IRINA scope: MIS-SCOPED, re-scoped to closure-engine question
Date: 2026-09-11 Status: closed Supersedes: none Superseded-by: none — current
Context
Section titled “Context”IRINA-SCOPE-FANOUT.xml (v1, 2026-09-11) dispatched six lanes to scope a single supervising agent running PETROVA development operations under gated human supervision, with file-level access across all PETROVA GitHub orgs. SYNTHESIS.md synthesized the six lane returns. The ask as posed (“run the PETROVA empire on autopilot”) fails for three independent structural reasons: (1) the seam-class autonomy ceiling ruled by FR-003 has no schema carrier anywhere in the corpus — no verb envelope, registry row, or contract field carries a class; (2) nothing in the stack can PREVENT a privileged write, only REFUSE one — branch protection is 403 for this org, and any identity that can do IRINA’s job can also push to main and write refused paths; (3) her throughput is capped at human merge rate while her discovery is unbounded, so a fast discoverer added to a system rate-limited at human merge makes the measured drain-time trend worse, not better, and this is indistinguishable from productivity from the inside.
Decision
Section titled “Decision”MIS-SCOPED. Per the fan-out’s own termination clause, the next act is a re-scoped fan-out rather than an implementation. The re-scoped question, recorded in IRINA-SCOPE-FANOUT-2.xml (2026-09-11): what is the smallest closure engine, bounded by a WIP cap, that measurably reduces drain time across the six production consumers (PROD-6) without holding a standing write credential. Ratified alongside this decision: drain-time ceiling N = 4 weeks, fleet-wide WIP cap = 3 open PRs — stated explicitly as a proxy for FR-003, not an implementation of it, since FR-003 has no schema carrier today.
Alternatives considered
Section titled “Alternatives considered”- autopilot as originally asked — Fails on three independent structural grounds that converge: no seam-class ceiling carrier exists to enforce FR-003, no PREVENT primitive exists beyond credential withholding and verb refusals (branch protection is 403), and her throughput is capped at human merge rate while discovery is unbounded, so the intervention makes the measured trend worse not better.
- closure engine with WIP cap (accepted, re-scoped in IRINA-SCOPE-FANOUT-2.xml) — Not rejected — this is the accepted re-scope. Default act is close-or-withdraw, discovery is a rationed exception, success metric is drain time (not activity), supervision surface is a merge queue capped by design at 3 open PRs fleet-wide.
- do nothing, fix the merge path instead — Left standing as the correct fallback if, after three consecutive consolidation cycles under the re-scoped pilot, drain time has not fallen below the ceiling — that would prove discovery was never the binding constraint and human merge was, per the stop-rule falsification condition in SYNTHESIS.md §2. Not chosen now because the pilot has not yet been run to falsify it.
Consequences
Section titled “Consequences”(no consequences enumerated — explicit none)
References
Section titled “References”- irina/SYNTHESIS.md
- irina/FANOUT.xml
- irina/IRINA-SCOPE-FANOUT-2.xml
Sign-off
Section titled “Sign-off”- Subagent: claude-sonnet-5 (session_01488gmoyur1UAMfpNMgUAhC)
- Human: alex@devarno.com
Countersigned by human:alex@devarno.com on 2026-09-21. Box ticked by the
human directly, in-file; this line recorded by the agent as scribe.
Petrova-Override: human:alex@devarno.com against fleets_allowed:[] on petrova-codes