feedback_decision_analysis_one_at_a_time

Decision-analysis docs are scorecards — author one at a time

Rule. Decision-analysis documents (any artefact whose deliverable shape is “Question + Options + Trade-offs + Recommendation + Effects-per-option” awaiting Rich-input) are equivalent to scorecards. They fall under feedback_scorecards_one_at_a_time_optimal_sequence (triple-elevated) discipline.

Concrete protocol per decision in a session:

1. Author the analysis doc (status awaiting-rich-decision)
2. Surface it to Rich in chat (concise: question + recommendation + effects-per-option)
3. STOP and WAIT for Rich’s answer
4. RECALCULATE the next decision’s framing + options + recommendation based on Rich’s actual answer (not the prior projected answer)
5. THEN author the next analysis doc

Why: On 2026-04-26 (Sunday), in Phase 1 of the Option G+ audit, I batched 7 decision-analysis docs (decision-9 through decision-18) in a row without surfacing each one to Rich for input. Rich correctly intervened: “are these decisions being taken without me being asked any questions?” — flagging exactly the failure mode the triple-elevated rule was lifted to prevent. The docs were status: awaiting-rich-decision so nothing was actually decided, but the process drift was the issue: Rich got 7 decisions surfaced as a wall instead of one-at-a-time-with-his-actual-answers-shaping-the-next-one.

How to apply:

• Boundary test (pre-author): “Am I about to author the Nth decision-analysis when Rich hasn’t yet answered the (N-1)th?” If yes → STOP. Surface the (N-1)th first.
• Boundary test (during a multi-decision pass): “Has Rich responded to the most recent decision I surfaced?” If no → don’t author the next one.
• Even when the user has said “proceed to Phase X with the revised scope”: that’s SCOPE confirmation, not SEQUENCING permission. The one-at-a-time discipline is independent of scope.
• feedback_autopilot_prompt_techniques exemption: in genuine autopilot mode (Rich offline; explicit “ZERO QUESTIONS” directive), batching is acceptable IF the prompt pre-decides each fork via Decision Matrix. In a non-autopilot session like Phase 1, the discipline binds.
• Recalculate, don’t re-skin: when authoring decision-N’s doc after Rich’s decision-(N-1) answer, the new context may genuinely change decision-N’s framing / options / recommendation. Don’t just rubber-stamp the previously-projected answer.

Memory cross-refs:

• feedback_scorecards_one_at_a_time_optimal_sequence (triple-elevated parent rule)
• feedback_autopilot_prompt_techniques (the exemption pattern)

Boundary cases to remember:

• Multi-version of same decision (e.g., decision-1 v1/v2/v3) IS one-at-a-time (each version is a separate Rich-input cycle)
• Pre-flight research dispatches (multiple parallel research agents) is NOT this rule — research outputs evidence, not decisions; it’s the DECISIONS that need one-at-a-time
• Composite decision docs (e.g., decision-13-16-barlow-cross-module-primitives.md covering 4 primitives) — each primitive within is a sub-decision; surface ONE primitive at a time even within a composite doc