Boris learned HTML 4.01 Strict

• Surface: boris + oliver, via a Droid session in a worktree (fac-boris-001, branch droid/implement-boris-issue-1003)
• Date: 2026-09-27
• Evidence: boris #1004 (merged), issue #1003, oliver #131, and the normative contract it shipped
• Verdict: promote-to-builds

The compiler and the renderer split a job between them. Oliver — the Zig HTML renderer boris builds on — grew an opt-in HTML 4.01 Strict serializer profile upstream. Boris issue #1003 then asked the right question first: is there a concrete consumer for a 1999-conformance mode, and if not, close it as not planned — "do not bump Oliver solely to acquire the option." The decision was made, and the implementation was delegated: one Droid session in a worktree took the issue end to end and came back green.

What shipped in boris #1004: a per-target --target-profile NAME=html4-strict flag; a fail-closed checker that inspects the exact assembled bytes of every page — layouts, generated nav, TOC, Aside, relations, backlinks, the Proof Pack, all of it — and refuses unchecked raw HTML passthrough; EHTML4STRICT diagnostics with page and layout context; a strict-compatible theme and fixture; and a normative contract that is unusually honest about what it is: a conservative valid subset, not an SGML DTD engine, with no accessibility conformance claimed. Release gate passed; an independent tidy run corroborated the fixture with no parser errors. A compatibility format, by the contract's own words — "not a better default for modern sites."

The meta part is that this blog is the natural first consumer of the feature, and the open question is whether that is a bad idea or a great one. The honest answer is both, and the full writeup with the handcuff bill lives at Boris whole-site HTML 4.01 Strict profile.