Reference
Policies Reference
Replay policies, fixture provenance, side-effect classes, capture/redaction policy, validity labels, and adapter capability status.
Policies are part of the evidence. Every replay, experiment, and report should name the policies that shaped the output.
{
"model": "recorded-only",
"tool": "pause-for-fixture",
"drift": "continue_to_terminal",
"maxSteps": 100,
"temperatureOverride": 0
}
| Value | Meaning | Current local engine |
|---|
recorded-only | Use captured model outputs. | Supported. |
determinism-probe | Repeat under same policy to detect variance. | Schema-supported; executor-dependent. |
pinned-live | Re-run with a pinned provider/model/config. | Fails without explicit executor. |
compatible-live | Re-run with a compatible live model. | Fails without explicit executor. |
provider-fixture | Use provider-supplied fixture output. | Schema-supported; executor-dependent. |
blocked | Do not allow model execution. | Useful safety state. |
| Value | Meaning | Current local engine |
|---|
recorded-only | Only use recorded tool outputs. | Errors on missing divergent fixtures. |
pause-for-fixture | Pause when a divergent output is missing. | Supported and recommended. |
analyst-fixture | Allow analyst-supplied fixtures. | Modeled through fixture provenance. |
simulated | Allow simulator outputs. | Modeled through fixture provenance. |
live-readonly | Allow live read-only tool execution. | Fails without explicit executor. |
live-explicit | Allow explicit live tool execution. | Fails without explicit executor. |
blocked | Block tool execution. | Errors on missing divergent tool output. |
| Value | Use when |
|---|
stop_on_first_divergence | You only want to isolate the earliest changed event. |
continue_to_terminal | Default for RCA because downstream effects matter. |
continue_until_budget | Future-facing mode for bounded long replays. |
| Value | Meaning |
|---|
recorded | Output copied from captured trace evidence. |
analyst_fixture | Output supplied by a human analyst or approved source. |
ai_fixture | Output simulated by an AI model. |
simulator | Output generated by a declared simulator. |
live_readonly | Output from a live read-only tool. |
live_explicit | Output from a live side-effecting tool with explicit approval. |
| Value | Meaning |
|---|
none | No runtime side effect. |
read | Reads local or external data without mutation. |
write | Writes local state or files. |
external_mutation | Sends, deploys, deletes, purchases, or mutates an external system. |
unknown | Not classified. Treat conservatively. |
Managed runtime helpers infer side effects from tool names as a fallback. Override this for important tools.
{
"defaultAction": "capture",
"rules": [
{ "path": "customerEmail", "action": "mask", "reason": "PII" },
{ "path": "authorization", "action": "drop", "reason": "secret" }
]
}
| Action | Current implementation |
|---|
capture | Capture after default pattern redaction. |
drop | Replace field with null. |
hash | Replace field with [HASHED]. |
mask | Replace field with [REDACTED]. |
encrypt, artifact_only, and metadata_only are not capture policy actions in v0.3.
Default redaction masks emails, phones, API-key-like tokens, JWTs, credit-card-like numbers, and secret-looking object keys.
| Label | Trigger |
|---|
confirmed_by_replay | Metrics exist and no stronger caveat is detected. |
sensitive_to_fixture | Fixture-related metrics are positive. |
model_nondeterministic | Variance-related metrics are positive. |
diverged_but_comparable | Any diff has diverged_but_comparable. |
non_comparable | Any diff has non_comparable. |
unsupported | No evidence supports stronger labels. |
| Status | Meaning |
|---|
native_full | Runtime primitives support native replacement or safe interception. |
managed_replay | Adapter can capture and steer, but runtime limitations remain. |
observability_only | Useful trace capture without deterministic replay. |
unsupported | Public API does not expose the primitive. |
Interception modes include native_replace, proxy_replace, post_result_replace, block_only, and observe_only.