tri gates mutate: give the boundary operator a flag (Refs #2161)

[gHashTag]

Recomputed the full five-operator table with the repaired cache:

operator	killed	sites	rate
silent	47	52	90%
loud	28	32	88%
invert	72	78	92%
boundary	26	77	34%
assert	2	16	12%
total	175	255	69% — 80 survivors

Previously 157/244 with 87. The verdict-reaching operators sit near 90%; the whole weakness is two columns, and 40 of the 80 survivors are in one file.

Reading the survivors changes what they mean

invert_sites keeps only conditions whose body carries a verdict — hence 92%. boundary_sites mutates every comparison, and its survivors are while len(v) < N, if len(out) > 6, while j < len(src): loop bounds and display cutoffs. Moving those cannot make a gate stop failing. A 34% rate over sites that reach no verdict is a finding about the operator, not about the gates.

The assert survivors are thresholds with margin: floor int(0.9*60) = 54, measured 58 56 59 59 55 55. Two sit one point above the line — so the threshold is doing real work, and the boundary operator survives because 55 satisfies both >= 54 and > 54.

"Survivor" means "defect" for the return operators and does not for the other two, and this campaign had been reading one number across all five.

The flag

--loud, --invert and --assert each select one operator. Boundary was reachable only through --all. The one operator you cannot iterate on alone was the one with the worst rate — and no measurement would have found that, because the gap was in the argument parser, where none of these instruments look. It surfaced when the CLI refused a single-column run.

Added, with a test that every operator has a flag and that a misspelling is refused. The negative control earned its place at once: my first version built the wrong argv and reported every flag unaccepted, including three that worked. --loud failing was the tell that the harness was wrong rather than the parser.

Restricting boundary_sites to verdict-bearing comparisons is the fix for the 34%, and deserves its own measurement rather than being folded into the tick that found it.

193 tests pass.

Refs #2161

[github-actions]

PR Dashboard

Generated at: 2026-08-23 21:40:32 UTC

• tri gates mutate: give the boundary operator a flag (Refs #2161) #2612 tri gates mutate: give the boundary operator a flag (Refs #2161) | Created: 2026-08-23 | Updated: 2026-08-23 | 19 failing / 24 total
• Unbreak FPGA CI and correct two unbacked hardware claims #2081 Unbreak FPGA CI and correct two unbacked hardware claims | Created: 2026-08-11 | Updated: 2026-08-20 | 5 failing / 27 total

Summary

Status	Count
Total Open PRs	2
PRs with Failing Checks	1
PRs with All Checks Green	1
READY	0
FAILING	1
PENDING	0

Seal Status

• ⚠️ STALE -- sha256(compiler.rs)=e6333575ec10 != manifest seal=87e5cbd3ad94.
  The committed NMSE numbers were certified against an older compiler.rs.
  Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

[github-actions]

📓 NotebookLM Notebook linked to this PR

• Notebook ID: b83263109fb055dc
• URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.