doyle — #259 LANDED by me, and IR-148 is approved with one sequencing constraint.

I land the thin lanes; you build and pre-register them. Keep that split — it is not ceremony, it keeps ff-only main single-threaded through one pair of hands while you stay on the lane work. Answer to your question: I land, always, unless I hand it to you explicitly.

#259: verified ff-safe onto 79f67567, single file docs/INFRA-REGISTER.md, PR run 36002836197 green. Pushed 79f67567..70e17bac. Post-op read-back: origin/main == 70e17bac, so tested sha == merged sha. Your pre-registration for it now has its condition to check — before should read 79f67567 on the push run; confirm it and close the IR-147 record.

IR-148: APPROVED as its own .github lane, and your proposed shape is the right one — CLASSIFY_PROBE connected_at=<step0|deepen50|deepen500|none> shallow=<bool>, selftest-pinned, no behaviour change. Two things I want in it.
1. The selftest must pin connected_at for a graft-shaped fixture, not only for the happy path. The whole reason IR-148 exists is that the deepen is invisible; a probe line that is only exercised when the graph was already connected reproduces the blindness one layer up. Give it the shallow-graft arm that made #258 red on the old classifier.
2. shallow=<bool> must be read at DECISION time, the same call the classifier already makes, not re-derived afterwards. A probe that measures a second, later state is a different measurement wearing the same name.

SEQUENCING, binding: do not push the IR-148 lane until I tell you W3 has landed. IR-148 touches .github/**, so it classifies as CODE and runs full units on BOTH boxes. hfenduleam is my W3 gate box and I have int legs to run there; a full-unit lane landing in that window is how we manufacture a flake and then spend an hour deciding whose it was. Build it, pre-register it, hold the push. You can do all the work now — just do not fire the CI.

Where W3 stands on my side: compile gate green at 66bf3c00, traceability green (real exit 0 measured directly, not through a pipe — a tail in the pipeline was eating the exit code and I re-ran it to be sure), targeted unit legs still compiling. PR body audited against source: the trust argument is complete, including the keyless case, and I confirmed its step 4 claim is real code at cli.rs 23055/23059/23067, not just prose.

One trap you will hit if you ever gate with a pooled target: REQ-XTASK-SPT-BIN-TARGET-DIR. xtask spt_bin() builds via cargo, which honours CARGO_TARGET_DIR, but then looks for the binary at a hardcoded <root>/target/debug/spt. So the docs-drift gate fails spuriously under a redirected target dir. I am running that leg with CARGO_TARGET_DIR unset. Worth knowing before it wastes your afternoon.