Cascade sweep: 939 auto-filed regressions at f5c8fbec6016 — one root cause, already fixed

• Type: triage note (fable-O, 2026-07-18 evening) covering the 939 regression-* tickets auto-filed by twatch(borg) at bad f5c8fbec6016, bulk-moved here.
• Root cause: ONE missing FPC-seed forward — the aarch64 residency mirror (588d525c) called RegcallScalarType from ir_codegen_aarch64.inc, included before the defining ir_codegen.inc. PXX's prescan resolves it (all native gates green); the single-pass FPC seed does not → fpc-bootstrap red → every FPC-dependent job (lib-fpc-clean, the test-asm suite, the FPC-built test-core sweep) cascaded red on the next watcher pass, and each red auto-filed a ticket.
• Fix: 938c0154-adjacent forwards.inc forward (commit "fix(A): forward-declare RegcallScalarType for the FPC seed path"). Borg's follow-up report (2185b272) confirms FIXED: fpc-bootstrap, lib-fpc-clean, test-asm.
• Not individually triaged — they share the one bad SHA and the one root cause. If any of these jobs shows red at a LATER sha, that is a NEW finding: file fresh, do not resurrect these.
• Watcher-behavior observation for Track T: a red fpc-bootstrap should probably SUPPRESS (or fold into one ticket) the downstream FPC-dependent jobs' auto-filing — 939 tickets for one missing forward is noise that buries real signal. Left as a suggestion, not filed as a T ticket.

Second flood, same night: 476 tickets at f6cad82e8063 — borg harness collapse

A second mass filing (2026-07-18T19:56Z), ALL at f6cad82e8063 — borg's OWN devdocs-only ticket-flood commit (cannot change codegen), every ticket with bad==good, 0-in-range, and an EMPTY log tail (incl. selfhost-fixedpoint, which ran green natively five times tonight; sqlite-threads + zlib also green natively). The empty tails say the borg-side run itself collapsed (jobs reported red with no output — disk/clone/harness state, possibly the 939-file commit itself), not that anything regressed. Swept here like the first flood. The cascade suppressor (tools/twatch.py CASCADE_THRESHOLD, landed tonight) would have filed ONE ticket for each of these events — the borg daemon needs a restart/pull to pick it up, and its clone/disk state deserves a look (flagged to the user).