plan(roadmap): cut v1.135 around PERF-P01 alone; TRACE-P02 → v1.136

[avrabe]

Verify-Filter: (has-tag "performance")

v1.135 ships the cascade timing baseline and nothing else.

Why minimal

Performance was the one axis on which this project could make no claim at all. engine-throughput has guarded the cFS engines since v1.56; the control path that actually flies had no timing baseline of any kind, and the Lean WCET proof had never been confronted with a measurement. PERF-P01 (#361) closes it:

full_cascade_tick   1.2413 us    <- all five stages, one tick
  iekf_propagate    1.1256 us       91 % of the cascade
  the other four                    ~108 ns combined

Deferred, with reasons in the artifacts

requirement	to	why
TRACE-P02	v1.136	next, and now precisely sized
MATHF32-P06	v1.139	not blocked — lower value than the four ahead
OCI-P03	v1.139	parked pending rewrite — do not implement as written

TRACE-P02 is no longer one undifferentiated sweep

Measured 2026-08-25:

population	count	cost
≥1 CI-runnable step	155	mechanical — promote from the gate result
bench-only only	36	judgement — named waiver or a real step
no steps at all	45	cannot be evidence; write steps or demote

And zero implemented requirements lack a verifier — the 22/247 figure is entirely unpromoted verifiers, not missing evidence. That makes the bulk far cheaper than it looked and isolates the 45 as the real work.

OCI-P03’s warning, restated because release order makes it easy to pick up

Upstream wit-bindgen 0.60 already generates [u8; 8], so the fixed-length-list case for the fork is gone. The fork is three minors stale (0.58 vs upstream 0.61.1) and nothing in relay depends on it.

The drift that actually matters is in the flight path: components pin wit-bindgen-rt 0.41.0 while the workspace is on 0.60.0 — a 19-minor spread, and 0.41 generates the bindings baked into every published artifact. Rewrite the requirement around that migration; it needs no fork.

One criterion deliberately left open

PERF-P01’s (d) — the PX4 head-to-head — is not discharged, and is now filed as #362 rather than left as an unmet criterion nobody re-reads. We can say what our cascade costs; we cannot yet say it costs less.

rivet validate PASS, exit 0.

🤖 Generated with Claude Code