[ibgreen]

Problem and behavior

C8 had no shared independent qualification matrix, and the G7 API/performance roadmap lacked reproducible equal-accuracy baselines. Qualification exposed three boundary failures: rigid transforms could overflow intermediate arithmetic despite finite outputs; radial tangent lifts could reject representable points on tiny ellipsoids; and spherical normal lifts could accept a one-ulp miss or lose accuracy near grazing contact.

This PR fixes those cases, adds 560 shared qualification rows, and establishes informational G7/C8 benchmark baselines without adding dependencies. The bounded qualification domain and intentional native convention differences are documented; this does not claim arbitrary finite-input certification or universal GeographicLib parity.

Evidence

• Independent 180-digit Decimal exact-input frame/pose/tangent controls, analytic boundary regressions, and pinned GeographicLib C++ 2.7 output controls.
• The three analytic regression rows fail on the old implementation and pass with these fixes. Ordinary overlapping controls pass on both revisions before benchmarking.
• Same qualification entry in Node, Chromium, Firefox and WebKit CI. Offline reference regeneration/provenance checks; CI does not build or download GeographicLib/PROJ.
• G7 separates setup/direct/reused lines, frame outputs, poses, velocity and tangent operations. Raw timings, spread, allocation estimates, runtime/compiler/source fingerprints and optional same-host native comparison are retained. No new speed gate or speculative API.
• Historical benchmark bundling preserves explicit package sideEffects metadata; selective retained-source graphs are checked against the pinned baseline.
• Existing accuracy, coverage, allocation and bundle gates remain unchanged. A separate C8 guard gives only the two numerical-fix classes a 2,000-byte minified / 1,000-byte gzip allowance; unrelated imports keep their limits.
• Existing Cesium and GeographicLib SPDX acknowledgements and package licenses retained; new helpers, oracle and native API caller are original MIT code. No new dependency, vendored source or lockfile change.

Validation

Full Node suite: 13,202 passed, 124 skipped. Lint, build, attribution/source-header checks, strict ESM/CJS consumer types, C8 package graphs/budgets and existing geodesic/seam/projection/allocation boundaries passed locally. Shared 560-row qualification passed Node, Chromium and WebKit locally and Firefox in CI. Review fixes cover pinned-baseline fetching in shallow CI, conservative historical package metadata, strict ESM/CommonJS consumer types, and validation of every actual optional-native answer before timing. A deliberately wrong finite native executable is rejected before measurement/report creation. The refreshed benchmark artifact includes all 72 actual-native qualification results. Latest-head CI is the final gate.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e9b0c4c79a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-24T22:33:50.146917Z	e9b0c4c	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The reviewed changes appear safe to merge; both previous findings are addressed.

Summary

This PR adds shared local-frame qualification and informational GeographicLib benchmark baselines.

• Local frames and tangent lifts keep more representable results.
• Local-frame checks now cover a shared, bounded qualification profile.
• G7 workloads now compare math.gl with historical and native baselines.

This review checked the code and stored results; it did not rerun the tests.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Independent and stored native controls] --> B[Check candidate accuracy]
  A --> C[Check overlapping historical accuracy]
  D[Optional native executable] --> E[Check every stored output field]
  B --> F[Run rotated benchmark samples]
  C --> F
  E --> F
  F --> G[Record timings, spread, and optional heap estimates]

Loading

_{Reviews (2) · Last reviewed commit: "Refresh baselines after validating nativ..." · Reviewed by Greptile}

[coveralls]

Coverage Report for CI Build 40140281525

Coverage decreased (-0.03%) to 94.226%

Details

• Coverage decreased (-0.03%) from the base build.
• Patch coverage: 75 of 75 lines across 3 files are fully covered (100%).
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	25893
Covered Lines:	24976
Line Coverage:	96.46%
Relevant Branches:	16002
Covered Branches:	14500
Branch Coverage:	90.61%
Branches in Coverage %:	Yes
Coverage Strength:	37559.4 hits per line

---

💛 - Coveralls