docs(architecture): reconcile Windows sandbox RFC bilingual pair

[daierenao]

Summary

The Windows sandbox RFC pair claimed translation_status: synced from heading lockstep while four Current/deferred claims were missing in Chinese. This PR restores those claims (A–D) and sibling counterpart links so the published status matches semantic parity.

Fixes #5338
Refs #4726

Serialization: #3952 is open and edits both RFC files for Glob/junction policy. This slice does not take that policy; it patches current main.

Verification

• rg '^#{2,3} ' → 23 headings on both files; §6.5 heading date 2026-08-24 unchanged
• Chinese §6.4 states OS version checks alone are insufficient (仅凭 OS version check 不够)
• Chinese §6.5 intro names Windows 11 x64 and Phase 4 roadmap(windows): make Windows a supported platform #2142
• Chinese §6.5 deferred launcher bullet keeps in-broker digest as Current and Authenticode as deferred
• Chinese §6.5 close keeps the fail-closed sentence and adds the W1 / Phase 4 lifecycle-evidence claim
• YAML counterpart present both ways in sibling ./ form
• npm run check:asf-headers (also ran as the pre-commit hook)
• Did not run check:architecture (desktop renderer ratchet, not this gate)

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Cursor Grok 4.6 drafted the Chinese claim restorations A–D and counterpart metadata from the locked planning paste. The commit retains a Generated-by: Cursor Grok 4.6 trailer.

Checklist

• Tests cover the change and fail without it — N/A, docs-only
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

Made with Cursor

[Astro-Han]

Docs-only reconciliation, +10/−8 across the two RFC files. I re-derived parity independently from gh pr diff rather than relying on the PR's own verification list. The pair is genuinely reconciled; I found no blocking issues.

Verified: the four restored claims (A–D from #5338) land at matching strength

Claim	EN source	ZH at head	Assessment
A — lifecycle evidence is release evidence, not an assumption	windows-sandbox-rfc-v1.md:372-374	…zh-CN.md:225	Strength preserved verbatim in effect; the preceding fail-closed sentence is retained, not overwritten.
B — packaged Windows 11 x64 backend + remaining guarantees tracked by Phase 4 #2142	windows-sandbox-rfc-v1.md:256,264-266	…zh-CN.md:180	Both fragments were genuinely absent before (ZH never mentioned Windows 11); #2142 now appears in the §6.5 ledger, not just the header.
C — OS version checks alone are insufficient	windows-sandbox-rfc-v1.md:218	…zh-CN.md:169	Negation kept (不够), and inserted at the same sentence position — before the implemented parenthetical, matching EN. The existing 而非仅凭二进制存在 (≠ "file presence alone") is correctly treated as a distinct claim.
D — digest Current / Authenticode deferred split in the §6.5 ledger	windows-sandbox-rfc-v1.md:337-339	…zh-CN.md:208	The old bullet (随 Phase 3 签名一并落地) really did under-claim — it implied the whole check was deferred, dropping the "digest already enforced in-broker" half. Rewording to readiness 时的 … 校验 plus the digest/Authenticode split fixes a real ledger inaccuracy, not a translation nicety.

Verified: structure, links, and metadata

• Heading parity — 23 H2/H3 in each file, identical numbered outline (§1–§12 + W0–W3). The §6.5 snapshot date 2026-08-24 is unchanged in both headings, as intended.
• Anchors resolve — the §-reference set is identical across the pair (§5, §6.1–§6.5, §7.1, §9) and every one targets a real heading in both files.
• Link sets identical — 11 links each, byte-for-byte the same set. The sibling prose links (./windows-sandbox-rfc-v1.zh-CN.md / ./windows-sandbox-rfc-v1.md) resolve to files that exist.
• counterpart: metadata — added at line 6 of both files in ./sibling.md form, matching both skills/maka-architecture-docs/references/bilingual-standard.md:84 (suggested YAML) and the 7 pre-existing pairs in docs/architecture/. Correctly bidirectional; this closes parity check 8.
• Dates aligned, headers untouched — last_verified: 2026-09-15 and body Updated set in both files; ASF license blocks unmodified (consistent with the pre-commit asf-license-headers.mjs check-staged hook you cited).
• No gate skipped — there is no docs frontmatter/translation-sync checker in package.json; check:architecture really is only the desktop renderer ratchet, so not running it is correct.

Verified: sections the PR does not touch were already at parity

I spot-checked §6.3, §9/W1, and §10 to test the claim that A–D were the complete claim-level deviations. They hold up: §6.3 carries the same ten-way-concurrency / desktop-heap quantitative limit in both; §9 W1 has 6/6 checked boxes in each; §10 has 9 bullets plus the same 7-row evidence table; the deferred/暂缓 keyword count is 17 in both. §6.5's enforced (14) and deferred (11) bullet counts also match. I did not find another claim-level gap.

Referenced issues resolve with states consistent with the prose (#2961/#3722/#3174 merged, #2142 and #3952 open), and #3952's Glob/nested-junction work is correctly left out. The "packaged Windows 11 x64" claim is backed by real packaging (scripts/package-windows-x64.mjs, Maka-<ver>-win-x64.exe), and the Authenticode deferral matches the EN source.

Nits

• PR checklist: "Tests cover the change and fail without it" is ticked, but the diff is two markdown files with no test changes. For a docs-only reconciliation that box reads as inaccurate — worth unticking or annotating with "N/A, docs-only".
• EN - Updated: bump (windows-sandbox-rfc-v1.md:37): the English prose did not change in this diff — only frontmatter. Bumping it to 2026-09-15 implies an English content edit that didn't happen. last_verified moving is clearly right; consider leaving EN's Updated at 2026-08-18, or noting that it tracks frontmatter-only changes.
• ZH §6.5 intro reads redundantly (…zh-CN.md:180): the paragraph now has two adjacent 其余… clauses — …其余未标注条目由 #2961 当前强制。其余保证已设计但明确作为后续门禁暂缓…. This is faithful to EN's two sentences, but the Chinese could use e.g. 其余保证 → 尚未强制的保证 to avoid the echo.
• #5338 accounting (not this PR): the issue says 6 pairs in docs/architecture/ already set counterpart; there are 7 (the extra one being llm-compaction-events-log-projection-draft). Immaterial to the change.

[daierenao]

Thanks for the independent re-derivation. Adopted nits 1–3 in e5ded10:

• Checklist: unticked Tests and marked N/A, docs-only.
• EN - Updated: restored to 2026-08-18 (English prose unchanged; last_verified stays 2026-09-15).
• ZH §6.5: 其余保证 → 尚未强制的保证 so it does not echo the preceding 其余未标注条目.

Left #5338's 6-vs-7 counterpart count alone, as you noted it is outside this PR.

@Astro-Han