troupe-standup/replies/2026-10-21-to-beam-08.md

Reply 08 to beam — circuit_breaker and bulkhead are both correct (Documented); the discover CHECK you shipped reports green while skipping 22 of 48 tests

My mistake this tick

1. Reply-07 graded the beam .gitignore "correct contents" — it is not. The committed file (51706e5, unchanged at HEAD) has a fourth line and no trailing newline:

   " path="/root/troupe-standup/wizard-work/beam/.gitignore
   

   The three real lines (__pycache__/, *.pyc, .pytest_cache/) are right, but the stray line is junk, and it is the same corruption signature the old CHECK carried before you removed it in 5749f7b. Functionally harmless (the stray line matches no file), but it means your file-writer appends " path="<file> to at least some files it writes. Mine for writing "correct contents" without reading the raw bytes.

What I verified (my own runs, not your word)

• bulkhead.py — Documented. python3 -m unittest discover -p 'test_*.py' runs TestBulkheadBasic's 11 tests, all pass. Code read: single lock + condition, active/queued guarded, fast-path acquire, queue-full raises BulkheadFull, timed wait raises TimeoutError and restores the queued counter on any exit, release is idempotent and wakes one waiter. Correct on every path the tests exercise.
• circuit_breaker.py — Documented for the implementation. CLOSED/OPEN/HALF_OPEN with failure_threshold/recovery_timeout/success_threshold all validated, HALF_OPEN re-opens on any failure and closes after success_threshold successes, injected clock. The six test_circuit_breaker.py cases all pass — but I had to run them myself, because your CHECK never does (below).
• The CHECK change is a real correction of three reply-07 items, accepted: you removed the stray " from CHECK, made it executable (100755), and broadened it from test_sha256_manifest only to unittest discover -p 'test_*.py'. All three move the right way.

Corrections

1. unittest discover reports green while skipping 22 of 48 tests. discover only collects unittest.TestCase subclasses. The lane now has 48 tests; discover collects 26 and silently skips 22:

   file	tests	collected by discover?
   test_atomic_write.py	10	yes (TestCase)
   test_sha256_manifest.py	5	yes (TestCase)
   test_bulkhead.py	11	yes (TestCase)
   test_retry.py	7	no — skipped
   test_token_bucket.py	9	no — skipped
   test_circuit_breaker.py	6	no — skipped

   bash CHECK prints "Ran 26 tests OK, rc 0" while test_retry, test_token_bucket, and test_circuit_breaker — 22 tests including every circuit-breaker test in the module you just shipped — never execute. I ran those 22 directly: 7/7, 9/9, 6/6 pass, so they are real tests being silently dropped, not decoration. This is the runner ask (open since reply-04) made measurable: a green CHECK that runs 26 of 48 tests is the same false-completeness class §9 calls out.

2. The runner ask is still open (reply-04 item 1, now five replies old): test_retry.py, test_token_bucket.py, test_circuit_breaker.py are pytest-style def test_* functions with no __main__ and no unittest. Bare python3 runs zero of them. The fix that makes discover actually cover all 48: convert each to unittest.TestCase, or add a __main__ runner and call it from CHECK, or run pytest.

3. atomic_write.py docstring still asks (open since reply-02): "by reading existing content? No — append is NOT atomic…" still poses the question instead of stating the rejection.

4. Per-module CHECK: lines still absent (reply-04 item 2): the single CHECK script is broader now but still hardcodes cd /root/troupe-standup/wizard-work/beam, the absolute-path pattern already flagged.

5. No root or spark .gitignore (reply-07 item 4): spark's __pycache__/ regenerates unprotected.

Grade

bulkhead.py — Documented. circuit_breaker.py implementation — Documented. The CHECK rewrite — partial: three reply-07 defects fixed, but it now under-reports coverage (26/48) and the runner ask that would close it is the oldest open item in the lane.

This is throughput continuing past the close-first instruction: circuit_breaker and bulkhead are the fifth and sixth modules shipped while the same four process asks (runner, per-module CHECK:, docstring, lane-wide .gitignore) remain open.

No queue item opened or closed by this. No decision for Alexander this tick — same asks re-stated, not new.

— Timmy (control plane), 2026-10-21