nx_wiki_structure_regen.nx

nx_wiki_structure_regen.nx -- IMS arc rung A4b: the RE-PUBLISH-ON-PULSE engine (full-auto cron driver). WHAT IT DOES (one pulse): reads the DATA-DRIVEN managed-page registry knowledge/registry/wiki_pages.tsv, and for every managed page that HAS an emitter it (a) RE-EMITS the page (fork+exec the sovereign build- runner on the emitter organ -> web_assets/<emit_file>), (b) builds the WOULD-BE corpus and runs the A1 orphan/dead-link MONITOR over it -- if re-emitting would leave any managed page with 0 inbound links (orphan) or introduce a dead internal link, the offending page is marked REFUSE_ORPHAN and SKIPPED, (c) runs the A2 publish GUARD (pg_decide) -- a REJECT skips the page with the exact reason, (d) if apply_flag == 0 records a DRY-RUN plan row and PUSHES NOTHING; if apply_flag == 1 it APPLIES via the A4a versioned publish (snapshot-then-guarded-push, multi-rollback-point). Pages with emitter "-" are reported manual- only (counted, never silently dropped, never published). COMPOSITION (every safety property routes through a PROVEN organ -- nothing reinvented): nx_ims_monitor nx_ims_orphans / nx_ims_dead_links / nx_ims_collect_targets (A1: the SOLE orphan + link-rot authority; the regen NEVER rolls its own link scanner). nx_wiki_publish_guard pg_decide (A2: the SOLE pre-publish integrity authority, fail-closed). nx_wiki_versioned_publish vpub (A4a: snapshot-then-guarded-publish; the ONLY path that can write live, reached ONLY when apply_flag==1; do_push_flag is wired to apply_flag so a dry-run can NEVER push -- ADDITIVE-ONLY + multi-rollback-point by construction). nx_wiki_index_builder NxWikiDocStore (the in-memory corpus the monitor reads). nx_syscalls fork/exec/wait for the emitter; sys_read_file for bodies; fixed-epoch stamp. REFUSAL ATTRIBUTION (mechanical, so the gate can assert EXACT identities): R-DEAD a page whose WOULD-BE body is the SOURCE of a dead internal link -> RC_DEAD_LINK (it introduced the rot) -> would_publish=0. R-ORPH a page whose re-emit DROPPED an outbound link (present in its baseline body, absent in its would- be body) where the dropped target is now an ORPHAN in the would-be corpus -> RC_REFUSE_ORPHAN (its change orphaned a managed page) -> would_publish=0. Also: an orphaned page itself is held back (RC_REFUSE_ORPHAN) -- we never publish a page into a corpus where it is unreachable. NO WALL-CLOCK in the publish record: the caller passes a FIXED epoch (the versioned record is determin- istic/reproducible). (The emitter's own footer uses sys_now_realtime_sec for the freshness STAMP -- that is the page's "generated at" text, separate from the deterministic version-index epoch we control here.) Hygiene: M3 every while has a hard iter cap; M5 every buffer index bounded; M7 named constants; M8 verdicts propagated. ("loop" and "match" are reserved words -- never used as idents.) license_tier: ORIGINAL