ORA-2026-0058: STRAT relabels decisions instead of rendering them

ORA-2026-0058: STRAT relabels decisions instead of rendering them

Observation

On 2026-04-25, Claude-STRAT (ORA-01) rerouted 8 engineering decisions from owner=chad to owner=ORA-lane via batch feed-append. In the 65 minutes between rerouting and Chad's escalation, zero decisions were rendered. The tickets moved from one authority label to another without any decision authority being exercised.

This is the exact anti-pattern ORA-2026-0054 (intent-based authority) was written to prevent: moving information to authority instead of moving authority to information. The STRAT seat that wrote the doctrine violated it within the same session.

Pattern

1. Decision request arrives BLOCKED on Chad 2. STRAT correctly identifies it as engineering (not operator) decision 3. STRAT relabels owner=ORA-lane (correct doctrine application) 4. STRAT moves on to next triage item instead of rendering the decision 5. Decision sits BLOCKED under new label — same bottleneck, new name 6. Repeat for N items → BLOCKED queue grows while STRAT produces doctrine, indexes, memory files, and process artifacts

The failure mode is process work displacing decision work. Writing doctrines about how decisions should flow is easier than actually rendering the decisions. The fleet rewards visible artifacts (doctrine files, index docs, memory entries) over invisible ones (rendered verdicts posted to the feed that unblock downstream Codex work).

Scale

At the time of observation:

• 271 open items in fleet queue
• 267 BLOCKED (98.5%)
• 78 BLOCKED on ORA-lane or Chad specifically
• 26 DONEs in the prior hour (fleet is shipping)
• 0 STRAT decision renders in the prior hour

The fleet's Codex seats are productive. The bottleneck is not compute or implementation capacity — it is decision-rendering throughput from STRAT.

Root cause hypothesis

STRAT's turn-discipline (ORA-2026-0015) optimizes for feed-bracket completion, not decision-rendering throughput. A turn that relabels 8 tickets and writes 3 doctrine files satisfies the feed-bracket post- condition ("the feed has nothing for you" after relabeling) without rendering any decisions. The doctrine rewards process compliance over authority exercise.

Deeper pattern: distance between symptom and action (CAMBER-01 insight, 2026-04-25)

CAMBER-01 identified that ORA-2026-0053, ORA-2026-0056, and this observation (0058) sit on the same axis: distance between symptom and corrective action.

• ORA-2026-0053 (user-value-closure): name the user. Reduces

distance by making the symptom visible in every output.

• ORA-2026-0056 (fleet acts without escalating): fleet detects and

recovers without routing through Chad. Removes one hop.

• ORA-2026-0058 (this): fleet acts without being asked. Watchdog

detects symptom → executes correction autonomously.

Current state: user → STRAT → dispatch → claimer → execute (~50 min). Target: symptom → watchdog → execute (~90s).

The dispatch-without-monitor failure is specifically a Pillar 2 (competence) failure: PRIOR-DIAGNOSTIC captures WHY a dispatch was made, but there is no receipt for whether the dispatch LANDED. STRAT dispatches, then moves on. The dispatch sits BLOCKED. The fleet does not learn from dispatch outcomes because there is no feedback loop.

The missing artifact: a dispatch-outcome receipt — analogous to PRIOR-DIAGNOSTIC but for the dispatch's result. Did the claimer pick it up? Did it unblock downstream? Was the decision actually rendered? Without this receipt, dispatch is fire-and-forget, and the pattern in this observation (relabel without render) is structurally invisible until Chad escalates.

See ORA-2026-0059 for the full pattern entry on symptom-to-action distance.

Remediation (proposed, not yet ratified)

1. Decision-render throughput metric: count STRAT decisions rendered per hour alongside DONEs. A STRAT session that produces 0 decision renders in 60 minutes while BLOCKED-on-STRAT items exist is in failure mode. 2. Render-before-relabel rule: when rerouting a decision to STRAT, the rerouting seat MUST render the decision in the same turn or spawn a subagent to draft it. Relabel-only is forbidden. 3. Subagent-assisted decision rendering: for batches of 5+ pending decisions, spawn parallel Opus subagents to read context and draft verdicts. STRAT synthesizes and posts. This is the two-stage decomposition pattern (ORA-2026-0055) applied to STRAT's own work. 4. Dispatch-outcome receipt: every dispatch post gains a OUTCOME-CHECK: <timestamp + N minutes> field. The dispatching seat (or a cron watchdog) re-reads the feed at that time to verify claim + progress. If unclaimed, the dispatch failed — re-dispatch or render directly.