You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Workspai already reuses unchanged project/provider output. The remaining risk
is correctness: deletes, renames, shared contracts, or provider-version changes can
leave stale entities or proofs unless incremental output is continuously checked
against a clean build.
PCC re-observation coverage for deletion, Graph overlay aliases, and artifact
resurrection detection.
Remaining work
Turn equivalence into a release invariant using the shared provider corpus and randomized
mutation sequences. Compare semantic content while excluding only declared
run-local timestamps/metadata.
Acceptance criteria
Add, edit, delete, rename, and repeated dirty edits are covered.
Cross-file, shared-schema, and cross-project invalidation are covered.
Incremental output equals a clean full build for entities, relations, proofs, and omissions.
Provider/schema/normalizer version changes invalidate the correct scope.
Incomplete Git evidence reports a fallback instead of false reuse.
Cancellation preserves the previous valid graph atomically.
JSON explains reused, rebuilt, and full-fallback scopes.
Randomized mutation runs pass on Linux, macOS, and Windows.
Completion evidence
Publish semantic diff output for the fixed corpus, randomized seed/reproduction
commands, and a measured one-file change on a 10,000-file fixture. Correctness is
the blocker even if a full build is temporarily faster.
Alternatives considered
The following alternatives or adjacent concerns are intentionally outside this issue:
redesigning the existing cache protocol;
performance claims before correctness passes;
distributed graph storage.
Expected impact
Large workspaces can reuse unchanged graph evidence without risking stale
entities, relations, or proofs after source and provider changes.
Problem statement
Workspai already reuses unchanged project/provider output. The remaining risk
is correctness: deletes, renames, shared contracts, or provider-version changes can
leave stale entities or proofs unless incremental output is continuously checked
against a clean build.
Proposed solution
Current foundation
incremental-project-cachegraph provider protocol;resurrection detection.
Remaining work
Turn equivalence into a release invariant using the shared provider corpus and randomized
mutation sequences. Compare semantic content while excluding only declared
run-local timestamps/metadata.
Acceptance criteria
Completion evidence
Publish semantic diff output for the fixed corpus, randomized seed/reproduction
commands, and a measured one-file change on a 10,000-file fixture. Correctness is
the blocker even if a full build is temporarily faster.
Alternatives considered
The following alternatives or adjacent concerns are intentionally outside this issue:
Expected impact
Large workspaces can reuse unchanged graph evidence without risking stale
entities, relations, or proofs after source and provider changes.