Skip to content

Evidence map

Generated evidence boundary. This page is generated from the checked-in selection, review, source-lock, promotion, and cohort manifests. It does not run Bifrost and it does not copy provisional scores.

The shortest honest answer is currently the published real-project-v1 result: its historical v0.2.0 page reports the measured Bifrost/reference comparison. Published result status is derived from the immutable bundles supplied to this generation run; no score is copied into this evidence map.

SliceFrozen identitySelection and review tierFrozen denominatorResult publication
Prospective v1real-project-v1 · v0.2.0prospective_pre_registered · human_adjudicated_agent_panel36 cases (12 per profile)Historical release; current v1 result
Prospective v2real-project-v2 · v0.3.4prospective_pre_registered · human_adjudicated_agent_panel36 cases (12 per profile)Immutable release v0.3.4; generated report provenance is recorded below
Reviewed legacy corelegacy-promotion-v1-balanced-core · v0.3.1retrospectively_selected · legacy_promoted110 cases (10 × 11 languages)Immutable release v0.3.1; generated report provenance is recorded below

The denominators above are not interchangeable. In particular, v1 and v2 are prospective source-only selections, while the legacy core is a separately frozen retrospective promotion of analyzer-informed development cases. A later report may present a documented stratified aggregate, but it may not flatten these trust tiers into one accuracy score.

Each profile remains visible before any aggregate. The numbers here are selected-case counts, not analyzer outcomes.

SliceLanguageCandidate/reference profileRepositoriesCases
v1gogopls412
v1pythonpyright412
v1typescripttypescript-language-server412
v2javaeclipse-jdtls412
v2rustrust-analyzer412
v2cppapple-clangd-21412

The immutable promotion is legacy-promotion-v1-balanced-core. Its balanced core is 110 cases, with 42 overflow candidates and 6 controls kept outside the correctness denominator. The source-only legacy inventory contains 158 cases.

LanguageBalanced-core cases
cpp10
csharp10
go10
java10
javascript10
php10
python10
ruby10
rust10
scala10
typescript10

The entries below are derived only from checksum-verified release bundles. The generated result pages remain the score authority; this index records their exact release, snapshot, manifest, and report identities without retyping score totals.

SliceReleaseRevisionFreeze manifest SHA-256Report artifacts
v1v0.2.06ea6056fa6b3eb52a656a2b4a62c57956771de783258e3269c98baa980883e969df97e1c9f4920503072ca8797e2600e2b268614bifrost 702a73c97d86a4c9c6b2b576c3d411306788581e4659b1ff9e82fae491c7459b, gopls 679796a7a8f3d5b5eb44cad9a9ef06a2b512f4d53fdd503414842a4024026b94, pyright c6899baca9154f464c7baefe1984d0078175813138d6c428234f3573d0665099, typescript-language-server 8dda75467e0bd2498905abb8e8e6702c185e9af5954c5513a55248ecda9a2ad4
v2v0.3.4e9ffdf26a2a216c4c1fcc0dc810b0b498e726abec6eece246a748d210b31d646ca3f19dfe74c8d8fdffe11e43de6ddce081a3a0aapple-clangd-21 2e8167d21745617e6d507875bb190206d30954d5ee5f2ad873db4825ff514c3f, bifrost 0341a8f8aee26d5f8ccd88b48f1cd665efc2a49ed914418346ccd7227c62a03e, eclipse-jdtls b0b41090ff1d6f1f4393761148c181800926bb09eb213ca91e2d3bc17a45b2ca, rust-analyzer b039a8be300568355668808de8a8161ae1e2735a770395aeb47d8e55753be2a4
legacyv0.3.18393dafe6b9af644c33cb8268b8c265810498dd46c64323d6c3289ab7a3934f288f3f75c765379b34d3268ad991ca1398a35d980apple-clangd-21 fe15d95580e5837f540070ca7f95a61e4baa79eb4c8c12a60b40a3e10995239b, bifrost 593cc3580d5a8e97ca6d3852629c181f0d362eff1f9ae17e3d14c5a43e99cd7a, eclipse-jdtls c6367a23bc6d1d65fe7242505b491fc8054beb2fec111c86abae290cbd8a32f4, gopls f3abe0cd3abf4d82c78941c5e8572cc28a28599e9be44ed1e8c39868a17f96fa, intelephense 29669d93db3f097ad3d12739636dd272b1599354cce07393b4415a92727a57df, metals ea522be8abfc92fdd9bfe2b91047b8e7a94f209c95f6e54e11f0546c9b1fedeb, pyright 33f89e7f54252bcd3aed32a47f203b3bd03b3cca88e755121254340270915fd7, roslyn 16b7815b9cd482270470e0cbc218e740eae1512b76ef14a1ce9e72302325d132, ruby-lsp 027c6761459f6032d0ac93fdfe92d64e924d1895671b798e50999c4570f2e8eb, rust-analyzer 1c0ba625f2a3ae8d5a9332d6d866e72a4b53188d3d166dfacd83df4a6142d293, typescript-language-server 2919729c212746b9c301dc29551285e1cc9f6147707c6f79c90d7996a3693acb

The checked-in development corpus contains 158 cases. The reviewed legacy core accounts for 110; the 48 cases outside that core comprise 42 frozen overflow candidates and 6 unsupported/not-planned controls. This remainder is retained for regression and diagnosis; it is not silently added to v1, v2, or the legacy denominator.

  • Prospective v1 and v2 keep separate profile/language denominators; v2 permits only documented stratified aggregation.
  • The legacy manifest records retrospectively_selected, legacy_promoted, source_only, and analyzerOutcomeUse: forbidden; re-review cannot make the source contract preregistered.
  • Controls and overflow remain explicit partitions and cannot enter the balanced-core score.
  • Score tables must be generated from a checksum-verified immutable report artifact bound to the matching manifest. This page records published provenance when such a bundle is supplied and otherwise reports readiness, never guessed scores.

Manifest provenance is machine-readable in docs/src/data/evidence.json and is checked in CI with scripts/generate-docs-evidence.py --check. See the current v1 result, the historical development result, and the human ground-truth audit for retained historical evidence.