test(sampler): pin the parameter SET, not a bare count

[shipyard-local]

main is red

Any lane that builds examples currently fails. PulpSampler registers ten parameters; the contract test still asserts nine.

registered ids: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
declared ids:   1, 2, 3, 4, 5, 6, 7, 8, 9

The tenth is kSamplerHeritageClockRatio, added with the sampler heritage kit (39c7dac, #6396).

Which side is wrong: the test

The parameter is real and fully wired, so it is not the thing to remove:

• declared — pulp_sampler.hpp:46, kSamplerHeritageClockRatio = 10
• read — pulp_sampler_render.cpp:94, state().get_value(kSamplerHeritageClockRatio)
• exercised — test_pulp_sampler_heritage_stream_domain.cpp:505/525/541 drive it at 2.0 and 4.0
• fully specified — "Heritage Clock Ratio", range 0.25–4.0, default 1.0, step 0.01

#6396 did edit this contract file — it updated the resource-estimate assertion and missed the count one. So this is a stale assertion, not a rogue parameter.

Why assert the set instead of bumping 9 → 10

A bare count fails as 10 != 9, which names nothing. Re-reading nine as ten would leave the identical trap for the eleventh parameter. Asserting the ordered set makes the failure self-describing — the two INFO lines above are the actual output of the negative control.

Why nothing caught it

The required PR gate builds with PULP_BUILD_EXAMPLES=OFF, so no PR or merge_group check compiles this file. nightly-full-build.yml does build examples ("Deliberately NO -DPULP_BUILD_EXAMPLES=OFF"), so the lane that catches it runs only after landing — and release-cli builds examples too, which is where a stale example break surfaces if nobody is watching nightly.

That gate hole is worth addressing separately. Adding examples to the PR gate is the wrong fix — they are the heavy targets, which is why they are off. The candidates are making the nightly failure reach someone, or building examples in merge_group. The latter needs care: merge_group is serial at max_entries_to_build: 1 and its check_response_timeout is already shorter than a legitimate validation.

Verification

• pulp-sampler-test, full binary, with this change: 4613 assertions in 186 test cases, all passed
• Negative control: omit the tenth id from the declared set → fails, printing the two lines above, then passes again when restored
• tools/scripts/gates.sh origin/main: green

Found by the VST3 hosted-editor session, which deliberately did not fix it to avoid smuggling a sampler change into a VST3 PR. They left open whether the test or the parameter was wrong; the wiring evidence above answers it.

---

🔎 Provenance

Agent	claude
Machine	m3
Tab	Set up fleet v3 project with Orchestrate in new workspace
Directory	/Volumes/Workshop/Code/pulp-mq
Session	1dd2217d-ca26-4de4-b90e-6360ee343606

Resume

claude --resume 1dd2217d-ca26-4de4-b90e-6360ee343606

Jump to this tab

cmux surface focus 22A04F4D-188B-4306-91D2-ECAEADCC7F13

Relaunch (any agent)

cmux surface resume get --surface 22A04F4D-188B-4306-91D2-ECAEADCC7F13

Restore URL — https://claude.ai/code/session_01363VPyLiV9Er9DJ1pcmC9c

_{stamped 2026-07-22 00:27 UTC}