Metrics, Diffs, And Effects Reference

Replay comparison outputs, metrics, effect statuses, effect types, and validity labels.

Diff Kinds

output
trace
tool
state
timeline
metric
causal

Each diff has projectId, replayId, kind, comparability, patch, optional summary, and metadata.

causal is a labeled comparison record, not proof that downstream runtime behavior was recomputed. Treat it according to the replay mode and executor support used for that run.

Comparability

exact
aligned
diverged_but_comparable
non_comparable

Metrics

Metric	Meaning
`first_divergence_step`	First changed event index; `-1` means none.
`tool_sequence_distance`	Normalized distance between tool event sequences.
`fixture_dependency_count`	Number of fixture events used.
`tool_argument_changed`	`1` when tool args or arg refs changed.

Metric provenance values:

deterministic
analyst_annotation
simulator_derived
provider_signal
model_judged

Effect Statuses

supported
not_supported
inconclusive
invalid

Effect Types

tool_args_changed
output_changed
fixture_sensitive
early_divergence
non_comparable
no_effect

Validity Labels

confirmed_by_replay
sensitive_to_fixture
model_nondeterministic
diverged_but_comparable
non_comparable
unsupported