Verdict: No actionable materially smaller equivalent Layer A design found. The +4,467-byte installed/unpacked growth crosses the 3 kB independent-review threshold, but the delivered inert owner prerequisites explain the generated growth. This review makes no claim that Layer B's refusal is active.
- Exact base
c8b944ca0116d577941fb47ba82c84504f9b3075; exact PR headf4b869ab2b1855f17540f75b32c08a13cffae508. - Passing size job: run
39685268166, job119517604558, artifactsize-reportID12386036554(uploaded ZIP SHA2566e1639ede0397d74be7fc3dff3c6adcacb3bc5a5d05b0013e399d275aa0e632b). Raw CI reports aresize-report-base.jsonandsize-report.jsonin this directory;accounting.jsonhas their hashes and complete file diff. - The Actions checkout began at merge commit
1b05db863, but the workflow explicitly detached to the base SHA for its base report and to the PR head SHA for its current report. These are base/head measurements, not merge-tree measurements. The report JSON lacks embedded commit IDs; the workflow, job log excerpt (job-evidence.txt), run metadata and copied artifact establish the attribution.
| Metric | Base bytes | Head bytes | Delta bytes |
|---|---|---|---|
| Installed including dependencies | 5,522,475 | 5,526,942 | +4,467 |
| Package unpacked | 5,521,746 | 5,526,213 | +4,467 |
| Download tarball | 1,674,653 | 1,675,711 | +1,058 |
Generated dist/src JavaScript |
4,295,907 | 4,300,374 | +4,467 |
| Sum of per-file gzip sizes | 1,460,815 | 1,461,710 | +895 |
The package has 649 files on both sides and dist/src has 487 JavaScript files on both sides. Exactly 16 existing dist/src/*.js pack entries changed; their signed deltas sum to +4,467 bytes. No pack entries, dependencies, lockfile changes, tests, docs or assets account for the growth. Installed overhead outside the package is 729 bytes on both sides. Package manifest edits only change internal exports. The source diff is 54 files, +1,239/-254 (1,493 gross), including production +549/-244 (305 net), tests +680/-6, fixtures +6, metadata +4/-4.
| Generated file | Byte delta |
|---|---|
dist/src/sdk-batch.js |
+1,450 |
dist/src/src.js |
+793 |
dist/src/device-claim-recovery.js |
+621 |
dist/src/device-claims.js |
+560 |
dist/src/session2.js |
+255 |
| Remaining 11 changed JS files, net | +788 |
The 16-file exact ledger is in accounting.json. Generated chunk names do not provide source-level byte attribution. Source inspection ties the leading changes to explicit registry resource traits and public standalone projection, read-only claim settlement and durable recovery inspection, the request-local device-selection seam, and lease expiry/hold inspection. The registry policy helper mostly moves existing behavior from registry.ts; inspectExistingClaimSettlement serves read-only inspection and normal acquisition, while the lease predicate serves preview and consumption. The Android teardown inspection remains behind a dynamic import. No new eager dependency edge or duplicated owner implementation is evident from the source diff. Removing explicit resource traits, recovery/lease preview, or test cases would drop required A/B parity rather than provide an equivalent smaller design. The generated growth is modest relative to those distinct owner contracts; byte-level causation inside a minified chunk cannot be established from this artifact alone.
Seven same-runner samples per command: --version base median 36.645 ms (35.757–39.340), head median 38.512 ms (36.124–40.227); --help base median 73.559 ms (71.299–74.412), head median 74.356 ms (71.592–75.802). These overlapping short samples do not establish a causal startup regression. The exact samples are in the raw report JSON.
Finding: None. This is a measured-growth assessment, not a replacement for the separate full source reviews or future B review. No source, test, build, device, PR comment, or CI state was changed here.