Source access: Linked repository evidence may require project access. This report distinguishes merged code, historical verification and product release acceptance.
11PRs merged into hyper-saas in October Metadata-confirmed, not an ecosystem total
0.1.75Extension version/changelog cut Both-store publication not established
Still openCanvas #823 and Inspector #831 Candidate evidence, not shipped features
What actually reached main
These eleven PRs have October merged_at timestamps. Merge metadata takes precedence over stale “not merged” wording in PR descriptions. Counts are limited to this repository and window; no file, line or commit totals are inferred by adding overlapping PR diffs.
Author accounts in this verified set: alex-mextner on eight PRs and smoothbricker on three. Account names are not silently mapped to people or used as a complete contribution census.
Product candidates: useful progress, still open
OPEN · NOT MERGED
Canvas: keep selection and manual edits intact
#823 at a4caec76 separates read-only selection observation from mutation authority, keeps newer selection from being overwritten by an earlier move, and preserves manual text around move/save. AI recovery remains read-only and starts only after an explicit Fix with AI click.
The source-bound result manifest records byte-exact restore, selection retention and zero AI requests before the click versus one afterward. The PR records 188 focused tests, 13/13 native assertion sets and ten consecutive typing → move/write → typing → save scenarios. CI run 38942173036 independently reports success for this head. These are historical proofs, not a fresh VS Code run in this reporting pass.
Remaining: integrate the stacked branch, reassess its current release readiness, and complete real host/SaaS, Undo/Redo and overall product acceptance. The linked typing/selection screenshot is an existing artifact; it was not visually re-inspected for this report.
OPEN · NOT MERGED
Inspector: localize subscriptions, count the complete cost
#831 moves 36 style values from root-panel state to section-local subscriptions. Its 17 October source-bound report measures the same controlled Inspector workflow at source e8dd27e4.
Metric
Parent
Field-local candidate
Root-panel calls per 1,000 edits
1,013
13
Whole-layout calls per 1,000 edits
1,001
2
Main-thread work per 1,000 edits, CPU ×1
1,314.97 ms
799.52 ms
Application + added library code
3,282 lines
3,191 lines
Actual right-webview gzip
373,130 B
375,534 B
Interpretation: the tested stress workflow does less broad UI work, but total code savings are small and the bundle grows. An equally localized native control is comparable; the equal-eight-write comparison does not establish a CPU advantage. This is not measured physical typing latency, developer productivity or live AST/HMR acceptance.
OPEN · NOT MERGED
Live examples: actual completion controls and keyboard continuity
#822 adds real issue-completion checkboxes, separate selection, focus restoration and duplicate-action protection. The PR describes Chromium/WebKit coverage, but its final full gate and deployed-lab acceptance are not established here. The latest metadata shows it still open at 20ef733c, now targeting main after #809 merged; older stacked-branch wording is not used as current status.
State-lib: a bounded application result
#832 integrates complete-entry specialization and work elimination. Its research report reports completion-handler medians at 1,200 visible Boards rows of 63.2 → 1.0 ms in Chromium and 93 → 1 ms in WebKit. The comparison uses pinned application artifacts and a synthetic API. It does not cover the entire edit/move/Undo/Redo workflow or production traffic.
The cost is retained: the complete state-lib entry is 42,166 gzip bytes versus 37,132 for the equally optimized baseline; small fixtures retain extra heap. Handler and next-frame timings are not field INP or completed paint. The historical measurements were not rerun after the 30 October replay. Its replay note lists focused compiler/Node/site checks and explicitly says eight Wasm contracts were not run.
Engineering approach: preserve one Runtime; make operation ownership explicit; compare complete application entries and retain adverse results. Upstream library integration does not close the separate application-adoption and staging tasks.
Verification and release ledger
Evidence is separated into repository state, historical test execution and delivery acceptance. No product test suite or benchmark was newly executed to prepare this report.
HISTORICAL HOSTED PASS
#810: integrated toolkit evidence
CI and Security Scan passed. The actual Tests job log was read in the prior evidence pass: 7,834 app passed, 7 skipped, 0 failed; 1,040 native passed; all 79 site gates passed.
Those counts belong to tested merge-ref 1d526fb9 (head d4cca87 plus base 81d0b05), not all later main revisions. The PR records explicit owner acceptance for its bounded reviewed tree; it is not blanket acceptance of the month’s work.
HOSTED FAILURES VISIBLE
#837 and late-October #809
The refresh confirms CI run 41059658812 and six other returned workflows failed at #837 head 539fe4c. CI run 41236201667 and six other returned workflows also failed at #809 head ecd65533.
#837 describes a local verification fallback. The precise hosted failure cause was not independently established from job annotations here. A merged branch or a PR’s local verification narrative is not labeled as passing hosted CI.
RELEASE CUT ≠ STORE PROOF
0.1.75: matrix and publication limits
#825 is merged. Its reported E2E matrix is 1,815 passed, 26 failed, 33 flaky, 1,228 skipped; this is not a fully green matrix. The PR attributes failures to pre-existing issues or flakes, and documents false-positive multi-select checks in the older version.
Registry APIs/pages in the prior check did not yield usable proof of both stores’ actual 0.1.75 versions. Publish run, release SHA and both registry versions remain to be reconciled.
SEPARATE ADMISSION WORK
Canvas, Inspector and application rollout
Canvas #823 and Inspector #831 are open, even where saved tests or CI are green. Companion harness #169 also remained an independent open PR in the previous evidence snapshot.
Fresh tree-specific readiness checks and human acceptance remain required before a new production release of the product. The report does not establish that those gates have been completed.
Next steps
Establish the delivery baseline. Match 0.1.75’s release SHA, publish workflow and actual versions in both stores. Diagnose late-October hosted failures and tie any accepted local fallback to the exact final tree.
Finish Canvas acceptance. Integrate and recheck the candidate, then demonstrate typing/selection/Undo/Redo and shared SaaS behavior against the intended release tree. Present evidence for owner acceptance before deployment.
Complete state-lib admission. Resolve remaining whole-entry, adoption and staging criteria; repeat source-bound measurements after integration. Retain the native comparison and the bundle/heap regressions rather than claiming universal superiority.
Close lab and E2E gaps. Verify #822’s final source and deployed-lab behavior, then address multi-select false positives and the separate ownership-safe E2E harness.
Coverage and boundaries
Core evidence: GitHub PR metadata, selected source-bound reports/manifests and historical CI results from hyperide/hyper-saas. The 1 November refresh confirmed all eleven October merges and the highlighted candidate states. An October-updated-PR search for alex-mextner/hyper-canvas-draft returned no matches; this is not a claim of no commits or no work there.
No unrelated bot projects, rig results or conloca results are included. No total LOC, file count, review count, productivity percentage or universal completion claim is invented. Evidence links may require repository access. The findings do not establish a new product deployment or blanket product acceptance.
