@@ -2341,6 +2341,69 @@ This general global-binding optimization materially improves two unrelated
23412341external sources while preserving the internal regression guard; B4 and B5
23422342remain active.
23432343
2344+ The exact-SHA hosted workflows for
2345+ ` a8726fb030e34ffd72d38240a8c6d5b017cbfe41 ` all completed successfully. CI
2346+ passed the comparison, Test262 subset, and full check jobs. Exact Test262
2347+ coverage remained 42,671 pass, one fail, zero timeout, and one actionable gap.
2348+ The hosted broad preview retained 25/25 cases, 3/3 valid blocks, and passing
2349+ linearity, but its 1.0179x candidate/base result was classified inconclusive:
2350+ several unrelated cases moved together on the variable hosted runner. The
2351+ candidate/QuickJS-NG diagnostic was 0.3574x. More importantly for the campaign
2352+ boundary, the hosted external preview still failed B5 by a wide margin at
2353+ 12.608x QuickJS-NG for 5/5 JetStream cases, 7.736x for 7/14 Kraken cases, and
2354+ 11.684x for 23/26 SunSpider cases. Hosted broad raw/report SHA-256 are
2355+ ` e40fbd919b692b2ba31f0b8c08588a65a3258646e19535f9e58550c8c2039514 `
2356+ and ` f4b208db24a1b567506e9a7571973b4a6f1308966276128205a538fe4ce4f113 ` ;
2357+ hosted external raw/report SHA-256 are
2358+ ` 68081a1cf2f8a0b841241d5740db57b3059fa2b54697f21d9f9d87584e631e8b `
2359+ and ` 94414e047fd7fb53be81b35dd98949ef3ee348957e8396f1e936f308fbccc284 ` .
2360+
2361+ The forty-third v2 unit follows the external string profile into primitive
2362+ named-property reads. Looking up a data method on a string, number, boolean,
2363+ bigint, or symbol previously materialized every live frame binding into a
2364+ ` CallEnv ` , even though primitive ` [[Get]] ` resolves the realm's intrinsic
2365+ prototype and frame bindings named ` String ` , ` Number ` , and so on are
2366+ irrelevant. The direct path now uses an empty-frame realm view. Accessor
2367+ descriptors still take the observable slow path. A cached dynamic-realm marker
2368+ keeps indirect eval and dynamically constructed functions on the complete
2369+ frame view; this guard was added after the full runtime suite caught the
2370+ cross-realm case during development. Tests cover all five primitive families,
2371+ frame-name shadowing, a prototype getter side effect, and the pre-existing
2372+ marked-realm behavior. The implementation contains no workload name, source
2373+ path, iteration count, checksum, or expected result.
2374+
2375+ Against base binary SHA-256
2376+ ` 8b566b3cb05d6570273b20248654403ba92af6288ecf487bd127c437b631a70b ` ,
2377+ candidate binary SHA-256
2378+ ` 2a7d910a70158e6c92f22674447bbbe56d515b6e4918d11ecce47b22cb609364 `
2379+ improved two independent external string programs in eleven-block interleaved
2380+ runs: ` string-validate-input ` measured 0.778007x base and ` string-base64 `
2381+ 0.637258x. ` crypto-md5 ` and ` crypto-sha1 ` were neutral at 1.000493x and
2382+ 0.999578x, so they are not counted as wins. Complete corrected-candidate and
2383+ same-window base external previews retained identical coverage at 5/5
2384+ JetStream, 10/14 Kraken, and 23/26 SunSpider. Common-case candidate/base suite
2385+ geometric means were 0.981036x, 0.971765x, and 0.949689x; candidate/QuickJS-NG
2386+ remained far from B5 at 9.456082x, 5.507269x, and 6.260206x. Candidate external
2387+ raw/report
2388+ SHA-256 are
2389+ ` 739ed4b89f9c96218e71fe4b6c14b68fe107bb00d7b4ec517dc6411e3afa9ed5 `
2390+ and ` 7fcbc241d7416d2aaae25b4ca559a77238590bc792ff15d43e78dc08a6735b09 ` ;
2391+ base raw/report SHA-256 are
2392+ ` f39138977bab8708d557d150d02ddf9a030dd1da92a935a3cb64da9151396483 `
2393+ and ` 0c622af88447819a17350cef152a2ee89288db23fa4d6ca29a6f9bb53b11d654 ` .
2394+
2395+ The corrected candidate's complete broad physical plan produced all 1,625
2396+ records, 225/225 eligible formal measurements, zero non-OK samples, and 600
2397+ passing linearity samples. Candidate/base was 1.001860x overall; family ratios
2398+ ranged from 0.973575x for string through 1.006763x for call. The intended
2399+ ` string_slice ` case measured 0.973575x base. Candidate/QuickJS-NG was 0.194900x
2400+ overall, but allocation still failed B4 at 1.135609x. The strict analyzer
2401+ correctly rejected the receipt-less dirty development binaries, so these are
2402+ regression diagnostics rather than a fixed-hardware claim. Broad raw SHA-256
2403+ is ` c60e47562d42e47cc83c8c6e60bd07aa71be1406e7b603ebbc36c7236c582342 ` .
2404+ This unit generalizes across independent external string programs while
2405+ keeping the broad regression guard neutral; B4 and B5 remain active.
2406+
23442407## Historical Broad V1 Baseline
23452408
23462409The first complete baseline was recorded on 2026-07-15 at commit
0 commit comments