RS3C-04 — Catalog-Sync Code Behavior Reconstruction (dot-catalog-sync) — 2026-06-21

Macro: RS3C (Mục tiêu B) · Deliverable: 04 of 13 · source-aware, line-cited Source-of-record: RS3C-02 §3, operational dot-catalog-sync sha256 7dd84cda… (264 lines), byte-identical to deployed OPERATIONAL (DOT-015). Date: 2026-06-21 · read-only · 0 mutations

Headline: catalog-sync is a read/scan + report tool whose only write is meta_catalog.record_count (aggregate integer). It never writes dot_tools. This resolves the dual-writer question by source (RS3C-05).

---

1. Behavior reconstruction table

#	Claim (code-derived)	Source line/fragment	Confidence	Risk	Note
C1	Entrypoint = main "$@"	L264	HIGH	—	single entry
C2	Args: --type=<dot_tools|pages|collections>, --all, --help/-h, --local/--cloud	L208–219	HIGH	low	requires a type or --all else exit 1 (L216–219)
C3	Scan paths: dot/bin/dot-* (exec), web/pages/*.vue, Directus /collections	L84 find "${DOT_ROOT}/bin" -name "dot-*" -type f -perm +111; L110 web/pages; L131 /collections?limit=-1	HIGH	low	filesystem + REST reads
C4	Read vs write — READ: filesystem find, REST GET	L84, L52, L131, L173	HIGH	—	discovery only
C5	WRITE — only meta_catalog.record_count	L68 api_request PATCH "${BASE_URL}/items/meta_catalog/${item_id}" "{\"record_count\":${count}}"	HIGH	low	the sole mutation: an integer count on meta_catalog
C6	NEVER writes dot_tools	no POST/PATCH/DELETE to /items/dot_tools anywhere in L1–264	HIGH	—	not a registry writer → no dual-write on dot_tools (RS3C-05)
C7	Upsert/insert/update/delete: PATCH-update only (record_count)	L68; lookup by code L52	HIGH	low	no insert/delete; no upsert; updates one numeric field
C8	Conflict/clobber: only overwrites record_count (intended)	L47–75 update_record_count	HIGH	low	if meta_catalog entry missing → warn + return (L56–66), no creation
C9	dot_tools relationship: counts & lists, no mutation	L77–98 scan_dot_tools: find → count → update_record_count "CAT-006" → echo list	HIGH	—	reconciliation reporter for CAT-006, read-only on the registry
C10	Registrar relationship: shares scan dir, different sink	both scan dot/bin/dot-*; catalog-sync writes meta_catalog, registrar writes dot_tools	HIGH	low	no in-process call; non-overlapping write targets
C11	Timing/on-deploy: invoked with args; no internal scheduler	L205–262; dot_tools.DOT-015 trigger_type=NULL; CAT-006 metadata "on-deploy"	MED	low	scheduling is external (deploy hook), not in source; last_executed 2026-03-31
C12	Metadata derivation: counts only	L86, L112, L141	HIGH	—	no status/operation/paired_dot treatment
C13	status/operation/paired_dot treatment: NONE	absent	HIGH	—	catalog-sync does not touch DOT lifecycle fields
C14	Auth: DOT_TOKEN via dot-auth, REST bearer	L230–240, L37/L42	HIGH	med	same token family as registrar; scope unproven from this file
C15	TLS: --globoff only (no -k)	L36, L41	HIGH	low	unlike registrar (which uses -k), catalog-sync does not skip TLS verify
C16	Model-A counts (--all): counts 6 collections via aggregate, PATCHes their record_count	L151–184 update_model_a_counts	HIGH	low	extends record_count writes to CAT-000/001/003/004/005/009

2. Source-aware boundary conclusion (feeds RS3C-05)

• dot_tools writer set = { dot-dot-register } only. Catalog-sync is not a second writer to the registry (C6).
• meta_catalog.record_count writer set = { dot-catalog-sync } (and possibly others, out of scope). This is a count cache, disjoint from registry rows.
• The RS3B-02 hypothesis "catalog-sync may be a second dot_tools writer (potential dual-writer)" is DISPROVEN by source. The companion hypothesis "catalog-sync is read-only" is partially wrong — it does write, but only meta_catalog.record_count, never dot_tools.
• Therefore: no dual-writer clobber risk on the DOT registry; the only cross-effect is that catalog-sync's CAT-006 count can lag/disagree with dot_tools (a reporting drift, not a registry mutation).

3. Status

• Behavior reconstruction: COMPLETE_SOURCE_DERIVED.
• Dual-writer-on-registry: DISPROVEN_BY_SOURCE (catalog-sync writes only meta_catalog.record_count).
• No unsafe registry-write behavior found in catalog-sync; its only write is a benign count field.
• Registration gate unchanged: REGISTRATION_HOLD · CAN_PROCEED = NO.