Record Growth Decision

Record Growth Decision

Create an auditable decision event that can be reviewed against later outcomes without rewriting history.

Inputs

Require:

• decision question, selected option, context, and owner;
• hypothesis;
• valid evidence IDs;
• counterevidence;
• at least one alternative;
• an explicit not_build list and reversibility assessment;
• review date;
• success and stop conditions;
• current causal confidence.

Workflow

1. Restate the decision question and selected option in a form a reviewer can accept or reject.
2. Link supporting evidence by ID.
3. Preserve counterevidence and unresolved unknowns.
4. List alternatives, including doing nothing when relevant, and record what will not be built.
5. Classify the decision as reversible, partially reversible, or hard to reverse.
6. State expected mechanism, success condition, stop condition, and review date.
7. Mark causal confidence as unknown, low, medium, or high.
8. Keep status at draft or awaiting_human_review until a person decides in a later user turn.
9. Append a scoped approvals.jsonl record covering the exact decision ID before setting status to approved.
10. Append the decision event. Never modify an earlier event to make a later result look expected.
11. At review time, compare mature outcomes with the original metric contract and record continue, change, stop, or hold.
12. Link a superseding decision instead of rewriting the original, and classify the new event with event_type — initial_decision, scope_change, success_metric_review, outcome_review, or reversal — so the log states what kind of change it is, not only which record it replaces. Leave event_type out rather than guessing; an unclassified event is honest, a wrong one is not.

Introduction-loop decision example

After reviewing the first-five direct-seeding batch, record one explicit decision instead of assuming that referral is the next action:

• Question: should the team test a draft-only introduction loop for this segment?
• Metric rule: name the loop coefficient numerator, denominator, cohort, and maturity window from define-growth-metrics.
• HOLD condition: name the observed condition that prevents a request, such as no repeat value, recipient-fit evidence, or support capacity.
• Resume condition: name the new evidence, owner, and review date needed to leave HOLD.
• Reward experiment: choose not selected, draft for later review, or a separately approved experiment. A reward is never implied by a referral objective and may not be promised, sent, or funded without scoped human approval.

Record the resulting continue, hold, change, or stop as a later append-only event. Do not retrofit the original first-five decision to make the next loop look pre-approved.

Boundaries

• Let the model identify missing evidence, alternatives, and counterarguments.
• Use deterministic checks for IDs, required fields, status transitions, and append-only history.
• Require a person to approve the decision and accept risk.
• Do not equate shipped, launched, or deployed with successful.
• Do not increase confidence because a statement appears in multiple model summaries of the same source.
• Do not treat an immature cohort, missing query, or unrelated interview quote as outcome evidence.
• Export an hplan intake only as draft_for_gate or ready_for_gate_review; never claim the hplan gate decision.

Outputs

Create or append:

• decisions.jsonl
• approvals.jsonl only after a person approves in a later turn
• hplan-intake.json as a pre-gate intake, never as an hplan gate decision
• decision-summary.md
• review-queue.md
• introduction-loop-decision.md

At review time, once a linked action's metric has matured enough to compare, append to outcomes.jsonl per workflow step 11. It is not a completion requirement for this skill — a decision can be approved and routed onward long before its outcome matures — so it stays out of the bullet list above and the router's completion check, the same way connect-customer-channels stays out of that check for a different reason.

For an introduction or referral objective, additionally create introduction-loop-decision.md. It captures the loop-coefficient definition, HOLD and resume conditions, reward-experiment choice, approval state, owner, and review date. It is a draft decision until a person approves in a later turn.

Read references/output-contract.md before writing them.

Add every new record to decisions.jsonl, approvals.jsonl, and, at review time, outcomes.jsonl, with the append-record command, never by writing or editing the file:

python3 scripts/stg.py append-record <artifact-directory>/decisions.jsonl '<json-object>'

It takes an exclusive lock and hash-chains each line to the one before it, so a later reader can tell whether history was rewritten. In a packaged runtime the same command is signal-to-growth append-record.

Stop conditions

Stop when evidence references are invalid, the owner or review date is absent, alternatives are missing, or approval is implied rather than explicit.

Verification

Keep evidence.jsonl and decisions.jsonl together, then run:

python3 scripts/stg.py validate-artifacts artifacts/
python3 scripts/stg.py export-hplan --artifacts artifacts/

Report draft, approved, executed, and outcome-recorded states separately.