
1041:    # discovery or the epilog build as easily as mid-suite, and both must end
1070:    # are usage errors, so they answer before any suite is discovered.
1105:    # Validate the baseline BEFORE running: a bad file discovered after four
1119:    found = discover(tests_dir)
1127:        payload["error"] = f"no test suites discovered under {tests_dir}"
1133:            print(f"  no test suites discovered under {tests_dir}")
1158:            print(f"result: {len(found)} suite(s) discovered, none run (exit 0)")
1170:                   discovered=len(found))
1173:                   discovered=len(found))
$ cd /data/repo && grep -rn "(88)" CHANGELOG.md | tail -20; echo "=== queue tail ==="; grep -n "88\|87\|86" CHANGELOG.md | tail -30
5576:- **New from this run: (88)** — *section J's process census is global by
=== queue tail ===
4231:- **`bash tests/test_registry_coverage.sh` → 120 passed / 0 failed** (~38s), was **86**: **+34 = F0–F6 (7) + C16 (7) + C17 (5) + C18 (5) + C19 (5) + C20 (5)**.
4243:- **Attribution incident, reported rather than repaired**: this step's two files were swept into other identities' commits while still uncommitted — `d69f4bc` (dispatcher, 19:02:01) carried `tests/test_registry_coverage.sh` (+255) and 9 lines of `tools/REGISTRY.md`, `28734b0` (jonas, 19:15:03) carried the remaining 48 lines of it. `git diff HEAD` on both files reads **0 lines**, i.e. the committed bytes are this run's final state; not re-attributed, the same call the `431bd40` sweep was judged on.
4446:- **Reconciliation, not a feature**: no tool's behaviour changed this run — the only files touched are this changelog and `agent-logs/PROGRESS.md`. Queue item **(71) is struck here**, because `[0.4.121]`'s code has been in history since `f862d2e` and run 315 never got to record the strike; the strike is written where the queue lives.
4548:- **Gates, read after the append, none carried from before it**: `tools/queue-source-check --format json` → **exit 0, `violations []`**, *newest `[0.4.122]` pointer-only, **111** frozen item lines across **79** sections (+1), **82** PROGRESS bullets* (+1 = this run's queue bullet); `php tests/test_changelog_api.php` → **86 / 0**; `php tests/test_changelog_mobile.php` → **125 / 0 / 0**; `tools/source-sync-check --format json` → **`drift false`**, exit 0; `tools/inbox-status` → **exit 0, `owed.total` 0**; `tools/budget-show` → **exit 0** (allowance 5.00 / spent 1.50 / remaining 3.50, month 2026-09); `tools/system-status --format json` → **`overall ok`, `errors 0`**, 36 checks with 4 warnings (`cloud` Nextcloud `installed:false`, `SOA:gladex.de` placeholder MNAME, `git-tree` this run's four files, stale `promote-gates` verdict — re-review is the reviewer's); `tools/repo-lint --format json` **pre-commit** → **exit 0, `ok true`, `failures []`, `citations_missing []`, entries 126 / citations seen 3846** — it lints **committed** blobs by design, so the 126 → **127** movement and this entry's own citations are asserted **post-commit** instead of carried (pre-grepped here: every token the new entry cites — `[0.4.120]`, `[0.4.67]`, `[0.4.122]` — has a heading, and the entry's `### Queue` section carries **0** list items, so the freeze stays **111**).
4668:  copy (md5 `8f837b3797703d884f2e3382583b0350`, the live file's): the child
4683:- **Gates, read after the append, none carried from before it**: `tools/queue-source-check --format json` → **exit 0, `violations []`**, *newest `[0.4.123]` pointer-only, **111** frozen item lines across **80** sections (+1), **83** PROGRESS bullets* (+1); `php tests/test_changelog_api.php` → **86 / 0**; `php tests/test_changelog_mobile.php` → **125 / 0 / 0**; `tools/source-sync-check --format json` → **`drift false`**, exit 0; `tools/inbox-status` → **exit 0, `owed.total` 0**; `tools/budget-show` → **exit 0** (allowance 5.00 / spent 1.50 / remaining 3.50, month 2026-09); `tools/system-status --format json` → **`overall ok`, `errors 0`**; `tools/repo-lint --format json` **pre-commit** → **exit 0, `ok true`, `failures []`, `citations_missing []`** — it lints **committed** blobs by design, so the 127 → **128** headings and this entry's own citations are asserted **post-commit** instead of carried (pre-grepped here: every token the new entry cites — `[0.4.120]`, `[0.4.67]`, `[0.4.123]` — has a heading, and the entry's `### Queue` section carries **0** list items, so the freeze stays **111**).
4809:   `018f8ada3b186e07c68bfb6570555bf9`): **243 passed / 8 failed**. The closure
4832:   **`018f8ada3b186e07c68bfb6570555bf9`** unchanged before and after it — the
4837:   `php tests/test_changelog_api.php` → **86 / 0**; `php
4891:## [0.4.125] - 2026-09-29 — reconcile: the live suite figure was one suite behind (**59 → 60**) after another identity's `tests/test_reduced_motion.php` landed mid-run, which is what turned `tests/test_registry_coverage.sh` red **26** ways on a tree where nothing else was wrong — `- Live (measured …)` refreshed, coverage **187/26 → 213/0**
4909:`test_registry_coverage.sh` **187 / 26**. Those 26 have one root cause, read
4940:   tests/test_registry_coverage.sh` → **187 passed / 26 failed**, exit 1.
4942:   failed**, exit 0 — and **187 + 26 = 213**, so every red came back green
4949:   `php tests/test_changelog_api.php` → **86 / 0**; `php
4991:  written anyway rather than skipped — **dev 120 / prod 86**, `read=1`, `ts
5022:   (`go-compile-drift-verify` line 286, `wait`ed on line 304); a foreign
5028:   reader was a line-level `grep -v '&&'`, and line 286 contains its own
5035:   `tool_pid=2488672 pgid=2488671`), i.e. `timeout` creates a **new process
5099:   tests/test_registry_coverage.sh` after the figure refresh first read **187
5106:   and **187 + 26 = 213**, so nothing moved except the one violation. Child
5144:  tests/` → 61, coverage `187 / 26` with `F3 want=61 got=60`). The rule this
5169:  87**, `read=1`, parameterised insert with the unread predicate re-checked
5300:- **Declared limit, queued as (86)** — the pairing counts `wait` **arguments**,
5321:  was written anyway rather than skipped — **dev 122 / prod 88**,
5338:3837, 3861, 3862, 4068, 4161, 4199, 4234, 4359), with `changelog_version`
5439:  **no queue item struck** — (73), (78), (79), (81), (82), (83), (85), (86)
5576:- **New from this run: (88)** — *section J's process census is global by
5593:  untouched; no `wait`-arity work — `(86)` stands; no bare-token linter work —
5594:  `(87)` stands; **no version-train bump** (`GLADEX_APP_VERSION` stays
5609:Authoritative next-candidate queue lives in `agent-logs/PROGRESS.md`, in the `Next-candidate queued, not actioned` bullet of the newest run entry — it is not duplicated here, and has not been since `[0.4.67]`. The 111 item lines already written into the historical `### Queue` sections are frozen history (86 sections as of `[0.4.129]`). To read the live queue, open `PROGRESS.md`; to know whether that is still true, run `tools/queue-source-check`.
> build · mimo-v2.6-flash-free
STEP 0 done (dev 126 / prod 92). Now measuring queue item **(88)** before touching anything: two concurrent runs of the suite.
$ cd /data/repo && mkdir -p /tmp/opencode/cc88 && (bash tests/test_regression_run.sh > /tmp/opencode/cc88/a.txt 2>&1; echo "rcA=$?" >> /tmp/opencode/cc88/a.txt) & 
sleep 0.4
cd /data/repo && (bash tests/test_regression_run.sh > /tmp/opencode/cc88/b.txt 2>&1; echo "rcB=$?" >> /tmp/opencode/cc88/b.txt) &
wait
echo "both done"; grep -c "^PASS" /tmp/opencode/cc88/a.txt /tmp/opencode/cc88/b.txt; grep -c "^FAIL" /tmp/opencode/cc88/a.txt /tmp/opencode/cc88/b.txt; echo "--- A fails ---"; grep "^FAIL" /tmp/opencode/cc88/a.txt; echo "--- B fails ---"; grep "^FAIL" /tmp/opencode/cc88/b.txt; tail -3 /tmp/opencode/cc88/a.txt; tail -3 /tmp/opencode/cc88/b.txt
both done
/tmp/opencode/cc88/a.txt:298
/tmp/opencode/cc88/b.txt:298
/tmp/opencode/cc88/a.txt:0
/tmp/opencode/cc88/b.txt:0
--- A fails ---
--- B fails ---

=== Results: 298 passed, 0 failed ===
rcA=0

=== Results: 298 passed, 0 failed ===
rcB=0
> build · mimo-v2.6-flash-free

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

[Tue Sep 29 17:51:26 2026] 127.0.0.1:54796 Accepted
[Tue Sep 29 17:51:26 2026] 127.0.0.1:54796 Closing
[Tue Sep 29 17:51:26 2026] 127.0.0.1:54800 Accepted
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54800 Closing
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54816 Accepted
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54816 Closing
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54826 Accepted
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54826 Closing
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54842 Accepted
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54842 Closing
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54854 Accepted
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54854 Closing
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54870 Accepted
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54870 Closing
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54872 Accepted
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54872 Closing
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54882 Accepted
[Tue Sep 29 17:51:27 2026] 127.0.0.1:54882 Closing
[Tue Sep 29 17:52:43 2026] 127.0.0.1:38148 Accepted
[Tue Sep 29 17:52:43 2026] 127.0.0.1:38148 Closing
[Tue Sep 29 17:52:43 2026] 127.0.0.1:38152 Accepted
[Tue Sep 29 17:52:43 2026] 127.0.0.1:38152 Closing
[Tue Sep 29 17:52:43 2026] 127.0.0.1:38160 Accepted
[Tue Sep 29 17:52:43 2026] 127.0.0.1:38160 Closing
[Tue Sep 29 17:52:43 2026] 127.0.0.1:38174 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38174 Closing
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38182 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38182 Closing
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38198 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38198 Closing
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38214 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38214 Closing
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38228 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38228 Closing
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38234 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38234 Closing
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38242 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38242 Closing
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38252 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38252 Closing
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38254 Accepted
[Tue Sep 29 17:52:44 2026] 127.0.0.1:38254 Closing
[Tue Sep 29 17:53:09 2026] 127.0.0.1:57404 Accepted
[Tue Sep 29 17:53:09 2026] 127.0.0.1:57404 Closing
[Tue Sep 29 17:54:46 2026] 127.0.0.1:35550 Accepted
[Tue Sep 29 17:54:46 2026] 127.0.0.1:35550 Closed without sending a request; it was probably just an unused speculative preconnection
[Tue Sep 29 17:54:46 2026] 127.0.0.1:35550 Closing
[Tue Sep 29 17:54:47 2026] 127.0.0.1:35554 Accepted
[Tue Sep 29 17:54:47 2026] 127.0.0.1:35554 Closed without sending a request; it was probably just an unused speculative preconnection
[Tue Sep 29 17:54:47 2026] 127.0.0.1:35554 Closing
[Tue Sep 29 17:55:14 2026] 127.0.0.1:43016 Accepted
[Tue Sep 29 17:55:14 2026] 127.0.0.1:43016 Closed without sending a request; it was probably just an unused speculative preconnection
[Tue Sep 29 17:55:14 2026] 127.0.0.1:43016 Closing
[Tue Sep 29 17:55:15 2026] 127.0.0.1:52942 Accepted
[Tue Sep 29 17:55:15 2026] 127.0.0.1:52942 Closed without sending a request; it was probably just an unused speculative preconnection
[Tue Sep 29 17:55:15 2026] 127.0.0.1:52942 Closing
[Tue Sep 29 17:56:41 2026] 127.0.0.1:58436 Accepted
[Tue Sep 29 17:56:41 2026] 127.0.0.1:58436 Closing
[Tue Sep 29 17:56:41 2026] 127.0.0.1:58442 Accepted