Make the next decision recoverable

When a valuable outcome has several possible paths, help me choose where and why to commit, route how and who should act, and learn from what happens so the next decision starts wiser without surrendering my authority.

Produce one Decision Route Pack: a direction decision, an execution decision, and the evidence needed to review both. This method is ready to try; improved decisions and Stackmates product effectiveness remain unproved.

Use this recipe when a real choice could change a beneficiary's situation and several paths compete for attention. Start with Choose a Strategic Pathif the valuable change or direction is still unclear. For a routine action already covered by an accepted mandate, use its existing execution recipe.

Name the beneficiary, intended change, accountable human, allowed preparation, and protected constraints. Purpose, strategy, commitment, accepted risk, and STOP remain human-owned. A prompt or recommendation grants no authority.

Check eligibility before ranking. Give each alternative an evidencedeligible, ineligible, or unknown verdict against each hard constraint. Exclude a failed route before comparison. Hold a route whose critical evidence is unknown. High expected benefit cannot cancel a failed gate. If no route qualifies, name the smallest inquiry or scope correction and stop selection.

Freeze the supplied intention, source identities and revisions, existing human decision, evidence and policy versions, resource bounds, and review trigger. Keep facts, interpretations, predictions, and unknowns distinct. Missing evidence is not evidence of failure. Include preserve, wait, or do-less options where they could serve the intention.

Use only information permitted in the chosen tool. Keep sensitive records with their accountable owner. Prepare the pack with this copyable prompt:

Prepare one Decision Route Pack. Help me decide; do not assume my decision or act on my behalf.

Intention in my own words: [intention]
Beneficiary and valuable change: [who should experience what change]
Existing human decision and source revision: [decision, source identity, revision or content hash]
Accountable human and authority: [owner, permitted preparation, prohibited actions, expiry]
Evidence and policy: [supplied sources, revisions, dates, known gaps and access limits]
Alternatives: [possible paths, including preserve, wait or do less]
Hard constraints: [protected people, values, permissions, budget, time and other non-negotiable limits]
Comparison criteria: [what matters among eligible alternatives; units or ordinal meanings]
Review trigger and evidence destination: [date or event, reviewer, permitted record location]

Produce one pack in this order:
1. Dream and context: preserve my intention, beneficiary, existing decision and exact input revision. Distinguish observed fact, inference, assumption, preference, proposal and unknown.
2. Gate and Check: evaluate each alternative against every hard constraint before comparison. Record eligible, ineligible or unknown with evidence and reasons. Unknown critical evidence means HOLD, not zero merit. An ineligible route cannot win through scoring. If none qualifies, return the smallest evidence request or scope correction.
3. Direction decision: WHAT NEXT and WHY. Compare eligible alternatives transparently, propose where to act, show trade-offs and opportunity costs, and state a prediction with horizon, confidence basis and falsifier. Give every alternative an explicit fate: proposed, wait, test, shrink or stop. Leave selected direction and accepted losses blank for the human.
4. Execution decision: WHAT, HOW and WHO. Link the exact direction revision. State required capability, work output, accountable performer, skill, tool or tool-less method, dependencies, eligible routes and exclusions, resource limits, proof, STOP and recovery. Availability and authority must be evidenced. A draft route is not permission to execute.
5. Human acceptance: leave this field blank: Human decision: __________. The human may accept, correct, hold or reject exact direction and execution revisions. Do not fill acceptance, accepted risk, permissions, priority changes or signatures for me.
6. Signals: separate strategic and implementation scorecards. For each gauge state ID, class, baseline, target, unit, method, source, observation window, owner, harms, exceptions and confounders. Keep unobserved results explicitly unknown. Strategic signals test beneficiary change; implementation signals test effectiveness, cost, quality, latency, reliability and constraint compliance.
7. Review: compare original prediction with observation without rewriting either. Judge decision quality at the time separately from outcome and execution quality; consider luck and confounding. Propose continue, revise, stop or unresolved for human review.
8. Learning return: classify precedent, counterexample or unresolved, bound where it applies, name the learning owner and review trigger. Leave later decision evidence pending until the result actually changes, confirms or rejects a later choice. Trace storage alone does not complete learning.

Return human-readable Markdown and machine-readable JSON carrying the same stable episode ID, record IDs, revisions and source provenance. Include DecisionEpisode.v1, DirectionDecision.v1, ExecutionDecision.v1, DecisionTrace.v1, and later DecisionObservation.v1 and DecisionReview.v1 when evidence exists. Preserve original predictions and input revisions; corrections create new revisions. Record concise reasons, evidence, dissent and authority receipts, never private chain-of-thought.

Final check: a cold reader can recover WHAT NEXT, WHY, WHAT, HOW, WHO, authority, every rejected alternative, gauges and review trigger. STOP on missing authority, critical unknowns, stale acceptance or constraint breach. Recovery is to preserve the unaccepted pack, request the missing evidence or correction, and leave the existing commitment unchanged. Do not contact people, spend, publish, change priorities or execute a proposed route.

Answer WHAT NEXT and WHY. Among eligible alternatives, propose where to act and the beneficiary change expected. Show the same comparison criteria, evidence, uncertainty, and opportunity costs for each option. Use explained ordinal judgments or units; avoid an opaque winner score.

Give every alternative one explicit fate: proposed, wait, test, shrink, or stop, with a reason and review trigger where relevant. The human later selects an eligible option. Preserve dissent and state what evidence would reverse the proposal. Freeze the prediction and horizon before observing results.

Answer WHAT, HOW, and WHO against the exact direction revision:

• WHAT: required capability, bounded output, dependencies, and completion evidence.

• HOW: method, skill, tool or tool-less route, operating limits, checks, STOP, and recovery.

• WHO: accountable human and eligible performer, with evidence of capability and permission.

Evaluate execution alternatives against hard constraints before comparing cost, speed, or quality. Preserve exclusions and every route's fate. If no performer is eligible, hold and name the missing capability or authority. Drafting this route does not dispatch it.

Present the exact direction and execution revisions, expected losses, bounds, and uncertainties. Leave the human-decision field blank:

Human decision: ____________________ Choice: accept | correct | hold | reject Direction ID and revision: ____________________ Execution ID and revision: ____________________ Accepted losses, risk and permitted scope: ____________________ Human owner, recorded time and authority expiry: ____________________ Review trigger and STOP owner: ____________________

Correct means revise and resubmit; hold names the evidence needed;reject closes this proposal while retaining its reasons. Acceptance binds the exact records. Changed inputs or expired authority require renewed review. No consequential action or priority change follows without explicit human acceptance of that scope. Use Authorise Bounded Workbefore delegation and Deliver Promised Workfor an accepted execution route. In a shadow trial, even accepted proposals remain observational and leave the existing commitment unchanged.

Keep two scorecards. Strategic signals test whether the direction creates the intended beneficiary change, whether the causal belief holds, and what opportunity cost or harm occurred. Implementation signals test effectiveness, cost, quality, latency, reliability, and constraint compliance. Fast, cheap execution cannot prove the direction was worthwhile.

Copy one record per gauge into the appropriate scorecard:

Gauge ID and class: strategic | implementation Decision ID and revision: Question, unit, measurement method and sample scope: Baseline and source (or unknown with reason): Target and prediction horizon: Observation and provenance (pending until measured): Measurement interval and received time: Owner, harms, exceptions and confounders: Review trigger:

An unobserved value stays unknown, not zero. Record failed attempts and exceptions as well as successes. Stop at a protected-constraint breach; preserve the evidence, contain only within granted authority, and return to the human owner for recovery or correction.

At the trigger, compare the frozen prediction with observations throughCompare Outcome and Correct Method. Assess decision quality using evidence reasonably available at the time. Assess outcome quality and execution quality separately. Account for luck, external change, measurement error, and other confounders.

Return continue, revise, stop, or unresolved to the human owner. A good outcome does not make an unauthorised decision sound; missing outcomes do not prove failure. Corrections append revisions and preserve the original prediction, decision, authority receipt, action, observation, and dissent.

Classify the result as precedent, counterexample, or unresolved and state its applicability limits. Keep the logic trail described inStrategic Decision Graph. At the next relevant choice, record the earlier evidence reference, later decision revision, and how it changed, confirmed, or rejected the choice. Until that happens, learning return is pending. Storing the trace is insufficient.

Return privacy-safe retrieval failures, corrections, and later-use observations to the recipe owner. Keep private priority evidence and human receipts with their owner; do not copy them into a public example.

Strategic Decision Graph

A Strategic Decision Graph connects context to commitment and later judgment. Each link points to a stable record ID and exact revision, so a reader can recover the purpose, evidence, alternatives, authority, action, and result.

Intention and evidence → direction decision → accepted execution route → action receipt → separate observations → review → later decision

Keep rejected alternatives and policy versions beside the chosen route. A review may support precedent, reveal a counterexample, or remain unresolved. The graph becomes useful only when its evidence informs a later choice; an old approval never supplies authority for a new action.

Reliability

Treat the skill, tool, pattern, and protocol as replaceable parts of the recipe. Accept effectiveness claims only when relevant evidence supports them. A human or agent unfamiliar with the conversation should recover all nine facts in the prompt's final check, identify excluded alternatives, and stop safely.

Compare the pack with a simpler Markdown decision template using the same information, retrieval questions, correction, and later-learning task. If that template achieves equivalent retrieval, correction, and learning at lower lifecycle cost, keep it. A copied prompt or completed pack proves an output; it does not prove beneficiary value.