Repair main red: interpreter primitive surface census

[briansrls]

Summary

Repairs the red main build in dag/test/claim/v1_interpreter_primitive_surface_witness_test.dag. Two witnesses were failing: the_surface_is_almost_entirely_derived and the_three_denominators_are_derived_and_distinct.

Root cause (measured by execution, not transcribed)

Two same-day commits each landed a genuine new derived arm in the v1 interpreter's v1_builtin_arms! macro (the single authority whose expansions generate both the dispatch match and the machine-readable roster from one token list), but only one of them touched the census pin, and it touched it incorrectly:

• Activate tracked Rust host observation with typed refusals #7562 added free_call.parse_stage0_cargo_manifest_bins inside v1_builtin_arms! — confirmed by locating the macro's open/close (macro_rules! v1_builtin_arms { ... } spans lines 9286–10565 in src/v1/stage0/src/v1_interpreter.rs) and finding the arm at line 9370, inside that span. It did not update this test file's pinned literals.
• Guarantee probe observation surface: compile_dag_diagnostic_census, and the first floor-class measurement #7575 added free_call.compile_dag_diagnostic_census, also inside the same macro span (line 10385), and did bump the pin (181→182) — but by incrementing the value already sitting in the file rather than re-measuring the surface, so it silently carried forward Activate tracked Rust host observation with typed refusals #7562's un-pinned addition instead of catching it. That's the actual defect: an increment stood in for a measurement.

Both arms are genuinely derived, not native/hand-added — exactly the kind of change this census exists to auto-track, not a weaker-surface regression. So the fix is to re-pin at the true measured values, not to weaken the equalities.

Measurement method

Built claim_batch locally (sccache off, bypassing the ctrl-build shim to avoid a known rustc ICE and to avoid CTRL_BUILD_MODE=remote's no-local-binary default). Wrote a scratch probe .dag module with equality witnesses bracketing each of the five candidate counts (e.g. v1_interpreter_derived_arm_count() == 183), ran it directly against the current worktree with claim_batch --source-root dag --source-root src/v2 --entry <probe> --functions <csv>, read which candidates PASSed, and deleted the probe before committing. This is the same mechanism every_dispatch_site_is_enumerated etc. use — no value here was transcribed from a peer session's claim or a CI failure message.

Measured: derived_arm_count=183, declared_arm_count=2, dispatch_site_count=7, distinct_arm_identity_count=174, authored_spelling_count=161, row_count=185 (183+2, cross-checked directly against v1_interpreter_row_count()).

Changes

• dag/test/claim/v1_interpreter_primitive_surface_witness_test.dag: the three pinned literals (182→183, 173→174, 160→161), and exact_count_witness_note's illustrative numbers updated to match (181/172/159/183 rows → 185/174/161/185 rows) — fixing only the code literals while leaving the note's prose stale would repeat exactly the defect that made this repair necessary.

Verification

All 24 witness fns in the file PASS by execution against the current worktree (local build, not CI-transcribed).

Blast radius

Grepped dag/ and src/v2/ for other transcriptions of these five counts — none found. The only other numeric mention near this carrier is an unrelated net-line-count note in dag/gunbc/v1_interpreter_primitive_surface.dag (added_hand_rust_accounting_note, about v1_interpreter.rs's +178 line delta), untouched here.

🤖 Generated with Claude Code