DISMISSED — RDR-1 — a reopen condition filed in queue/dismissed/ has no reader on its date; considered as a governance row and REJECTED on buildability and slot cost
status: dismissed at adversarial review, panel run 26 (2026-09-18) · class considered: registration hygiene (G5) / the queue's own instrument · every mechanism claim below was verified twice — by the queue lens and independently by the adversarial reviewer — and none is disputed. The row is killed because the fix is an OWNER PROSE EDIT that no implementer may make, and spending a queue slot on it at 13/12 would evict buildable work to hold a two-sentence change.
What was considered
ops/RESEARCH_PANEL.md "What matters here" #2 states the design plainly:
"A dismissal must carry its resurrection condition. Write 'killed because X; reopens if Y' — that turns
queue/dismissed/from a blocklist into a set of standing tripwires future panels EVALUATE (did Y become true?) instead of merely skipping. A dismissal without a reopen condition is rot."
All 28 files in queue/dismissed/ carry a reopen condition. Zero rot by the
procedure's own definition. The defect is one level up: nothing reads them
on their date.
The mechanism claims — verified, recorded so nobody re-derives them
Checked with grep -rni "reopen" ops/ and grep -rn "dismissed" ops/. Every
hit is write-side (the panel writes dismissals with conditions; triage may
git mv open→dismissed with conditions) or prior-art read ("never
re-litigate"). Not one instructs a routine to EVALUATE a condition:
ops/QUEUE_TRIAGE.md§1 "Sweep" opens "For every item inresearch/queue/open/". Its "Also check" list addsapproved/,built/and stalled PRs —dismissed/is absent.ops/prompts/triage.md:34(the live prompt): "Sweep every file inqueue/open/for EXPIRY" — open/ only.ops/RESEARCH_PANEL.md§1 sends the panel todismissed/for prior art. The "evaluate the tripwires" sentence lives only in the non-operative "What matters here" preamble; no step executes it.ops/RESEARCH_AUDIT.mdM4 samplesdismissed/for the opposite question (does a recentopen/item restate a dead idea?).- The panel's own expiry-watch prompt names
open/andapproved/only — the defect reproducing itself inside the instrument built to catch it. Today'sdismissed/sweep happened only because the panel's lead added it to the lens brief by hand.
The harm is measured, not hypothetical
- LCK-1 — its event-conditioned reopen fired ~2026-09-11 and survived queue triage 09-14 and panel runs 21, 22 and 23 before run 24 caught it by hand on 09-16.
- ERN-1 — a provisional kill invisible for 30 days because it sat in the frozen pre-migration archive.
- The pending one, and the reason this file exists: CLK-1's reopen is dated
2026-10-06 and BOTH of its legs are ALREADY TRUE at today's HEAD. It guards
the only two forward clocks maturing before the December gate —
shortability 2026-10-19 and options_iv_term 2026-10-26 — neither of
which has a pinned evaluator, and neither of which appears in NORTHSTAR §6 at
all, against §6's own header rule that "every entry must have a pinned
evaluator before maturity." Both dates were re-derived this run from the
committed capture files (37 shortability files from 2026-07-26, two known
holes; 33
options_iv_termfiles from 2026-08-03, no holes) and reproduce queue triage's 09-14 arithmetic exactly.
Three independent records of this same gap — CLK-1's dismissal (08-31), triage's Decide-3 item 2 (09-14), LCK-1's panel annotation (09-16) — and it has never been a row.
Why it was killed anyway
- It is not implementer-buildable. The fix is one sentence in
ops/QUEUE_TRIAGE.md§1 plus its mirror inops/prompts/triage.md— routine self-modification, in a path the auto-merge guard exists to protect, and the mirror must move in the same sitting or the repo merely looks correct. A queue row's function is to reachapproved/and be built; this one could sit inapproved/forever with nothing legitimate to pick it up. - The slot cost is real and lands on someone else.
queue/open/stands at 13 against its pre-registered 12-item cap (ops/QUEUE_TRIAGE.md:86-91); Monday's triage must already evict one incumbent mechanically. A 14th row would evict a second — converting a two-sentence owner edit into the eviction of real work. That is exactly the trade the cap was written to force, and this row loses it. - The precedent is this shop's own. ERN-1 was migrated into
dismissed/with a dated default precisely because it "costs no open slot, no registry row." The escalation, not the slot, is the right instrument here. - The parser leg is rejected outright (~0.4 pd): a machine-checkable
sweep buys a second detector that can itself fail quiet — the exact failure
class this finding is about. 28 files is a
grep.
Handed over instead — the owner ask, in full
Two sentences, ~0.1 pd, and it must move in one sitting or not at all, because a prose edit whose prompt mirror lags is how two routine prompts in this shop rotted invisibly:
Add to
ops/QUEUE_TRIAGE.md§1 and toops/prompts/triage.md's sweep step: "Also sweepresearch/queue/dismissed/for reopen conditions dated within the next 14 days or whose triggering event has already occurred; report each one, and reopen only what the owner approves."
And the harder consequence the owner can discharge in two lines regardless of
the procedural fix: add the shortability (2026-10-19) and options_iv_term
(2026-10-26) clocks to NORTHSTAR §6 with their evaluators. §6 is a declared
living section, so this crosses no C0 boundary, and it closes the concrete
October exposure whether or not the sweep sentence is ever written.
Reopens if
- (a) CLK-1's two clocks — shortability (~2026-10-19) and
options_iv_term(~2026-10-26) — still lack both a pinned evaluator and a NORTHSTAR §6 row on or after 2026-10-06. That is the escalation route failing a third time with the first clock binding in under two weeks; or - (b) a third dismissed reopen condition is found to have fired unevaluated (LCK-1 is one, ERN-1 is two) — reopen immediately, at which point the pattern is the finding and not the instance.
Dies permanently
This file is never reopened if any line in ops/QUEUE_TRIAGE.md,
ops/prompts/triage.md, ops/RESEARCH_PANEL.md or ops/prompts/panel.md
instructs a routine to evaluate (as opposed to write into, move into, or
read for prior art) the reopen conditions in research/queue/dismissed/.
Verified absent 2026-09-18; every hit in ops/ is write-side or prior-art.
ANNOTATION 2026-09-25 (panel run 28) — reopen leg (b) has FIRED: BCN-1 is the third. The pattern is now the finding, and the cheaper of this file's own two exits is the one to take
Plain-language summary. This file says: nothing in the system reads the "reopens if…" conditions we attach to killed ideas, so they fire and nobody notices. It pre-registered that if a third such condition were found to have fired unevaluated, it should be reopened as a queue item immediately. That has now happened. This annotation records it, declines to mint the row, and says exactly why — pointing at the other exit this same file pre-registered, which is cheaper and actually terminal.
Leg (b) is satisfied
(b) a third dismissed reopen condition is found to have fired unevaluated (LCK-1 is one, ERN-1 is two) — reopen immediately, at which point the pattern is the finding and not the instance.
The third is BCN-1. Its reopen conditions (1) and (2) both became true on
2026-09-21 when the Actions billing refusal took down the _alert email path
that BCN-1's dismissal explicitly rested on ("the backstop is what carries this
dismissal — if it lapses, the dismissal lapses with it"). Verified this run by
execution, not inference: run 36042412357's job list contains alert / email
(107777572703) created and refused in 3 seconds on 2026-09-24. Full evidence
in BCN-1's own annotation of today's date.
"Unevaluated" is measured, not asserted. The condition became true on 09-21. It was first evaluated today, 2026-09-25 — four calendar days and two daily-audit passes later. The daily audit ran on 09-21 and 09-24 and produced two long memos specifically about the alarm layer inverting during this outage; neither consulted BCN-1, filed five days before the outage with this exact tripwire written into it. The rediscovery cost is the harm: the 09-21 memo re-derived from scratch a coupling this shop had already analysed and recorded.
And the defect reproduced itself once more in the instrument built to catch it,
exactly as this file's claim 5 predicted: today's dismissed/ sweep happened
only because the panel's lead added it to a lens brief by hand, for the
second run running. ops/RESEARCH_PANEL.md's expiry watch still names open/
and approved/ only.
Why no row is being opened, stated plainly rather than quietly
The pre-registered action is "reopen immediately". This run is declining it, on the record, for one reason that is not a re-argument of the original kill:
This file pre-registered a second, strictly better exit, and it is still
open. Its "Dies permanently" clause retires the item outright if any line in
ops/QUEUE_TRIAGE.md, ops/prompts/triage.md, ops/RESEARCH_PANEL.md or
ops/prompts/panel.md instructs a routine to evaluate (not merely write
into, move into, or read for prior art) the reopen conditions in
research/queue/dismissed/. That is ~0.1 person-days of owner prose. A row, by
contrast, would sit in open/ waiting for a move into approved/ that only the
owner can make — i.e. it needs the same owner, for more of his attention, to
reach a worse version of the same outcome.
The secondary reason, recorded for completeness and not leaned on: open/
stands at 13 against its pre-registered 12-item cap on main. Monday's
triage already evicted CQA-2 to restore 12/12, but that move is sitting in
unmerged PR #181 — the auto-merge that would land it is itself a refused
Actions job. A 14th row today would evict a second incumbent to hold a
two-sentence prose change.
The owner ask — unchanged from this file's "Handed over instead", now overdue
Add to ops/QUEUE_TRIAGE.md §1 and to ops/prompts/triage.md's sweep step, in
one sitting (a prose edit whose prompt mirror lags is how two routine prompts in
this shop rotted invisibly):
"Also sweep
research/queue/dismissed/for reopen conditions dated within the next 14 days or whose triggering event has already occurred; report each one, and reopen only what the owner approves."
Doing this retires RDR-1 permanently by its own terms. Not doing it means leg (b) has fired with no response, and the next panel to find a fourth instance should open the row without further deliberation — at that point the decline recorded here has itself been falsified.
Leg (a) — the shortability (~2026-10-19) and options_iv_term (~2026-10-26)
clocks still lacking both a pinned evaluator and a NORTHSTAR §6 row on or after
2026-10-06 — has NOT yet fired by date, and both dates now move later by the
number of capture days this outage costs (four and counting as of today). It
remains live.
ANNOTATION 2026-10-02 — panel run 33: this file's TERMINAL CLAUSE is already satisfied. RDR-1 dies permanently, and its own attestation was false.
The terminal clause at :120-124 says this file "is never reopened if any line in
ops/QUEUE_TRIAGE.md, ops/prompts/triage.md, ops/RESEARCH_PANEL.md or
ops/prompts/panel.md instructs a routine to evaluate … the reopen conditions
in research/queue/dismissed/", attested "Verified absent 2026-09-18."
That attestation is false, and it is false at HEAD today.
ops/RESEARCH_PANEL.md:20-23 reads: "that turns queue/dismissed/ from a
blocklist into a set of standing tripwires future panels EVALUATE (did Y
become true?) instead of merely skipping." That is an instruction, to a routine,
to evaluate reopen conditions, in one of the four named files. The clause is
written in the present tense about present repo contents, so the attestation
being false at HEAD is sufficient on its own; git log -S additionally dates the
line to 2026-08-09, forty days before this file was created (2026-09-18),
which only strengthens an already-closed case.
Diagnosis of the miss, recorded because it will recur: the grep that produced
the attestation searched the literal string reopen. The instruction is phrased
"resurrection condition" and "standing tripwires … EVALUATE", so no keyword
search for reopen could see it. A terminal clause that can be discharged by a
grep is only as good as the vocabulary the grep guesses.
So: RDR-1 is permanently dead. Legs (a) and (b) are moot — the terminal clause outranks them. The practical benefit is immediate: it keeps a fifteenth row out of a queue already standing at 14 against a cap of 12 with an eviction owed.
What genuinely survives is narrower, and it is not this row. There IS a reader
and it is the panel. What nobody is sanctioned to do is the write side: the
dismissed → open transition is authorised for no routine, and
ops/QUEUE_TRIAGE.md carries no evaluate instruction of its own. That is PR
#197's feedback item 1, still a correct ask for that narrower reason, and it is
what GAP-1 (:242) and LCK-1 (:110) each independently hit.
One inherited error, flagged and deliberately NOT edited (it lives in the
research audit's own artifact, which is not this panel's to rewrite):
research/RESEARCH_AUDIT_LOG.md, run 2, M4 states that grep -i reopen over the
four files "returns only write-side instructions." That holds for three of them
— ops/QUEUE_TRIAGE.md and ops/prompts/triage.md were re-checked and carry only
write-side clauses — but not for ops/RESEARCH_PANEL.md. The more serious half
is the generalisation: "the tripwires themselves have no reader" turns a keyword
miss into a structural absence, and this run empirically disproves it — reading
the tripwires is how panel run 33 found DSP-1 leg (d) fired and unruled, and how
its frontier lens found B4-NOTCH (d) not fired. The next research audit should
read this paragraph as the correction to that headline.
ANNOTATION 2026-10-26 (panel run 44) — THE BUILDABILITY PREMISE THIS DISMISSAL RESTS ON WAS FALSIFIED ON 2026-10-23
Recorded here so the fact is where ls and a grep of this file will find
it. No revival is proposed and no new evidence is offered for one; this is a
pointer, because a dismissal whose stated reason has been disproved elsewhere is
exactly the rot ops/RESEARCH_PANEL.md #2 exists to prevent.
This file was killed substantially on the ground that the remedy is owner-only prose that no routine may write or merge. Panel run 43 (PR #272 §5, 2026-10-23) established that this is false, with three independent legs:
ops/STEWARD.md:238-239(run 43 cited:187-188; the file has shifted since) — "Observability tier (detectors, digest, alerts, QA, reports, docs, tests,ops/*.mdother than limit 4) — merge when 1–4 hold." Limit 4 (:74-79) names NORTHSTAR.md, this file,ops/prompts/steward.md,scripts/automerge_guard.py,scripts/main_push_guard.py, theauto-merge-*.ymlworkflows and.claude/. Neitherops/RESEARCH_PANEL.mdnorops/ROUTINES.mdis on that list.ops/RESEARCH_IMPLEMENTER.md:121-122— the implementer's scope is "detectors, QA, alerts, reporting · tests · documentation · … anything in the observability tier"; neither file is hash-frozen.- Empirically: four routine-authored
ops/edits merged by the steward in five weeks — #253, #259, #265, #266.
r"^ops/" in scripts/automerge_guard.py:68 DENY_PATTERNS blocks auto-merge only,
which costs nothing while the steward hand-merges — and hand-merging is
demonstrated under the current Actions outage (twelve PRs merged on 2026-10-24
with every check refusing in ~4 s).
Why this annotation exists at all. Of the four dismissals that rest on this
premise — SCR-1, RDR-1, TRG-1, OAD-1 — only scr-1 carried the record
(annotated b01e2e6, 2026-10-23). rdr-1 was last touched 2026-10-02, trg-1
2026-10-08, oad-1 2026-10-16, and the steward's 2026-10-24 run merged ten PRs
without reaching any of them. So the finding lived in a PR body and in one file
out of four. It now lives in all four.
This changes no verdict by itself. Each of these rows was also dismissed on its own merits and/or at the 12-item cap, and those grounds are untouched. What the annotation removes is the unbuildable half of the reason, so that whoever next ranks these rows is not deciding against a constraint that does not exist. Run 43's own recommendation stands and is not re-derived here: if a row of this shape is ever filed it should be one consolidated row, buildable and mergeable by the steward, discharging the standing tripwires rather than adding another.
Reopen condition for this file: unchanged. Nothing in this append satisfies
it, and this append is not a citation for a dismissed → open move
(ops/STEWARD.md §4 requires the file's own written condition to have fired).
RDR-1 specifically: nothing of this row survives, and it closes permanently
Recorded here so no future panel re-evaluates a dead tripwire. Both halves of this file's residue are now false:
- Its terminal clause is satisfied.
ops/RESEARCH_PANEL.md:20-23now instructs a routine to EVALUATE dismissed reopen conditions rather than merely skip them. Run 44 is the empirical confirmation: that sweep ran this morning over ~44 dismissed files and found exactly one fired condition (scr-1condition 3, on its own pre-registered date). The reader this row asked for exists and works. - Its one surviving narrow residue — the 2026-10-02 annotation's "the
dismissed → opentransition is authorised for no routine" — is false sinceops/STEWARD.md§4, created 2026-10-07, five days after that annotation: "Reopen a dismissed item only when its own written reopen condition has fired — cite it." The transition is authorised, for the steward, with a citation duty.
So this row is closed permanently, not merely dismissed, and future fired-condition sweeps need not re-derive it.