Description
docs/compatibility.md states the policy as “the current stable OPNsense release and the previous stable: today that means 26.1.x and 25.7”. Both halves are wrong as of 2026-09-20: upstream has tagged through 26.7.4, so current stable is 26.7.x and the named window is two releases out of date. The live evidence does not cover the stated window either - after OPN-0106 retires the prod profile and the lab consolidates, the only release-channel box is guest 106 tracking the current stable head, so previous-stable is verified from upstream source alone. Rob decided on 2026-09-20 to narrow the policy to the current stable main release only, rather than keep claiming a window nothing exercises. The page must claim what is actually verified and say how, because it is what an external contributor reads before filing a version bug. Check whether the page states the window in more than one place and whether any shim or exemption note justifies itself by the two-release window - the interfacesOverview release-vm entry cites stable/26.1 and was written when 26.1 was thought current, so its reasoning needs re-deriving rather than deleting.
Acceptance Criteria
- #1 docs/compatibility.md states a current-stable-only policy and names no release version that upstream has superseded
- #2 The page distinguishes what a live box verifies from what is derived from upstream source
- #3 Any exemption or shim note justifying itself by the two-release window is re-derived against the narrowed policy, and the interfacesOverview release-vm entry is corrected or removed
Definition of Done
- #1 just check
- #2 just gen (if any generated artifact changed) and the diff committed
Implementation Notes
Landed in 47e40235. Three pages carried the stale claim, not one: docs/compatibility.md, docs/faq.md and README.md all named “current and previous stable”. The two hits under docs/superpowers/ were left alone - that tree is gitignored scratch, confirmed at .gitignore:68.
Two things found while re-deriving, neither in the original scope. First, the page called the canary daily; that stopped being true when the schedule came off the workflow, so it now describes what the canary actually covers. Second, the shim-pruning paragraph promised that shims “get pruned” when a release ages out, which would be a bad promise to keep literally - it now says a cheap shape is usually kept, because tolerant-reader parsing is the whole reason an older box works.
AC3 came out bigger than expected. The interfacesOverview release-vm entry was removed as planned - stale reason, and redundant against a base missingOK that already covers rows[].link_typev6 - but nine MORE entries carry a “prune when 26.1 leaves the support window” trigger that this narrowing fires simultaneously: idsSettings, keaSubnets4, memoryStatistics, ndpTable, pfStatisticsByInterface, pfStatsInfo, protocolStatistics, quaggaOspfNeighbors, systemMbuf. I did not prune them and that is deliberate. Removing a missingOK entry alone does not remove a shim - it makes the canary ENFORCE a key no supported box sends, which manufactures exactly the standing-warning noise OPN-0096 had to clear once already. The real prune deletes the legacy struct field and drops best-effort reading for a box on the old shape, which contradicts the paragraph this same commit wrote. Each note now records that its trigger fired and points at OPN-0113, raised to take the nine decisions one at a time.
No CodeRabbit run: docs prose plus declarative ledger notes and one redundant entry removal, no branching logic. Validated through just check instead, exit 0.
Final Summary
Policy narrowed to the current stable release (26.7.x) across compatibility.md, faq.md and README.md, with a new section separating live-verified evidence from source-derived, in 47e40235. The interfacesOverview release-vm exemption is gone - stale and redundant. Nine further entries whose prune trigger this narrowing fired are deliberately NOT pruned, because removing a missingOK alone enforces a key no supported box sends and the real prune drops best-effort reading; their notes record the fired trigger and OPN-0113 carries the nine decisions. just check exit 0.