[devin-ai-integration]

What and why

The release-gate performance record for 0.12.0: docs/project/performance-release-0.12.0-vs-0.11.0.md, following the method of the 0.11.0 record (every benchmark package, benchstat over six interleaved rounds, whole-binary timings on generated and real models, process floor, the tests/stressmodel constellation), wired into mkdocs.yml (not-in-nav, after the 0.11.0 record), docs/project/README.md, the "latest is" link of docs/internals/performance.md, and a changes/unreleased/perf-0.12.0-record.added.md fragment.

The record measures three columns, because the toolchain moved from go1.25.11 to go1.27.3 between the releases and dominates a naive comparison: 0.11.0 as released, the 0.11.0 source rebuilt with go1.27.3, and the candidate — first develop at 1c6bb1972 (which holds #1699), the pre-fix measurement that found the regressions below, then the release candidate itself, cc1d477ba (develop ab88ea950 after the seven fixes + #1730), measured with the full method in "The release candidate" with the pre-fix column kept so the recovery is visible row by row. Every statement says which columns it rests on; source regressions are column 2 against 3 (and 2 against the release candidate).

What it finds, in short: the candidate is faster on nearly every row a user notices (loads −8…−13 %, WriteAPIJSON −10…−37 %, action chains −35…−59 %, LookupQualified 150 → 43 ns), and allocates more on a handful of paths, each bisected to its merge — #1462 (LoadModel/0), #1478 (every load row, +5 % bytes), #1514/#1461 (AssignmentLoopStep), #1619 (DerivedReadWriteRead), #1508/#1512 (the binding walks), #1626 (FeaturesOf, Instantiate, BatchSatisfy) — and the +31 MiB resident set on the largest generated model is #1441 holding a submission's parse trees through its analysis. The fix for each is a separate perf/<slug> PR; the record's Fixes section is filled in as they open, and its Re-runs section with the quiet interleaved re-run of the rows the loaded machine left in doubt.

The fixes (#1741–#1747) are listed as in the candidate with their merge SHAs; the export rows were re-probed on 5e0322e0c when #1760 landed on internal/translate/export; "After the measured heads" accounts for every merge up to develop b4d0a35c4 — by git diff --stat … -- '*.go' go.mod go.sum for the client/pom/workflow merges and by #1771's own six-round benchstat for the one runtime refactor. The record also says what it does not decide: the afternoon's quiet re-run ran under 3–8 % hypervisor steal, so the first pass's time verdicts stand, and one open finding (a -run-query run leaves the next -validate of the same binary with a cold library record cache) is named and not attributed.

How it was verified

python3 scripts/check-doc-links.py reports one broken link, docs/project/third-party-notices.md → ../assets/landing/libavoid-js.LICENSE.txt, which is pre-existing on develop (same result on a clean checkout of 54e590985) and untouched here; the new record's links all resolve. The raw per-round benchmark outputs (six rounds × three configurations, the release-candidate rounds and re-probes, whole-binary runs, bisect logs) are kept outside the repository and attached to the gate report, as the 0.11.0 record's were.

Checklist

• make test and make lint pass locally — documentation only; no Go source changes
• Tests added or updated for the change — not applicable
• Documentation extended where it already covers the surface (see CONTRIBUTING.md)
• Changelog entry added as changes/unreleased/<slug>.<section>.md, not as an edit to CHANGELOG.md
• baselines regenerated and make docs-counts run if a gate count moved — no gate count moved
• No internal work-item labels (waves, slices, F4, K5) in the body, docs, or changelog

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/15408827f8104c7fbbdbd1b2c852ed2c
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/15408827f8104c7fbbdbd1b2c852ed2c?variant=devin
Requested by: @HuiJun

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

🔍 Devin Review: 3 flags

Not posted on this PR by your GitHub settings — view them in Devin Review. (Configure)