Execution Receipts

10-Second overview

Execution Receipts are per-step artifacts for non-terminal Flow workflow steps. Each receipt records what executed and how it resolved. Context Layer stores receipts for retrieval.

Why Execution Receipts exist

Authority Reports summarize completed Flow workflows at session level and do not include per-step detail. Execution Receipts provide that per-step detail.

Receipt contents

Each Execution Receipt contains:

execution_id — Unique identifier for the invocation.
tier — Enforcement Tier decision (TIER_0, TIER_1, TIER_2).
trust_contract_status — Overall Trust Contract evaluation result (PASS, PARTIAL, FAIL).
output_revised — Whether the model output was revised during this step.
constraint_compliance — Per-constraint compliance results.
instruction_coverage — Per-instruction coverage results.
verification_dimensions — Execution counts including conflict_count, integrity_count, and violation_count.
model — Provider name, model key, latency in milliseconds, retry count.

Storage

The Context Layer produces an Execution Receipt for each step and stores it persistently. Receipts are retrievable per project and per session/step.

Plan access

Execution Receipts are not available on the Free plan. The Production plan includes Execution Receipts per step. The Scale plan includes Execution Receipts per step and access to the Request Debugger.

Accessing receipts

Each receipt row expands to show full step detail. Each receipt row provides an Open Debugger button. The button opens the Request Debugger for that step.

The Request Debugger is available on the Scale plan.

Receipts and Authority Reports

Authority Reports summarize the full workflow session. They include total calls, tier counts, enforcement events, and session lifecycle.

Execution Receipts provide per-step detail. They include what executed and how it resolved, what Enforcement Tier was assigned, whether the model output was revised, and what the model produced.

Non-claims

Execution Receipts record execution checks and outcomes, but they do not guarantee model correctness, semantic accuracy, or factual accuracy.