fix(corpus-ratchet): remove the t27c test step; it was measured on the wrong tree

.github/workflows/corpus-ratchet.yml

@@ -59,65 +59,17 @@ jobs:
       - name: Build t27c
         run: cargo build --release -p t27c
 
-      # Closes #2288. `cargo build` does not compile `#[cfg(test)]` modules, so
-      # eight workflows on master build t27c and not one of them ever compiled,
-      # let alone ran, a single test in it. Measured 2026-08-20: 1221 tests
-      # exist and 1221 of them had never executed in CI. They all pass -- so
-      # this lands as a plain gate, not a ratchet.
+      # A `cargo test -p t27c` step lived here for 34 minutes on 2026-08-20
+      # (ce6ea628b) and is removed again by #2292. It was landed on a measurement
+      # of 1221 passing tests taken in a working tree checked out at
+      # feat/wave-547/host-heapsort, NOT at master -- that branch's
+      # bootstrap/src/compiler.rs is 22,142 lines against master's 36,970. On
+      # master the same command is 1602 passed / 13 failed / 2 ignored, so the
+      # step went red on its first run here.
       #
-      # It goes here because the crate is already built one step above, so the
-      # marginal cost is compiling the test targets and running them: measured
-      # 1m43s cold / 1.75s warm, against this job's 45-minute ceiling and its
-      # ~314s ratchet run.
-      #
-      # Scope, stated openly -- this runs what `cargo test -p t27c` REACHES,
-      # which is not every test in the tree. 913 `#[test]` attributes live in
-      # bootstrap/src; 895 compile into this run. The missing 18:
-      #   * bootstrap/src/math_compare.rs -- 10 tests, and no `mod math_compare;`
-      #     exists anywhere in the crate, so rustc never opens the file. Built
-      #     standalone it is 9 pass / 1 fail (see #2288 follow-up); the failure
-      #     is a bad assertion, not a bad implementation, so it is NOT silenced
-      #     here and NOT wired in here -- wiring it in would land this gate red.
-      #   * bootstrap/src/proxy.rs -- 8 tests behind
-      #     `#[cfg(all(test, feature = "server"))]`. No workflow passes
-      #     `--features server`, so they are stripped before type-checking.
-      # Neither hole is created by this step; both are named so the green tick
-      # is not read as "every test in bootstrap/src ran".
-      - name: Unit tests (cargo test -p t27c)
-        id: unit_tests
-        run: |
-          set -o pipefail
-          cargo test -p t27c --release 2>&1 | tee unit_tests.log
-          rc=$?
-          totals=$(grep -E '^test result:' unit_tests.log \
-            | awk '{p+=$4; f+=$6; i+=$8} END {printf "%d passed, %d failed, %d ignored", p, f, i}')
-          bins=$(grep -cE '^test result:' unit_tests.log)
-          echo ""
-          echo "t27c unit tests: ${totals} across ${bins} test binaries."
-          {
-            echo "### t27c unit tests"
-            echo ""
-            echo "\`cargo test -p t27c --release\` -- **${totals}** across ${bins} test binaries."
-            echo ""
-            echo "Not covered: 10 tests in \`bootstrap/src/math_compare.rs\` (no \`mod\` declaration"
-            echo "anywhere, so rustc never compiles the file) and 8 in \`bootstrap/src/proxy.rs\`"
-            echo "(behind \`feature = \"server\"\`, which no workflow enables)."
-          } >> "$GITHUB_STEP_SUMMARY"
-          exit $rc
-
-      - name: Explain a unit-test failure
-        if: failure() && steps.unit_tests.outcome == 'failure'
-        run: |
-          echo "::error title=t27c unit tests failed::A Rust test in the t27c crate failed."
-          echo ""
-          echo "This is NOT the corpus ratchet and has nothing to do with"
-          echo "docs/reports/suite_expectations.json. Do not bless an expectation for it."
-          echo ""
-          echo "Reproduce locally with:"
-          echo "    cargo test -p t27c --release"
-          echo ""
-          echo "The failing test names are in the step log above and in the"
-          echo "corpus-ratchet-log artifact (unit_tests.log)."
+      # It should come back as a RATCHET over the 13 named in #2292, never as a
+      # plain gate: a gate that lands red gets disabled rather than obeyed.
+      # Do not re-add it without a measurement taken on master.
 
       - name: Run the corpus ratchet
         id: ratchet
@@ -163,5 +115,4 @@ jobs:
           path: |
             ratchet.log
             suite_summary.json
-            unit_tests.log
           retention-days: 14

docs/NOW.md

@@ -1,3 +1,63 @@
+# NOW -- the test gate landed red because the measurement was taken on the wrong tree (2026-08-20)
+
+Last updated: 2026-08-20
+
+## fix(corpus-ratchet): remove the `cargo test -p t27c` step landed 34 minutes ago
+
+The entry directly below this one is **wrong**, and this is the correction.
+
+`ce6ea628b` added a `cargo test -p t27c --release` step to `corpus-ratchet.yml` on
+the strength of "1221 passed, 0 failed, 0 ignored". That measurement was really
+taken and really said that. **It was taken in the wrong working tree.**
+
+The host checkout sits on `feat/wave-547/host-heapsort`. The gate runs on `master`.
+The two trees are nowhere near each other:
+
+| | `feat/wave-547/host-heapsort` | `origin/master` |
+|---|---|---|
+| `bootstrap/src/compiler.rs` | 22,142 lines | **36,970 lines** |
+| `#[test]` in `bootstrap/src` | 1,781 | 1,622 |
+| `cargo test -p t27c` main bin | 895 passed, 0 failed | **1602 passed, 13 failed** |
+
+The step went red on its first run on master. It is removed here. The 13 failing
+test names, and the plan to re-land this as a ratchet rather than a gate, are in
+#2292; #2288 and #2289 are reopened.
+
+Kept from `ce6ea628b`: the `Explain a failure` step stays scoped to
+`steps.ratchet.outcome == 'failure'`. As a bare `failure()` it answers a failing
+`Build t27c` with "add an entry to `docs/reports/suite_expectations.json`", which
+is wrong independently of any test step.
+
+### Honesty limits (BINDING)
+
+- **The number in the entry below -- 1221 tests, 0 failures -- describes a tree
+  that is not master and never was.** On master the same command is **1602 passed,
+  13 failed, 2 ignored** in the main binary. Both numbers are real measurements of
+  different things; only the second one is about the branch that gates merges.
+- **13 is a lower bound, not the failing count.** `cargo test` stops at the first
+  failing target, so the 21 test binaries in `bootstrap/tests/` never ran on master.
+  Their state is **unmeasured**. A ratchet baseline needs `--no-fail-fast` first.
+- **The local verification was real and proved nothing about master.** The step body
+  was run verbatim from the parsed YAML and exited 0; a shape-identical harness
+  confirmed it exits non-zero on failure. Both were true, on the wrong tree. Running
+  a command locally is not evidence about CI unless the checkout matches.
+- **This made things worse for 34 minutes, not merely no better.** `corpus-ratchet`
+  was *already* failing on master at `Run the corpus ratchet` (`eeb779e4c`,
+  `57a53005a`, `2255e4c32`, `bffd38982`, `400850702`). Putting a failing step in
+  front of it short-circuited the job, so the pre-existing ratchet failure stopped
+  being reported at all.
+- **`master` has no branch protection.** `GET /repos/gHashTag/t27/branches/master/protection`
+  returns `404 Branch not protected`. PR #2291 auto-merged at `03:57:13Z`; its own
+  `corpus-ratchet` run started at `03:57:16Z` -- it merged **three seconds before the
+  gate it was adding began to run**. Nothing in this repo that calls itself BLOCKING,
+  including `corpus-ratchet.yml`'s own `W632 (BLOCKING)` header, currently blocks
+  anything. That is not fixed here and is the larger finding.
+- **#2290 stands unaffected** -- `bootstrap/src/math_compare.rs` is still compiled by
+  nothing, and its `test_hybrid_v2_plateau` still asserts a bound 2.10x tighter than
+  the file's own goldens. No test was edited in either direction.
+
+---
+
 # NOW -- 1221 tests that had never run in CI now run in CI (2026-08-20)
 
 Last updated: 2026-08-20