feat(ci,tests): PR-4b — promote teaching onto the gate, and close pin 3's own blind spot

The reachability gap turns by PROMOTION for the first time: 15 -> 14. PR-3b's
earlier turn was by deletion, which is free. This one costs 22.1s on every push,
which is exactly why it is a decision rather than a default.

THE RECOMMENDATION WAS RE-DERIVED, NOT EXECUTED — and it did not survive as
written. PR-4b recommended `teaching` on the strength of G-22:
test_ratification_ceremony.py was red on main, in `teaching`, run by no gate.
That hole was already closed a different way — G-22's fix promoted that single
file into smoke. So `teaching` would have added 9 net-new files, NONE with a
demonstrated escape, and promoting on a justification that has already been
spent is precisely the failure this arc exists to close.

Promoted on a stronger, doctrinal basis instead.
tests/test_epistemic_invariants.py enforces AGENTS.md's "Teaching and mutation
safety" non-negotiables in substance — INV-22/INV-23: an unmarked proposal and
an unmarked reviewed example default to SPECULATIVE, the status enum has exactly
its four positions, a pack-mutation proposal carries no hardening flag. Durable
corpus mutation is the one irreversible thing in this system, and the suite
guarding it ran on no pre-merge gate. That justification does not require an
incident to have happened yet, which makes it the better one.

Measured: 22.1s / 122 tests across 9 net-new files (the tenth was already on the
gate). Both invocation sites updated, GATE_SUITES extended, `teaching` removed
from UNREACHABLE_BASELINE in the same edit, ratchet lowered <= 15 to <= 14 —
the both-directions discipline that has caught this author twice.

A HOLLOW SPOT IN PIN 3 ITSELF, found by sabotage-testing the promotion rather
than trusting it. _suites_invoked_by_gate_tiers() takes the UNION of
scripts/hooks/pre-push and scripts/ci/local-ci.sh, so a suite present in either
satisfies it. Removing `teaching` from local-ci.sh alone PASSED SILENTLY: the
two scripts could drift from each other invisibly while local-ci.sh states in
prose that it runs "the same four steps as scripts/hooks/pre-push". A developer
running one would get coverage the other advertises and does not deliver.

Closed by test_the_two_gate_scripts_invoke_the_same_suites, scoped to the
numbered (n/m) gate steps — a discriminator both files already use — so it reads
the gate tier without parsing shell control flow. The identical sabotage now
fires. Verified in three directions: removed from one script (fires, and did not
before), removed from both (the pre-existing assertion fires), restored (green).

THIS IS NOT THE CI-PARITY PIN N-9 KILLED, and the distinction is recorded at the
assertion site so it cannot be re-litigated a third time. That one asserted
TEST_SUITES["smoke"] == smoke.yml and was reverted in 50fa287d because
AGENTS.md:280 makes GitHub Actions billing-locked dead signals — parity with a
dead signal buys nothing. Both scripts here are LIVE local merge bars.

Gate is now four steps: smoke + warmed_session + deductive + teaching.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wcw2pnMBwyvmNyQg4uPEt4

docs/assessment/30-gap-register.md

@@ -112,6 +112,8 @@ Pin 1 exists because the two gates have **already** drifted (see H-12): the loca
 
 **What pin 3 deliberately does not do — and what is now Shay's to decide:** *which* of the 19 belong on the gate. That costs gate time and is a real decision. `teaching` is the obvious first candidate: it just shipped a red ratification pin that no gate ran. Recorded as the open half of this entry.
 
+**CLOSED 2026-07-28 — `teaching` promoted (PR-4b), gap 15 → 14, the first turn by promotion rather than deletion. Sabotage-testing that promotion also found pin 3's own union blind spot (the two gate scripts could drift from each other undetected) and closed it.**
+
 **The gap is 15, not 19, and the open half is now a scheduled plan item.** PR-3b (Wave 1, per R-7) deleted four aliases with zero `--suite` references anywhere — `refusal`, `margin`, `rotor`, `inner-loop` — taking the curated set 21 → 17 and the unreachable set **19 → 15, by deletion**, which costs no gate time at all. That is the ratchet turning the way it is meant to. The remaining fifteen are each named in AGENTS.md, CLI help, a README or an ADR, so deletion is not available for them and promotion is the only way the number falls further. The decision was a parenthetical in this entry and in PR-4 — which is precisely how fifteen suites came to be unreachable without anyone choosing it — so it is now **PR-4b** in `50-execution-plan.md` §4, sized, recommended (`teaching` alone, first), and prepared so ruling costs one word.
 
 *One real gap the entry was circling, now fixed:* the parity pin lived in `fast`, which the pre-push gate does not run — **the pin guarding the gate did not run on the gate**. It is now in `smoke`.

docs/assessment/50-execution-plan.md

@@ -284,7 +284,17 @@ Promoted out of a parenthetical in PR-4, because a parenthetical is how fifteen
 
 **Explicitly not recommended:** promoting all fifteen. That is the ceremony failure this plan rejected twice already (749 exclusion reasons; nineteen justifications in one sitting). Nor is deletion the default — PR-3b deleted the four with *zero* references anywhere; the remaining fifteen are all named in AGENTS.md, CLI help, READMEs or ADRs, so deleting them breaks documented commands.
 
-**Verification:** `teaching` invoked from `scripts/hooks/pre-push` and `local-ci.sh --tier gate`, `GATE_SUITES` updated, `teaching` removed from `UNREACHABLE_BASELINE` in the same edit (`test_suite_reachability.py` fails in both directions if either half is missed — it has already caught this author twice), and the ratchet's `<= 15` lowered to `<= 14`. The added gate seconds reported as a number in the PR body.
+**LANDED 2026-07-28 — and the recommendation was re-derived rather than executed, because its original justification had been spent.**
+
+This item recommended `teaching` on the strength of G-22: `test_ratification_ceremony.py` was red on `main`, in `teaching`, run by no gate. **That hole was already closed a different way** — G-22's fix promoted that single file into `smoke`. So `teaching` would have added **9 net-new files, none with a demonstrated escape**, and promoting on a spent justification is the failure this arc exists to close.
+
+**Promoted on a stronger, doctrinal basis instead.** `tests/test_epistemic_invariants.py` enforces AGENTS.md's *"Teaching and mutation safety"* non-negotiables in substance — INV-22/INV-23, that an unmarked proposal and an unmarked reviewed example default to **SPECULATIVE**, that the status enum has exactly its four positions, that pack-mutation proposals carry no hardening flag. Durable corpus mutation is the one irreversible thing in CORE, and the suite guarding it ran on no pre-merge gate. That justification does not depend on an incident having happened yet, which makes it the better one.
+
+**Measured: 22.1s for 9 net-new files** (the tenth, `test_ratification_ceremony.py`, was already on the gate). Both invocation sites updated, `GATE_SUITES` extended, `teaching` removed from `UNREACHABLE_BASELINE` in the same edit, ratchet lowered `<= 15` → `<= 14`. **The reachability gap turned by promotion for the first time** — PR-3b's earlier turn was by deletion, which is free; this one costs, which is why it is a decision.
+
+**A hollow spot in pin 3 itself, found by sabotage-testing the promotion.** `_suites_invoked_by_gate_tiers()` takes the **union** of `pre-push` and `local-ci.sh`, so a suite present in either satisfies it. Removing `teaching` from `local-ci.sh` alone **passed silently** — the two scripts could drift from each other invisibly, while `local-ci.sh` states in prose that it runs "the same four steps as `scripts/hooks/pre-push`." Closed by `test_the_two_gate_scripts_invoke_the_same_suites`, scoped to the numbered `(n/m)` gate steps so it reads the gate tier without parsing shell control flow. The identical sabotage now fires.
+
+**This is not the CI-parity pin N-9 killed**, and the distinction is recorded at the assertion site so it is not re-litigated: that one asserted parity with `smoke.yml`, which `AGENTS.md:280` makes a billing-locked dead signal. Both scripts here are live local merge bars, and a developer running either is entitled to the coverage the other advertises.
 
 ### PR-5 · The flag-default register · **M** · G-8/H-6
 `docs/specs/flag_register.md`: all **28** default-off flags plus the 4 default-on — current default, deliberate-posture vs accumulated-hesitancy, **what evidence flips it**, and named profiles (one-shot / eval / continuous-life) per R-4. Declared in the table, not the call site (ADR-0263 Rule 5). A pin asserts the register lists exactly the flags `core/config.py` defines — so a new flag cannot land unregistered. Includes the F-6 accrual resolution from R-3 and the N-6 docstring correction.

scripts/ci/local-ci.sh

@@ -54,7 +54,7 @@ usage: local-ci.sh [--tier TIER] [--allow-interpreter-fallback] [--list] [-- ARG
 
 tiers:
   smoke   smoke suite only                     (~1 min)  = CI smoke.yml parity
-  gate    smoke + warmed_session + deductive   (~2 min)  = the pre-push gate
+  gate    smoke + warmed_session + deductive + teaching  (~9 min) = the pre-push gate
   full    the entire tests/ tree, parallel     (~10 min) = the merge bar
 
 options:
@@ -165,10 +165,11 @@ case "${TIER}" in
         step "smoke suite" suite smoke
         ;;
     gate)
-        echo "[local-ci] tier=gate — same three steps as scripts/hooks/pre-push"
+        echo "[local-ci] tier=gate — same four steps as scripts/hooks/pre-push"
-        step "(1/3) smoke suite" suite smoke
+        step "(1/4) smoke suite" suite smoke
-        step "(2/3) warmed_session lane" pytest_ tests/test_warmed_session_lane.py -q --no-header
+        step "(2/4) warmed_session lane" pytest_ tests/test_warmed_session_lane.py -q --no-header
-        step "(3/3) deductive suite" suite deductive
+        step "(3/4) deductive suite" suite deductive
+        step "(4/4) teaching suite" suite teaching
         ;;
     full)
         jobs="$(nproc 2>/dev/null || echo 4)"

scripts/hooks/pre-push

@@ -53,23 +53,23 @@ if [ "${read_any}" -eq 1 ] && [ "${only_deletes}" -eq 1 ]; then
     exit 0
 fi
 
-echo "[pre-push] CORE local-first gate: smoke + warmed_session lane pin + deductive"
+echo "[pre-push] CORE local-first gate: smoke + warmed_session lane pin + deductive + teaching"
 
-echo "[pre-push] (1/3) smoke suite (core test --suite smoke) ..."
+echo "[pre-push] (1/4) smoke suite (core test --suite smoke) ..."
 if ! uv run core test --suite smoke -q; then
     echo "[pre-push] BLOCKED: smoke suite failed. Fix before pushing." >&2
     echo "[pre-push] (emergency bypass, discouraged: git push --no-verify)" >&2
     exit 1
 fi
 
-echo "[pre-push] (2/3) warmed_session consistency lane ..."
+echo "[pre-push] (2/4) warmed_session consistency lane ..."
 if ! uv run python -m pytest tests/test_warmed_session_lane.py -q; then
     echo "[pre-push] BLOCKED: warmed_session lane failed (T13 regression class)." >&2
     echo "[pre-push] (emergency bypass, discouraged: git push --no-verify)" >&2
     exit 1
 fi
 
-echo "[pre-push] (3/3) deductive suite (core test --suite deductive) ..."
+echo "[pre-push] (3/4) deductive suite (core test --suite deductive) ..."
 if ! uv run core test --suite deductive -q; then
     echo "[pre-push] BLOCKED: deductive suite failed — deduction serving is a" >&2
     echo "[pre-push]          LIVE capability (ADR-0256), not an experiment." >&2
@@ -77,5 +77,16 @@ if ! uv run core test --suite deductive -q; then
     exit 1
 fi
 
+echo "[pre-push] (4/4) teaching suite (core test --suite teaching) ..."
+if ! uv run core test --suite teaching -q; then
+    echo "[pre-push] BLOCKED: teaching suite failed — durable mutation is governed by" >&2
+    echo "[pre-push]          AGENTS.md non-negotiables (INV-21..INV-30). An unmarked" >&2
+    echo "[pre-push]          proposal or reviewed example defaulting to anything but" >&2
+    echo "[pre-push]          SPECULATIVE is a corpus-truth defect, and corpus mutation" >&2
+    echo "[pre-push]          is the one thing here that is irreversible." >&2
+    echo "[pre-push] (emergency bypass, discouraged: git push --no-verify)" >&2
+    exit 1
+fi
+
 echo "[pre-push] gate PASSED — push proceeding."
 exit 0

tests/test_suite_reachability.py

@@ -45,13 +45,14 @@ _ROOT = Path(__file__).resolve().parents[1]
 #: that silently matched nothing would make this pin vacuous — the exact
 #: failure mode (N-1) that has produced three false conclusions in this repo.
 #: So the set is declared here and *verified* against the shell below.
-GATE_SUITES: frozenset[str] = frozenset({"smoke", "deductive"})
+GATE_SUITES: frozenset[str] = frozenset({"smoke", "deductive", "teaching"})
 
 #: Curated suites with members and no gate caller.
-#: 19 at first measurement; **15 after PR-3b (Wave 1) deleted four unreferenced
+#: 19 at first measurement; **15 after PR-3b (Wave 1)** deleted four unreferenced
-#: per-phase aliases** — `refusal`, `margin`, `rotor`, `inner-loop`. That is the
+#: per-phase aliases — `refusal`, `margin`, `rotor`, `inner-loop` — the ratchet
-#: ratchet turning the way it is supposed to: the gap shrank by deletion rather
+#: turning by deletion, which costs no gate time at all. **14 after PR-4b promoted
-#: than by promotion, which costs no gate time at all.
+#: `teaching` onto the gate (2026-07-28)** — the first time it turned by PROMOTION,
+#: which is the direction that actually buys coverage and the direction that costs.
 #: This list may only SHRINK. A suite leaves it by being invoked from a gate
 #: tier — not by being deleted from this file.
 UNREACHABLE_BASELINE: frozenset[str] = frozenset(
@@ -70,7 +71,6 @@ UNREACHABLE_BASELINE: frozenset[str] = frozenset(
         "pulse",
         "runtime",
         "sensorium",
-        "teaching",
     }
 )
 
@@ -164,14 +164,16 @@ def test_baseline_has_no_stale_entries() -> None:
 def test_recorded_gate_gap_matches_the_measurement() -> None:
     """Pin the number so the gap moving is a reviewed decision, not a drift.
 
-    15 of 17, after PR-3b. (19 of 21 before it.) Shrinking this is the point — it shrinks by putting
+    14 of 17, after PR-4b. (15 after PR-3b; 19 of 21 before it.) Shrinking this is
-    a suite on the gate, which costs gate time and is therefore a decision
+    the point, and it shrinks two ways: by deleting a suite nobody calls (free) or by
-    someone has to make deliberately.
+    putting one on the gate (costs gate time on every push, and is therefore a
+    decision someone has to make deliberately). PR-4b was the first of the second
+    kind — `teaching`, at a measured 22.1s for 9 net-new files.
     """
-    assert len(UNREACHABLE_BASELINE) <= 15, (
+    assert len(UNREACHABLE_BASELINE) <= 14, (
-        f"gate-unreachable suites grew to {len(UNREACHABLE_BASELINE)} (was 15; "
+        f"gate-unreachable suites grew to {len(UNREACHABLE_BASELINE)} (was 14; 15 "
-        "19 before PR-3b deleted four unreferenced aliases). The ratchet only "
+        "before PR-4b promoted `teaching`, 19 before PR-3b deleted four unreferenced "
-        "turns one way."
+        "aliases). The ratchet only turns one way."
     )
 
 
@@ -182,3 +184,47 @@ def test_the_two_gate_suites_actually_exist() -> None:
     for name in GATE_SUITES:
         assert name in suites, f"gate tier invokes undefined suite {name!r}"
         assert suites[name], f"gate suite {name!r} is empty"
+
+
+#: A gate step in either script, e.g. ``(4/4) teaching suite ... suite teaching``.
+#: The ``(n/m)`` prefix is the discriminator: ``local-ci.sh`` defines other tiers
+#: whose steps are unnumbered, so this scopes the parse to the gate tier without
+#: needing to parse shell control flow.
+_GATE_STEP_RE = re.compile(r"\(\d+/\d+\)[^\n]*?\bsuite[= ]+([a-z0-9-]+)")
+
+
+def _gate_suites_in(rel: str) -> set[str]:
+    text = (_ROOT / rel).read_text(encoding="utf-8")
+    return set(_GATE_STEP_RE.findall(text))
+
+
+def test_the_two_gate_scripts_invoke_the_same_suites() -> None:
+    """``pre-push`` and ``local-ci.sh --tier gate`` must not drift apart.
+
+    Both scripts state that they are the same gate — ``local-ci.sh`` says so in
+    prose ("same four steps as scripts/hooks/pre-push"). Two records asserting the
+    same thing is the shape that goes stale, and until this pin the drift was
+    invisible: :func:`test_declared_gate_suites_match_the_shell` takes the UNION of
+    the two files, so a suite present in either satisfies it. Removing `teaching`
+    from ``local-ci.sh`` alone was sabotage-tested and passed — which is how this
+    gap was found.
+
+    **This is NOT the CI-parity pin that N-9 correctly killed.** That one asserted
+    ``TEST_SUITES["smoke"] == smoke.yml`` and was reverted in ``50fa287d`` because
+    ``AGENTS.md:280`` makes GitHub Actions billing-locked dead signals — parity with
+    a dead signal buys nothing. Both scripts here are live local merge bars, and a
+    developer running one is entitled to the coverage the other advertises.
+    """
+    hook = _gate_suites_in("scripts/hooks/pre-push")
+    local = _gate_suites_in("scripts/ci/local-ci.sh")
+    assert hook, "parsed no numbered gate steps from scripts/hooks/pre-push"
+    assert local, "parsed no numbered gate steps from scripts/ci/local-ci.sh --tier gate"
+    assert hook == local, (
+        "the two gate scripts have drifted — a push through the hook and a run of "
+        "local-ci.sh --tier gate would cover different suites:\n"
+        f"  pre-push only: {sorted(hook - local)}\n"
+        f"  local-ci only: {sorted(local - hook)}"
+    )
+    assert hook <= set(GATE_SUITES), (
+        f"the scripts invoke suites GATE_SUITES does not declare: {sorted(hook - set(GATE_SUITES))}"
+    )