DOCS LANE READY FOR GATE — PR #167, head b7135d66, stacked on W5 9c3a129a.

FINDING SHAPE: W1-W5 put the entire IO contract in shells/frames.md + the GENERATED cli/reference.md and nowhere else (token census over docs-site/src for now-signal, shortform, DISPATCH_RESULTS, COMMUNE_FAIL, [io], SPT-NOW hits exactly those two files). So every gap was an ABSENCE, not drift. I filled the surfaces a harness author reads from and pointed the neighbours at frames.md rather than restating it — 13 files, 237 insertions.

Biggest ones: manifest.md had no [io] / [io.now_signal] at all (the compliance gate is manifest-driven and was undiscoverable from the manifest reference); api.md documented api state with no payload flags and had neither now-signal nor hint; the integration checklist had no IO row in any group or in the floor; docs/MANIFEST.md had no [io]; and llms.txt indexed shells/frames.md NOWHERE — that predates the milestone but became load-bearing the moment the IO contract landed inside it. Also two coherence fixes: shells/overview.md said a link carries ONE core-pushed thing (now two, opposite ends of the taxonomy), and lifecycle/overview.md's commune bullet did not know its failure mode is now signalled.

ONE CODE-ADJACENT CHANGE, flagging rather than asking: Io::shortform's doc comment — and the published schema description generated from it — said the field opts out of @<...@> parsing. It gates BOTH markers; delivery.rs guards dispatch_ingest_shortform AND dispatch_ingest_seal_mints off that one bool. Comment corrected, manifest.schema.json regenerated, drift gate green, no behaviour change. Revert is one commit if you want it out of a docs lane.

IR-52 RIDER, EXPLICIT AS ORDERED: this lane does NOT touch the docs-site CLI reference generator. No CLI surface changed, no xtask edit, reference.md is byte-identical. The conditional rider does not ride, so IR-52 STAYS OPEN in the register.

NOT TOUCHED DELIBERATELY: CONTEXT.md:1114's stale default-on sentence. Your e326ad4b fixes it main-side and is not an ancestor of this stack — re-fixing here would duplicate at assembly.

GATES at b7135d66, each read from its own exit file / captured line: xtask check OK exit 0 (docs drift + curated llms links); traceable-reqs check exit 0; manifest schema drift test 1 passed exit 0; clippy --workspace --all-targets exit 0 / 0 warnings on Windows; same on kitsubito with SHA=b7135d668f13... echoed in the same command, CLIPPY_EXIT=0. No test legs — the lane has no product behaviour for an e2e to bind, so I am not naming an IR-67 e2e for it; say the word if you want one anyway.

Doc stages for the milestone REQ set were ALREADY satisfied by frames.md at 9c3a129a; the new sections carry additional doc tags at their real evidence for REQ-IO-SHORTFORM-GATE, REQ-NOW-SIGNAL-SPEC, REQ-IO-INGEST-STATE-PAYLOAD, REQ-NOW-SIGNAL-VERB.

Acceptance you named was the drift gate PLUS a blind consumer read — I did the consumer read myself against source (spec precedence, without-beats-only, max_lines, hint's real usage shape which I had wrong at first draft and corrected off the generated help). A second reader who has not seen the code is the arm I cannot supply.