test(gates): refuse to run the 45–55 suite when ajv is unresolvable (it reported 4 gate-53 defects that do not exist)

[rubenvdlinde]

What this fixes

test_gate_45_to_55_acceptance.sh reports four gate-53 defects that do not exist when ajv is not resolvable — and one of its arms passes for the wrong reason while doing so.

Evidence

Fresh clone of main (c51a225, git rev-list --left-right --count origin/main...HEAD = 0 0), 0-byte stderr on both runs.

Without NODE_PATH — 4 failures:

FAIL — gate-53 failed without naming EventRoster
FAIL — gate-53 accepts removing the component and its registry entry together → expected 'PASS', got 'FAIL'
FAIL — gate-53 leaves a PRE-EXISTING orphan advisory (WARN), not blocking → expected 'PASS', got 'FAIL'
FAIL — gate-53 swallowed the pre-existing orphan entirely

With NODE_PATH set — ALL GREEN, exit 0.

So gate-53 is correct and the suite is not stale. Both helpers behave properly in isolation: check_manifest_crossref.js on the D3 fixture exits 0 and emits the expected WARN for the pre-existing orphan. The gate fails closed on [gate-53] ... ajv not resolvable from <lib> — refusing to run fail-open, which is the right behaviour — but it makes every downstream assertion describe a gate that never ran.

This was previously reported upstream as "the suite expects blocking behaviour for pre-existing orphans that #250/#260/#280 deliberately made advisory". That reading is wrong in both halves: the suite expects advisory, and the shipped gate delivers advisory.

The part worth keeping

🔑 Arm D1 still printed ok. It expects FAIL and got FAIL — because gate-53 failed closed, not because it detected the orphan. A false pass inside the suite built to catch false passes.

That is why this aborts (exit 2, zero verdict lines) rather than warning and continuing. A wiring fault must not be able to produce a verdict at all.

The check itself

⚠️ node -e "require('ajv')" is not a valid check on its own — Node resolves upward from the cwd, so it can succeed against a node_modules belonging to an ancestor directory rather than to the gates package. (Measured the same day: an agent's ajv check passed for exactly this reason.)

So ajv is resolved from the helpers' own directory — the same place the runner resolves it from — and the absolute path is printed as the evidence:

preflight — ajv resolves to /…/node_modules/ajv/dist/ajv.js

The path is the evidence; the exit code cannot tell you which ajv answered.

Proven both directions

arm	result
ajv absent	exit 2, 0 verdict lines emitted, wiring message on stderr naming the fix
ajv present	exit 0, ALL GREEN, empty stderr

No gate logic changed. No assertion relaxed. One file touched.