diff --git a/.github/workflows/cli-tri.yml b/.github/workflows/cli-tri.yml
index bbb4f44759..db2806b9b9 100644
--- a/.github/workflows/cli-tri.yml
+++ b/.github/workflows/cli-tri.yml
@@ -72,8 +72,40 @@ jobs:
       - name: install yosys
         run: sudo apt-get update -qq && sudo apt-get install -y -qq yosys
 
+      # --nocapture --test-threads=1, and the log tee'd out, so the ratchet
+      # below can measure something `cargo test` otherwise throws away.
+      #
+      # Three smoke-gate tests need a Lean toolchain that no workflow installs
+      # (#2370). They do not use #[ignore] -- they println! a reason and return.
+      # So libtest counts them as PASSED and the summary line reads
+      # "0 ignored", and `cargo test` CAPTURES the stdout of a passing test, so
+      # the reason never reaches the log at all. Without --nocapture there is
+      # nothing to measure. Without --test-threads=1 the reason cannot be
+      # attributed to a test, because libtest lands it on the "test NAME ... "
+      # line it has already opened and parallel workers interleave those.
+      #
+      # set -o pipefail keeps tee from swallowing a real test failure.
       - name: cargo test -p tri
-        run: cargo test -p tri
+        run: |
+          set -o pipefail
+          cargo test -p tri -- --nocapture --test-threads=1 2>&1 | tee "$RUNNER_TEMP/tri-test.log"
+
+      # A green run of this job is consistent with those three never having
+      # executed. This step puts the count and the reasons in the job summary so
+      # a reader of a green run cannot mistake it for coverage, and ratchets the
+      # set: it FAILS if the set grows. A fourth test starting to skip silently
+      # is the regression nobody would otherwise see.
+      #
+      # It deliberately does not fail on today's three. Turning existing skips
+      # red is the owner's policy decision and a skip is not a failure; #2370
+      # stays open for the real fix, which is installing the toolchain.
+      #
+      # The baseline is emitted by this same script (--emit-baseline) from a real
+      # test log. It is never hand-written: a hand-typed "3" would restate what
+      # we already know and would not survive a rename.
+      - name: runtime skip ratchet
+        if: always()
+        run: python3 cli/tri/skipwatch.py "$RUNNER_TEMP/tri-test.log" cli/tri/skip-baseline.txt
 
       # `tri rtl check` reports numbers, and a binary that runs but reports
       # nothing would pass the two steps above. yosys is already on PATH from
diff --git a/cli/tri/skip-baseline.txt b/cli/tri/skip-baseline.txt
new file mode 100644
index 0000000000..dcae633d07
--- /dev/null
+++ b/cli/tri/skip-baseline.txt
@@ -0,0 +1,7 @@
+# Generated by cli/tri/skipwatch.py --emit-baseline.
+# Do not hand-edit: regenerate from a real test log.
+# Each line is a test that skipped ITSELF at runtime,
+# with its reason. libtest reports every one as passed.
+fpga::tests::test_measured_to_lean_standalone_builds_in_temp_lake_package	skip: lake not on path
+fpga::tests::test_smoke_gate_json_synthetic_validate_lean_standalone	skip: lake not on path
+fpga::tests::test_smoke_gate_validate_lean_standalone_matches_snapshot	skip: lake not on path
diff --git a/cli/tri/skipwatch.py b/cli/tri/skipwatch.py
new file mode 100644
index 0000000000..d320c8c74b
--- /dev/null
+++ b/cli/tri/skipwatch.py
@@ -0,0 +1,223 @@
+#!/usr/bin/env python3
+"""Ratchet on the set of `cli/tri` tests that skip themselves at runtime.
+
+Why this exists (#2370, and the narrow issue this landed under):
+
+Three smoke-gate tests need a Lean toolchain, no workflow installs one, and they
+do not use `#[ignore]`. They `println!` a reason and `return`, so libtest counts
+them as PASSED. Two consequences, both measured on master:
+
+  * `cargo test -p tri` reports `173 passed; 0 failed; 0 ignored`. The word
+    "ignored" is 0. The aggregate cannot tell you a phase never ran.
+  * `cargo test` captures stdout of passing tests, so the printed reason is not
+    in the log at all. `--nocapture` is required to see it, and
+    `--test-threads=1` is required to attribute it to a test name, because
+    libtest emits the reason onto the `test NAME ... ` line it already opened.
+
+So a green `cli-tri` run is consistent with those tests never having executed.
+This script makes that visible and ratchets it: it fails when the skip set
+GROWS, which is the regression nobody would otherwise see -- a fourth test
+starting to skip silently.
+
+It deliberately does NOT fail on the skips that exist today. Turning those red
+is the owner's policy call, and a skip is not a failure. #2370 stays open for
+the real fix (install the toolchain).
+
+Usage:
+    skipwatch.py <test-log> <baseline-file>            # ratchet (exit 1 on growth)
+    skipwatch.py <test-log> <baseline-file> --emit-baseline   # write baseline to stdout
+
+The baseline is produced by THIS script in --emit-baseline mode. It is never
+hand-written: a hand-written "3" would only restate what we already know and
+would not survive a rename.
+"""
+
+import os
+import re
+import sys
+
+# libtest, under `--nocapture --test-threads=1`, opens a line as
+# "test some::path::name ... " and then lets the test's own stdout land on it.
+TEST_LINE = re.compile(r"^test ([A-Za-z0-9_:]+) \.\.\. ?(.*)$")
+# The verdict libtest appends once the test returns.
+VERDICT = re.compile(r"^(ok|FAILED|ignored)$")
+# The in-repo convention for a self-skip, in either case, with : or -.
+SKIP_MARK = re.compile(r"^\s*skip\b\s*[:\-]", re.IGNORECASE)
+
+# Volatile substrings that must not enter the baseline: absolute paths differ
+# between a runner and a laptop, and pids/sizes differ between runs.
+ABS_PATH = re.compile(r"/[^\s,;)]+")
+LONG_NUM = re.compile(r"\b\d{3,}\b")
+
+
+def normalise(reason):
+    """Reduce a skip reason to something stable across machines and runs."""
+    text = ABS_PATH.sub("<path>", reason.strip())
+    text = LONG_NUM.sub("<n>", text)
+    return re.sub(r"\s+", " ", text).strip().lower()
+
+
+def parse(log_lines):
+    """Return (completed, {test_name: normalised_reason}).
+
+    `completed` is False when the log has no `test result:` line, i.e. the run
+    did not finish. That is NOT an empty skip set -- absence is not a value --
+    and the caller must refuse to ratchet on it.
+    """
+    completed = any(line.startswith("test result:") for line in log_lines)
+    skips = {}
+    i = 0
+    while i < len(log_lines):
+        match = TEST_LINE.match(log_lines[i])
+        if not match:
+            i += 1
+            continue
+        name, tail = match.group(1), match.group(2)
+        # The test's stdout: whatever shared the opening line, plus every line
+        # up to libtest's verdict or the next test.
+        block = []
+        if tail and not VERDICT.match(tail.strip()):
+            block.append(tail)
+        j = i + 1
+        while j < len(log_lines):
+            line = log_lines[j]
+            if (
+                TEST_LINE.match(line)
+                or VERDICT.match(line.strip())
+                or line.startswith("test result:")
+            ):
+                break
+            block.append(line)
+            j += 1
+        for line in block:
+            if SKIP_MARK.match(line):
+                skips[name] = normalise(line)
+                break
+        i = max(j, i + 1)
+    return completed, skips
+
+
+def render(skips):
+    return sorted("%s\t%s" % (name, reason) for name, reason in skips.items())
+
+
+def load_baseline(path):
+    try:
+        with open(path, encoding="utf-8") as handle:
+            raw = handle.read().splitlines()
+    except OSError:
+        return None
+    return sorted(
+        line.rstrip()
+        for line in raw
+        if line.strip() and not line.lstrip().startswith("#")
+    )
+
+
+def summary(text):
+    sys.stdout.write(text + "\n")
+    path = os.environ.get("GITHUB_STEP_SUMMARY")
+    if path:
+        with open(path, "a", encoding="utf-8") as handle:
+            handle.write(text + "\n")
+
+
+def main(argv):
+    if len(argv) < 3:
+        sys.stderr.write(__doc__)
+        return 2
+    log_path, baseline_path = argv[1], argv[2]
+    emit = "--emit-baseline" in argv[3:]
+
+    try:
+        with open(log_path, encoding="utf-8", errors="replace") as handle:
+            lines = handle.read().splitlines()
+    except OSError as exc:
+        sys.stderr.write("cannot read test log %s: %s\n" % (log_path, exc))
+        return 0 if not emit else 2
+
+    completed, skips = parse(lines)
+    observed = render(skips)
+
+    if emit:
+        sys.stdout.write("# Generated by cli/tri/skipwatch.py --emit-baseline.\n")
+        sys.stdout.write("# Do not hand-edit: regenerate from a real test log.\n")
+        sys.stdout.write("# Each line is a test that skipped ITSELF at runtime,\n")
+        sys.stdout.write("# with its reason. libtest reports every one as passed.\n")
+        for line in observed:
+            sys.stdout.write(line + "\n")
+        return 0
+
+    if not completed:
+        # The build or the run died. We have no measurement, so we make no claim.
+        print("::notice title=skip ratchet::test run did not complete; skip set NOT evaluated")
+        summary("### Runtime skip ratchet\n\nTest run did not complete — skip set not evaluated.")
+        return 0
+
+    baseline = load_baseline(baseline_path)
+    if baseline is None:
+        print("::error title=skip ratchet::baseline %s is missing" % baseline_path)
+        return 1
+
+    baseline_set, observed_set = set(baseline), set(observed)
+    new = sorted(observed_set - baseline_set)
+    gone = sorted(baseline_set - observed_set)
+
+    lines_out = ["### Runtime skip ratchet", ""]
+    lines_out.append(
+        "`%d` test(s) skipped themselves at runtime. libtest counts every one of "
+        "them as **passed** and reports `0 ignored`, so a green run here is *not* "
+        "evidence these ran." % len(observed)
+    )
+    lines_out.append("")
+    if observed:
+        lines_out.append("| test | reason |")
+        lines_out.append("| --- | --- |")
+        for line in observed:
+            name, _, reason = line.partition("\t")
+            lines_out.append("| `%s` | %s |" % (name, reason))
+        lines_out.append("")
+
+    print(
+        "::notice title=skip ratchet::%d test(s) skipped at runtime and were "
+        "counted as passed (baseline %d)" % (len(observed), len(baseline))
+    )
+
+    if gone:
+        lines_out.append(
+            "%d baseline entr(y/ies) no longer skip — tighten the baseline by "
+            "regenerating it:" % len(gone)
+        )
+        for line in gone:
+            lines_out.append("- ~~`%s`~~" % line.replace("\t", "` — `"))
+        lines_out.append("")
+        print(
+            "::notice title=skip ratchet::%d baseline entr(y/ies) no longer skip; "
+            "regenerate the baseline to tighten it" % len(gone)
+        )
+
+    if new:
+        lines_out.append("**FAILED — the skip set grew.** New runtime skip(s):")
+        for line in new:
+            lines_out.append("- `%s`" % line.replace("\t", "` — `"))
+        lines_out.append("")
+        lines_out.append(
+            "A test that skips itself is reported as passed. Either make it run "
+            "on this runner, or record it in `%s` by regenerating the baseline "
+            "and say in the PR why the coverage loss is acceptable." % baseline_path
+        )
+        summary("\n".join(lines_out))
+        for line in new:
+            name, _, reason = line.partition("\t")
+            print(
+                "::error title=skip ratchet::new runtime skip: %s (%s)" % (name, reason)
+            )
+        return 1
+
+    lines_out.append("No new runtime skips. Baseline holds.")
+    summary("\n".join(lines_out))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv))
diff --git a/docs/now/2026-08-22-cli-tri-runtime-skip-ratchet.md b/docs/now/2026-08-22-cli-tri-runtime-skip-ratchet.md
new file mode 100644
index 0000000000..a9f67c4d64
--- /dev/null
+++ b/docs/now/2026-08-22-cli-tri-runtime-skip-ratchet.md
@@ -0,0 +1,62 @@
+# NOW — a green cli-tri run said nothing about three tests that never ran
+
+Last updated: 2026-08-22
+
+## Runtime self-skips are now counted, named, and ratcheted (Closes #2377)
+
+- Branch: `skipvis/lean-smoke-gate`
+- Issue: #2377
+- **#2370 stays OPEN.** This does not fix it and does not pretend to.
+
+### What landed
+
+`cli/tri/skipwatch.py` — parses `cargo test` output for tests that skip *themselves* at
+runtime, writes the count and reasons to `$GITHUB_STEP_SUMMARY` and a `::notice`, and
+fails when the set **grows**.
+
+`cli/tri/skip-baseline.txt` — the three of them, emitted by `skipwatch.py
+--emit-baseline` from a real test log. Not hand-written. A typed `3` would restate what
+#2370 already established, would not survive a rename, and would detect nothing.
+
+`.github/workflows/cli-tri.yml` — the test step gains `--nocapture --test-threads=1` and
+tees its log; a new `runtime skip ratchet` step reads it.
+
+### Why both flags are load-bearing
+
+The three Lean smoke-gate tests do not use `#[ignore]`. They `println!` a reason and
+`return`. So libtest counts them as **passed** — the run reports `173 passed; 0 failed;
+0 ignored` — and `cargo test` *captures* the stdout of a passing test, so the reason
+never reaches the log. `grep -icE 'skip'` over the full default log returns 2, and both
+are test *names*, not reasons.
+
+`--nocapture` is what makes the reason exist in the log. `--test-threads=1` is what makes
+it attributable: libtest lands the reason on the `test NAME ... ` line it has already
+opened, and parallel workers interleave those.
+
+### Biting
+
+A fourth deliberately-skipping test was planted in `fpga::tests` and the suite re-run.
+`cargo test` reported `172 passed; 2 failed; 0 ignored` — the mutant counted as passed,
+`ignored` still zero, which is precisely the defect. The ratchet reported `4 test(s)
+skipped at runtime ... (baseline 3)`, named the new one, emitted `::error`, and exited 1.
+The mutant was then removed; `cli/tri/src/fpga.rs` is byte-identical to master.
+
+### Honesty boundaries (BINDING)
+
+- **No Lean toolchain was installed.** `elan` + a Lean toolchain is multi-gigabyte and
+  this machine has been at ~1.1 GB free all session, hitting literal zero eleven times
+  today. The real fix for #2370 is still the install, and it is still not done.
+- **This is a ratchet, not coverage.** It cannot make a skipped test run. It only makes
+  the skipping legible and stops the set from growing unnoticed.
+- **It only recognises the in-repo skip convention** — a leading `SKIP:` / `skip:` on a
+  line of test stdout. A test that bails out early while printing something else, or
+  while printing nothing at all, is invisible to it exactly as it is invisible today.
+  That is a real hole, not a detail.
+- **The baseline was generated on macOS**, with the demo bitstream present and yosys 0.63.
+  If the ubuntu runner's skip set differs, the ratchet reddens on its first run rather
+  than passing quietly — that is the intended direction of the failure.
+- Two tests (`test_smoke_gate_json_synthetic_verify_lean`,
+  `test_smoke_gate_json_theorem_matrix_is_computed`) fail on this laptop against local
+  yosys 0.63. They are green on the runner and were not touched.
+- The verifying job is `cli-tri`'s `build`, which is **not** a required check, so
+  auto-merge will not wait for it. It is watched on master after the merge.