Decision Record v1
Review the verdict, policy version, evidence, hashes, replay handle, and verify link before trusting the action.
Live proof artifact
This is the canonical Decide proof: a revenue-sensitive software action asks whether it should proceed, the Decision API returns a deterministic record, and the handle follows the downstream CRM or billing action.
This artifact is a representative proof run. It shows the integration API contract customers evaluate before production traffic: request, verdict, evidence, storage fields, execution receipt, outcome, intelligence readout, replay behavior, and production handoff.
Use this packet when a buyer, implementer, or security reviewer asks what exactly gets inspected before production access. It keeps the live proof tied to the same Decision Packet v1 evidence, verifier, and operational review path.
Review the verdict, policy version, evidence, hashes, replay handle, and verify link before trusting the action.
Confirm the authorized billing or CRM mutation was attempted and bound back to the original decision.
Attach the observed result so policy effectiveness and anomaly review have real lifecycle evidence.
Export Decision Packet v1, verify hashes and signatures, and forward the proof without requiring console access.
The core product surface is small on purpose: one action boundary, one API call, one decision record your system can store.
Request
{
"question": "Approve 15% annual-plan discount exception?",
"mode": "single",
"context": {
"workflow": "pricing_exception",
"owner_role": "revops",
"margin_floor": "passed",
"owner_rule": "verified",
"account_segment": "growth",
"source_system": "crm"
}
}
Response
{
"decision_record_version": "decision_record_v1",
"verdict": "yes",
"request_id": "req_9f3c",
"decision_id": "dec_43b2",
"policy_id": "pricing_exception",
"policy_version": "v3",
"policy_bundle_hash": "b81c...",
"evidence": ["MARGIN_FLOOR_OK", "OWNER_RULE_VERIFIED"],
"action": "approve_discount",
"input_hash": "8d51...",
"output_hash": "48aa...",
"record_hash": "7f3a...",
"replay_url": "/api/decision/dec_43b2/replay"
}
request_id + decision_idThe strongest proof is not only that Decide can answer. It is that the answer becomes a durable protocol trail: record the authorization, receipt the attempted mutation, report what happened, then score the policy against outcomes.
dec_43b2 approves the discount with evidence, hashes, replay, and verify links.
exec_71af records billing mutation, executor, target object, and before/after state hashes.
out_9a3c reports the successful discount and observed margin after action.
Effectiveness is healthy and anomaly review is stable for pricing_exception:v3.
| Proof artifact | Endpoint | Buyer-readable readout |
|---|---|---|
| Decision Record v1 | POST /api/decide |
Verdict yes, record_hash 7f3a..., receipt_hash c91e..., replay and verify links attached to the action. |
| Action Execution Receipt | POST /api/decision/dec_43b2/execution |
Billing mutation discount.create executed against sub_1042 with state hashes and matching action binding. |
| Outcome Record v1 | POST /api/decision/dec_43b2/outcome |
Outcome succeeded, observed margin_after_discount: 0.182, and outcome hash linked to the original receipt. |
| Policy intelligence | GET /api/decision/policies/pricing_exception/effectiveness |
Effectiveness score 0.94, recommendation healthy, anomaly recommendation stable. |
The proof is not just the first answer. The stored decision_id gives reviewers a way to compare the original record against the declared policy state later.
dec_43b2 against pricing_exception:v3
Replay compares verdict, evidence, and record hash, so review starts from the original policy run instead of chat history or screenshots.
Replay comparison
{
"decision_id": "dec_43b2",
"replay_against": "pricing_exception:v3",
"compare": ["verdict", "evidence", "record_hash", "policy_bundle_hash"],
"replay_status": "matched",
"drift": []
}
A customer can wire this without adopting a new workflow UI. Decide sits at the action boundary.
A discount, payout, access change, policy check, or agent step asks to proceed.
The payload is evaluated against context, owner rule, and guardrails.
The caller stores the full Decision Record v1 object next to the action.
Disputes use the original record instead of reconstructing a thread.
This page is meant to remove buyer ambiguity: it shows what is purchased, where it integrates, and what evidence compounds.
Pricing exceptions affect margin and revenue quality before anyone talks about broad governance.
The integration can start as one POST /api/decide call that records why the system acted.
Each action creates a replayable record of why software acted, which becomes harder to replace over time.
Use this sequence to explain the artifact in under a minute.
Before requesting production access, confirm these items with the workflow owner and implementation owner.
| Question | Required answer | Why it matters |
|---|---|---|
| What action is gated? | One exact action such as approve discount, deny payout, route to owner review. | Prevents broad platform drift. |
| What facts enter the payload? | Owner, source system, guardrails, risk context, and any policy evidence. | Determines whether the verdict is useful. |
| Where is the record stored? | The full Decision Record v1 object attached to the downstream action. | Creates replay and audit value. |
| What happens on missing evidence? | Fail closed, route to owner review, or request more context. | Keeps high-risk automation from guessing. |
Production access should make this proof callable by your system without widening the scope beyond the first action boundary.
Server-side x-api-key calls scoped to one production boundary.
Monthly record tier, quota headers, telemetry, and replay visibility.
Export Decision Packet v1 and verify hashes, signatures, execution receipts, and Outcome Records before forwarding to a buyer or security reviewer.
First live readout confirms stored records, review/no routing, and replay checks.
Only after the first boundary is trusted, apply the same API contract to another gated action.