# DISMISSED — OAD-1 — nothing says what a routine OUTAGE does to a dated condition that matured while the routine was dark, and the one rule that exists is conditioned on a declaration nobody can make retroactively

status: dismissed on the day it was raised · panel run 39 (2026-10-16), queue-and-clocks candidate C3, dismissed at the chair's adversarial pass · class: governance / registration hygiene (G5) · effort would have been ~0.1 pd of prose

## The claim, in plain language

Thales has a rule for what happens to a deadline when the owner is **away on
purpose**: `ops/ABSENCE.md:30-32` — *"expired items get flagged, never
escalated to 'lapsed' — a deadline cannot lapse at a person who declared they
are away."*

It has **no rule** for what happens when a *routine* is dark. The absence rule
is conditioned on a **declaration**. An outage is not declared, cannot be
declared in advance, and cannot be declared afterwards without becoming the
post-hoc edit C0 forbids. So a condition that came due inside a dark window
has no governing rule, and the next reader must either invent one — which is
gate-shopping — or read the date as lapsed, which silently discards the
condition.

`grep -rniE "outage" ops/*.md` filtered for expiry/lapse/clock language returns
**zero hits**. `ABSENCE.md`'s own banner says it is *"largely superseded
2026-10-07"* and *"must be revised"* before December.

## Two live instances, both created by the 2026-10-12 → 10-15 outage

- **`dismissed/trg-1`** leg (a) is keyed to *"the 2026-10-12 report"*, which
  does not exist — `outbox/triage` has no commit since 10-05. Its permanence
  clause (*"if the 2026-10-12 and 2026-10-19 reports both carry the sweep,
  this dismissal becomes permanent"*) is therefore **unsatisfiable as
  written**: a condition designed to close itself can now never close.
- **`dismissed/hire-1`**: *"Reopen **automatically** if frontier-cadence
  counter (c) fires on 2026-10-12."* The substantive count reached 8 of 8 in
  the window ending 2026-10-11, the declaring routine (triage) was dark on
  10-12, and "automatically" names no actor. This is the actor-less
  auto-reopen `trg-1:57` already flagged, now with its date behind it.

## Why it is dismissed rather than opened

**1. The durable home is agent-denied, and that exact argument already killed
a sibling.** The rule belongs in `ops/ABSENCE.md`, matched by
`scripts/automerge_guard.py:68` (`^ops/`). No routine can merge it. The
adversarial reviewer killed TRG-1 on precisely this ground — *"spending a
queue slot on something nobody in the loop can execute is implementation
fantasy"* — and that reasoning binds here. The two mitigations offered when
this was raised (the queue is at 10/12 so no eviction is spent; the rule could
live in `queue/` and be routed to the owner as a `NEEDS-OWNER:` line) make the
row **cheaper**, not **executable**, and cheapness is not the bar.

**2. Its own kill criterion fires in three days, which is faster than any
build.** The criterion proposed with it was: *if Monday 2026-10-19's triage
handles `hire-1` and `trg-1` correctly with no written rule, the rule is
unnecessary.* Triage's own procedure already carries the instruments — the
14-day dismissed sweep (`ops/QUEUE_TRIAGE.md:94-102`, whose first real
execution is that same Monday) and the counter-(c) declaration with its
pre-committed action (`:117-133`). Opening a row to pre-empt a test that
resolves on its own in three days is the quota behaviour `NORTHSTAR.md` §3
forbids.

**3. The 2026-10-15 ruling is evidence the gap is already being closed without
a written rule.** Four conditions matured in or near the dark window and all
four were answered by the owner on 10-15 (`b927884`), each pinned before its
criterion was computed. The observed failure rate of "a matured condition gets
mishandled after a dark window" is, so far, zero.

## Reopen condition — this is the Monday test, pre-registered now

Reopen as a row if **any** of these becomes true:

1. **The 2026-10-19 triage run does not name `hire-1` and `trg-1` in its
   decide-list**, or names them and treats either date as simply lapsed — i.e.
   the condition is discarded rather than flagged-and-evaluated. First
   evaluation: the 2026-10-19 outbox report. If that report does not exist
   either, condition 2 applies instead.
2. **A second dark window of ≥2 consecutive weekdays occurs** (any routine,
   any cause) **and** at least one dated condition matures inside it — at
   which point this is the second instance of the class and the pattern is the
   finding, per `rdr-1:128`.
3. **Any routine or session resolves a condition that matured in a dark window
   by moving its date**, rather than reading it as-of its original date on the
   evidence then available. That is the gate-shopping this row exists to name,
   and one instance is enough.

## The three-line rule, recorded here so it is not re-derived

Kept verbatim so that whoever does get to write `ops/ABSENCE.md`'s December
revision — which that file already demands — does not start from scratch, and
so that condition 1 above has something concrete to be measured against:

> (i) A dated condition whose evaluation date falls inside a window where its
> evaluating routine produced no beacon is **FLAGGED, not lapsed**, and is
> evaluated by the first run of that routine after the window.
> (ii) Its date does **not** move — the condition is read as-of its original
> date, on the evidence available then.
> (iii) A clause whose closure requires a report from inside such a window is
> **re-cut by the owner or struck**, never deemed satisfied.

Routed to the owner as a one-line ask rather than built: it is three lines of
prose in a file no agent can merge, and `ABSENCE.md` needs its December
revision regardless.

## Adjacent ground — do not duplicate

**CLK-1** (fired 2026-10-06) is clocks with no §6 row or evaluator. **RDR-1**
(permanently closed 2026-10-16 by its own never-reopen clause, now satisfied by
`ops/QUEUE_TRIAGE.md:94-102`) is nobody *reading* dismissed conditions.
**TRG-1** is one prompt omitting one duty. **RAD-1** (opened this run) is which
inputs a pinned criterion counts. This row was the distinct question of dated
conditions being **unrulable after a dark window**, and it is parked, not
answered.
