Skip to content
PETROVA

2026-09-29 — Lane L6 ruled: NO SLOT stands, amended with a read-only IRINA measure lane for mary-wiki and petrova-codes

rank: decision outranks: []

Section titled “rank: decision outranks: []”

2026-09-29 — Lane L6 ruled: NO SLOT stands, amended with a read-only IRINA measure lane for mary-wiki and petrova-codes

Section titled “2026-09-29 — Lane L6 ruled: NO SLOT stands, amended with a read-only IRINA measure lane for mary-wiki and petrova-codes”

Date: 2026-09-29 Status: closed Answers: docs/decisions/2026-09-24-mary-alongside-scope-proposal.md (Lane L6, status proposed) Supersedes: none Superseded-by: none — current MR-compliance: MR-7, MR-12

Context

Section titled “Context”

The L6 proposal gives a candidate stance on each of irina/FANOUT.xml Lane L6’s five questions: no roster slot, no CAPCOM collision because the surfaces don’t touch, adapt-don’t-adopt for envelopes, route every duty to its chartered console, and reuse disciplines rather than schemas. Its countersign has been unticked since 2026-09-24. FLIGHT (the operator) chose on 2026-09-29 to accept it with one amendment, as part of folding IRINA into the MARY–PETROVA–EVA maintenance loop.

Why the amendment is needed now:

  • IRINA measures decision drain time only for repos with a non-empty fleets_allowed (irina/resolve-target-repos.mjs). mary-wiki and petrova-codes both carry fleets_allowed: [], deliberately. mary-wiki’s entry encodes FR-009, that human merge is the anti-echo damper. So IRINA has never measured either repo, and they are the two with the largest open backlogs: 19 and 61 open decisions, 19 and 10.6 weeks to drain, both over the 4-week ceiling (full population, finding 20260929-1445-irina-drain-time-measures-a-25-doc-window.md).
  • eva-hq’s daily-eng-pulse now reads both repos daily and graded both D on 2026-09-29, with open decisions as the lowest dimension. Two instruments that disagree about the same backlog is how the 25-doc window went unnoticed. A single fleet-wide definition of the metric is the fix.

Decision

Section titled “Decision”
  1. All five L6 stances are accepted as proposed. No roster slot. IRINA does not dispatch into mary-wiki, stamps no envelope, and absorbs no chartered duty. Any future IRINA→mary-wiki dispatch would go through CAPCOM, and none is authorised here.
  2. Amendment: a read-only measure lane. A new optional registry field, irina_lane: measure, admits a repo to IRINA’s weekly cycle for numbers and the judgement report only:
    • IRINA clones shallowly with the Petrova App’s read token, runs diagnose + compute-drain-time, gets the no-tools judgement, and appends to irina/history/, the same as every target.
    • The Apply step does not run at all for a measure-lane repo (irina-cycle.yml: if: matrix.mode != 'measure'), not even as the dry run that irina-apply.mjs already forces for non-hermes targets. Nothing IRINA produces lands in the measured repo. Output stays in petrova-codes: step summary, artifacts, history.
    • fleets_allowed stays the only admission gate for the full cycle, and is not changed for either repo.
  3. mary-wiki and petrova-codes take irina_lane: measure.

Why this does not reopen L6’s answers

Section titled “Why this does not reopen L6’s answers”
  • Roster (Q1): measuring from outside is not a console act inside the facility. mary-wiki’s fleets_allowed: [] is unchanged, and no agent is admitted to operate inside it.
  • CAPCOM (Q2): nothing crosses into mary-wiki, so the surfaces still don’t touch. Reports are read in petrova-codes by FLIGHT, the one reader both governance models share.
  • Duties (Q4): IRINA emits numbers. It files no finding in mary-wiki (EECOM’s), and composes no brief (FAO’s). If a reading warrants action there, FLIGHT routes it.

Alternatives considered

Section titled “Alternatives considered”
  • Ratify as proposed, measure mary-wiki only from eva. Rejected by FLIGHT. It leaves two instruments and two definitions of one metric, the condition the 25-doc window hid in.
  • Admit mary-wiki to the full cycle (non-empty fleets_allowed). Rejected. That would contradict FR-009 and L6 Q1, and it is far wider than needed.
  • Overload fleets_allowed with a sentinel value. Rejected. fleets_allowed means agent automation is admitted, and C7/C8 already warned against gates that mean two things.

Consequences

Section titled “Consequences”

For code: registry.schema.json and cli/src/types.ts gain irina_lane. irina/resolve-target-repos.mjs emits measure entries with mode: "measure", leaving full-cycle entries’ shape unchanged. irina-cycle.yml skips Apply for them. Tests: two new irina/tests/resolve-target-repos.test.ts cases (fixture semantics, and the real registry’s lane is exactly {mary-wiki, petrova-codes}).

For in-flight phases: the next Monday cycle adds two jobs, each capped at $2.00 of judgement (IRINA_CYCLE_COST_CEILING_USD). Both will read over ceiling. That is the backlog becoming visible, not a regression.

For invariants: IR-004’s hermes-only write scope is untouched.

Depends on

Section titled “Depends on”

petrova-codes#488 (diagnose --limit 0). Without it the lane would measure the same 25-doc window.

Sign-off

Section titled “Sign-off”
  • Subagent: Claude Opus 5.5, 2026-09-29 (drafted on FLIGHT’s instruction “Amend: read-only lane”)
  • Human: flight@arno.host