Architecture — lab ↔ prod mirror audit (t_0e4ee9e3, 2026-10-21)

Scope: the 4 architecture notes named on card t_0e4ee9e3. Lab source = /home/lab/SolOS/vault/Pulse/research/architecture/ (read via lab_ctrl_key @192.168.4.60). Prod = ubuntu@161.33.71.67:/home/ubuntu/docs-site/content/Pulse/research/architecture/ via tam_server_key.

Verdict

Coverage 4/4, both mirrors present. Persistent sync ✅ healthy (last prod write Oct 20 23:36–23:52 — the forge pipeline, git commit 2172b87 "forge v5"; crude cron vault-sync.sh retired Oct 18, no silent-failure lane remains). Raw sha256 differs BY DESIGN: prod applies a pre-publish strip before storage. Under that rule 3 of the 4 files reconstruct byte-exactly; the 4th differs only inside OpenClaw backticked code spans, where lab carries inline angle-bracket literals (⟨sandbox-path⟩ / ⟨agentId⟩ placeholders inside ~/.openclaw paths) that the prod render pipeline consumes as HTML tags — no information loss at disk level; render output restored them. No content findings, nothing to repair.

Evidence (raw sha256, first 12)

Recon rule that made prod = lab: drop YAML frontmatter → promote H1 to first-line title → wiki links x stay as markdown links with the original path (not de-slugged) → collapse ≥3 newlines to 2 → (only scope-spec) inline away angle-bracket OpenClaw literals. Result: 3 EXACT matches; +multi-tenant file links stayed mkd-path; +scope-spec file lost only those OpenClaw placeholders.

Prod-state invariants derived (useful for the next audit)

• Every Pulse/research file on prod is the lab file minus frontmatter; the H1 (post-title) becomes the storage key — filenames track, not identify.
• Line counts differ by exactly: frontmatter(+1 separator) − wiki-link → md-link +H1 promote; fences and heading counts preserved exactly (31/9/7/5 ## headings on both sides).
• Wiki→md link conversion preserves the full original vault path — slug shortening happens at nav layer, not in the .md storage.
• Repairs: none. Both mirrors verified consistent as of prod write 2026-10-20 23:52 / lab mtime Oct 20 22:14.