Decision experiment and evidence record

Decision ID / experiment ID / version / owner / method reviewer:
BUSINESS QUESTION and decision the result will change:
PRIMARY OUTCOME and economic horizon:
BASELINE: value, definition, date and uncertainty:
HYPOTHESIS and commercial mechanism:
TREATMENT: exact version, population, eligibility and delivery:
COMPARATOR: status quo, repaired process or explicit no-treatment condition:
ASSIGNMENT UNIT and method:
ESTIMAND: whose effect over what horizon, usually all assigned eligible units:
LEADING INDICATORS and why they are informative:
ECONOMIC METRIC and Finance cost convention:
ATTRIBUTION LIMITATION: what platform/journey credit cannot establish:
INCREMENTALITY METHOD and identifying assumptions:
FEASIBILITY: useful effect, variance, sample/power or identification limits:
DURATION: enrolment, outcome maturation and planned readout:
INTEGRITY: spillover, missingness, sample ratio, concurrent changes, compliance:
GUARDRAILS and immediate safety stops:
ANALYSIS: tested method/code, multiplicity and interim-review rule:
DECISION RULE: local scale/validate/revise/stop thresholds fixed before exposure:
COST AND CAPACITY: learning cap, analyst/reviewer and delivery time:

Choose randomised individual/account assignment if valid; geographic or other designs if exposure and spillover require them; qualified observational analysis when experimental identification is infeasible. Explain limitations. Do not delay necessary safeguards or obvious repairs to manufacture a control.

What was delivered versus planned, with assignment and outcome completeness:
Primary outcome, effect estimate and interval/range:
Economic result, full costs, cash consequence and guardrails:
Estimate basis: observed / attributed / modelled / forecast (record the applicable basis):
Identification: descriptive association / causal estimate / unresolved; method and assumptions:
Deviations, contamination, alternative explanations and limitations:
Contradictory platform, cohort, experiment, aggregate and Finance evidence:
Reconciliation: entity, metric, denominator, period, spend and assumptions:
Decision: continue / validate / revise / stop; owner and exact action:
Why this beats the available alternative or why the conclusion is unresolved:
Next readout and what evidence would reverse the decision:
Method sign-off / Finance sign-off / accountable decision owner:

Evidence ID / linked decision and Play version:
Type: fact / finding / hypothesis / decision / learning
Claim and exact population, product, treatment, geography and horizon:
Source, source date and reproducible result:
Method, sample, effect, uncertainty and limitations:
Confidence: High / Medium / Low; in which proposition?
Applicability / exclusions / conflicting findings:
Reviewer, review date and effective date:
Expiry or retest trigger; supersedes/contradicts references:
Next decision that used this evidence and resulting change:

30-minute readout: The method owner explains what can be concluded; Finance reconciles costs; the decision owner chooses an action. If evidence is immature, record the next review within the learning cap. Preserve negative and invalid results. An AI draft is never self-promoted to authoritative evidence.