Worksheet: choices, evidence and the next decision

H.A.R.D. Protocol 0.3 · Public Preview

Use paper or existing notes. The facilitator can maintain the detailed machine record. Completing this worksheet does not authenticate evidence or grant a favorable result.

Set the boundary

Artifact / revision / date: ______ Purpose / affected people: ______

Route: artifact_review / engineering_commitment / independent_evaluation

Stage: specification_handoff / prototype_handoff / implementation_review / runtime_release_review

Artifact population: ai_generated / runtime_ai / both / neither / unknown

Current scope: ______ Deferred or excluded items and reasons: ______

Reviewer kind: ______ Role overlaps and reference access: ______

Criteria fixed when, before or after viewing the artifact? ______ Owner of next decision: ______

Inspect one consequential choice

Question	Record
What is actually chosen, and where?	Choice ID, observed behavior and source location: ______
What purpose and criteria should it meet?	Outcome, affected people, priorities and acceptance conditions: ______
What reason was recorded at the time?	Documented / reported / unknown, source: ______
What alternatives were recorded or are newly proposed?	Historical alternatives with sources: ______. New alternatives: ______
What are the effects and tradeoffs?	Benefits, costs, assumptions, conflicting priorities and limits: ______
What evidence supports or questions the choice?	Sources and validation result: ______
What does a person decide now?	Current rationale, labeled new when appropriate: ______. Accepted / revise / pending: ______
What happens next?	Action or revisit condition: ______. Owner: ______

Repeat only for consequential choices. No need to replace an appropriate original choice. An unrecorded reason stays unrecorded; a plausible explanation is new analysis.

Record requirement and recovery evidence

Item and acceptance check	Scope and reason	Specified	Walkthrough	Implemented	Runtime tested
______	______	______	______	______	______

For each level, record pass, fail, unassessed or justified not_applicable plus a source location. Do not collapse the columns. Current unassessed items remain in the denominator. A future-scope exclusion cannot conceal a dependency necessary for the current claim.

Finding / expected vs observed / location: ______ Current applicability: ______

Proposed severity and basis: ______ Accepted, unsupported, duplicate or unresolved: ______

Confirmed severity, if actually reviewed: ______ Remediation and verification: ______

Complete the minimum artifact review

Declared-stage result: ______ Evidence that supports it: ______

Choices retained: ______ Open decisions: ______ Findings requiring repair: ______

Next action / owner / revisit trigger: ______

Reviewer metrics: N/A unless eligible under independent evaluation. Reason for each N/A: ______

Preparation / session effort, if recorded: ______ Do not invent retrospectively measured times.

This route does not require a measured baseline or ROI estimate. It does not issue an engineering recommendation or deployment approval.

If funding engineering, complete all six gates

Gate	Additional evidence
G1: What happens today?	Observed current case, connected steps, time, touches, failures, review, escalation, rework, volume and observation window: ______
G2: What needs to improve, for whom?	Actual need, alternatives, requirements, affected roles and impact checks: ______
G3: What happens on failure?	Connected states and recovery, dependencies, authority boundaries, data preservation and owner: ______
G4: What demonstrates behavior?	Requirement, exact artifact and requirement/context fingerprints, executed validation and result: ______
G5: What work remains for people?	Disjoint review, correction, escalation, rework and residual manual minutes; cost assumptions: ______
G6: What bounded step is justified?	Risk depth, human evidence-quality review, findings, owner, resource cap and next review trigger: ______

Apply the full risk rules before selecting QUICK6 or FULL. QUICK6 stops at the first failed or missing gate; later gates are NOT_EVALUATED, and known critical findings remain visible. Preserve a stopped record before a new linked FULL run. Missing baseline blocks a qualifying engineering recommendation and leaves ROI indeterminate.

Engineering recommendation / reason / next action: ______ Separate owner authorization: ______

Ask: In your own words, what does this result allow, and what remains undecided? Record the answer before explaining again.

Permission / anonymization / retention: ______ Previous run / changes: ______

Execution versions: protocol 0.3-preview.1 · MCP 0.3.0rc1 · Skill/contract 0.3.0-rc.1.

Engineering reasoning deepening

Complete this section for each consequential choice only after recording whether deepening applies.

Choice ID: ____________________

Does engineering deepening apply? yes / no

Why? ________________________________________________________________

Trigger(s): state mutation / external side effect / irreversible action / privileged or tenant boundary / unreliable dependency / repeat or concurrent execution / money or data loss / material scale or cost assumption / current promise depends on deferred work / ambiguous source of truth / other

Record each consequential assumption separately before the surface table. Leave unsupported claims unassessed.

Assumption ID	Statement	Status	Consequence if false	Evidence needed / retained evidence	Revisit trigger

For each relevant system surface, record the current model and status.

Surface	Current model or question	Status	Evidence or next evidence
Truth
Ownership
State
Boundary
Contract
Failure/recovery
Time/ordering

What would have to be true for this choice to be wrong? ____________________

Expected behavior or invariant: ________________________________________

Consequence if mishandled: ____________________________________________

Evidence that would resolve the challenge: _____________________________

Evidence level when assessed: walkthrough / implemented / runtime-tested / unassessed

Smallest coherent next slice: _________________________________________

Do not fill blanks with a generated rationale. Unknown is a valid review result.