01 · Discovery
Name the decision
Choose work your team already does repeatedly: consequential enough to matter, bounded enough to examine with real facts, rules, and accountable people.
Name the people and the authority
| Responsibility | Named role or team |
|---|---|
| Supplies or owns source facts | |
| Owns the rules or policy | |
| Evaluates alternatives | |
| Approves the decision | |
| Carries out the action | |
| Reviews, audits, or challenges it |
02 · Current state
Reveal the operating model
The decision already has a model, even if it is distributed across spreadsheets, applications, messages, policy documents, and people’s judgment. Make that model visible.
Source facts and revisions
| Fact | Source and owner | Time or revision | How corrections arrive | Reported, observed, or assumed? |
|---|---|---|---|---|
Ask: Which fact changes most often? Which is hardest to reconcile? Can the team identify exactly which version was used?
Rules and constraints
| Rule or constraint | Type | Where it lives today | Owner / version | Exception or expiry |
|---|---|---|---|---|
Types: hard constraint · policy rule · preference · accountable judgment
Private copies and interfaces
| Copy or interface | What it repeats or transforms | Who uses it | How it stays synchronized | Where disagreement appears |
|---|---|---|---|---|
| Spreadsheet, app, script, dashboard, email, AI tool, or human process | ||||
Trace one dependency chain
- Changed fact
- Affected calculation
- Rule or constraint
- Decision or alternative
- View or action
If the first box changes, can the team explain why the final box moved without reconstructing the process by hand?
03 · Changed fact
Inspect what follows
Use a real historical revision or a clearly labeled hypothetical scenario. Distinguish what is calculated from what is assumed, recommended, or authorized.
Trace the consequence
- Revision received
- Dependent results move
- Constraints are tested
- Alternatives change
- People or views are affected
Compare authored alternatives
| Alternative | Finding | Assumptions | Tradeoffs | External action required | Decision owner |
|---|---|---|---|---|---|
| Feasible / conditional / infeasible / unknown | |||||
“Unknown” is valid when facts, rules, or authority are incomplete. Do not turn missing evidence into false certainty.
Build the explanation record
Keep the claims separate
- Source factReported or observed by a named source
- AssumptionChosen for analysis or a scenario
- Authored ruleApproved logic used in the evaluation
- Calculated resultOutput from identified facts and rules
- RecommendationOption proposed for consideration
- Authorized decisionChoice made by recognized authority
- Observed outcomeWhat was measured after action
04 · Evaluation
Design a bounded evaluation
Evaluate one recurring decision under controlled conditions. Compare observed evidence with a named current-state baseline before proposing broader transformation.
Bound the work
| Include | Explicitly exclude or defer |
|---|---|
| One recurring decision | |
| Named users and decision owner | |
| Source facts, revisions, rules, and constraints | |
| Required views or outputs | |
| Historical, synthetic, shadow, or approved live mode | |
| Fallback process and stop condition |
Choose three to five measures
| Measure | Current baseline | Observation method | Evaluation result | Evidence owner |
|---|---|---|---|---|
| Time from source change to updated decision | ||||
| Manual reconciliation effort | ||||
| Agreement across role-specific views | ||||
| Ability to reconstruct why a result moved | ||||
| Correct application of agreed test cases |
Record the baseline first. Report results with method, period, and limitations. A target is not a measured outcome.
Safeguards
Readiness review
Bring this completed worksheet—and one real revision—to a Grid discussion. The goal is to determine whether one recurring decision can be represented, tested, explained, and governed as a shared operating model.