diff --git a/tests/adr-612-collision-characterization.test.cjs b/tests/adr-612-collision-characterization.test.cjs
new file mode 100644
index 000000000..a6e317829
--- /dev/null
+++ b/tests/adr-612-collision-characterization.test.cjs
@@ -0,0 +1,129 @@
+'use strict';
+
+/**
+ * Characterization test for the M-NN phase-ID collision — the PR-0 anchor
+ * requested in the #612 (bracket phase-ID convention) approval.
+ *
+ * See docs/adr/612-bracket-phase-id-convention.md, Decision 3.
+ *
+ * This test captures the CURRENT behavior of `normalizePhaseName`
+ * (src/phase-id.cts:66 → gsd-core/bin/lib/phase-id.cjs). It is NOT a bug the
+ * PR fixes — PR-0 is zero behavior change. It "proves the defect first":
+ * a fully-specified milestone/phase/subphase/plan token silently collapses to
+ * a bare two-digit integer under the milestone-prefixed (M-NN) grammar, because
+ * the hyphen is overloaded (milestone↔phase in one position, phase↔plan in
+ * another) and the anchored milestone regex cannot consume a trailing plan
+ * hyphen that follows a subphase dot.
+ *
+ * Mechanism (both regexes are in normalizePhaseName):
+ *   - Anchored milestone regex (src/phase-id.cts:71):
+ *       /^(\d+)((?:-\d+)+)([A-Z]?(?:\.\d+)*)$/i
+ *     On "2-01.02-01" it captures 2, then "-01", then ".02"; the trailing "-01"
+ *     (the plan) is not expressible as (?:\.\d+)* so the `$` anchor fails and
+ *     the match is rejected.
+ *   - Unanchored numeric fallback (src/phase-id.cts:79):
+ *       /^(\d+)([A-Z])?((?:\.\d+)*)/i
+ *     With no `$`, it matches only the leading "2" and zero-pads to "02";
+ *     everything after the leading integer is silently dropped.
+ *
+ * These assertions are the byte-identical current-behavior contract. They are
+ * expected to be REPLACED — not merely to keep passing — when PR-1 lands the
+ * bracket grammar (parsePhaseId/renderPhaseId/toDir) behind
+ * `phase_id_convention: 'bracket'`, at which point a full bracket token
+ * (`GSD.02-05.03-01`) parses to exactly one tuple and the collapse no longer
+ * occurs on the bracket path. Until then, this file documents the ambiguity.
+ *
+ * All assertions are BEHAVIORAL (call the exported function, assert its typed
+ * output). No source-grep.
+ */
+
+const { test, describe } = require('node:test');
+const assert = require('node:assert/strict');
+
+const phaseId = require('../gsd-core/bin/lib/phase-id.cjs');
+const { normalizePhaseName } = phaseId;
+const fc = require('fast-check');
+
+describe('ADR-612 PR-0 — M-NN phase-ID collision (current-behavior characterization)', () => {
+  test('the full 4-tuple token "2-01.02-01" collapses to bare "02" (the anchor defect)', () => {
+    // Intended meaning: milestone 2, phase 01, subphase 02, plan 01.
+    // The anchored milestone regex rejects the trailing plan hyphen after ".02",
+    // so control falls to the unanchored fallback, which keeps only the leading
+    // "2" and zero-pads it. A fully-specified identity collapses to bare "02".
+    assert.equal(normalizePhaseName('2-01.02-01'), '02');
+
+    // It did NOT round-trip the input: there is no single deterministic 4-tuple
+    // a parser can return under M-NN without out-of-band convention metadata.
+    assert.notEqual(normalizePhaseName('2-01.02-01'), '02-01.02-01');
+  });
+
+  test('the collapse is a general class, not a single case: "10-02.03-04" collapses to "10"', () => {
+    // Same failure mode: the trailing "-04" (plan) after ".03" (subphase) kills
+    // the anchored milestone match; the fallback keeps only "10".
+    assert.equal(normalizePhaseName('10-02.03-04'), '10');
+  });
+
+  test('the collapse appears ONLY when all four dimensions coexist (defect boundary)', () => {
+    // milestone + phase — anchored milestone regex matches, no collapse.
+    assert.equal(normalizePhaseName('2-01'), '02-01');
+    // + subphase (dot tail) — anchored regex still matches, no collapse.
+    assert.equal(normalizePhaseName('2-01.02'), '02-01.02');
+    // + plan (a hyphen after the subphase dot) — anchor fails, token collapses.
+    assert.equal(normalizePhaseName('2-01.02-01'), '02');
+  });
+
+  test('"02-04" is a cross-subsystem semantic ambiguity, not a parse collapse', () => {
+    // normalizePhaseName reads "02-04" as milestone 02 / phase 04 and returns it
+    // intact — this token does NOT collapse.
+    assert.equal(normalizePhaseName('02-04'), '02-04');
+    // The SAME string is the phase-02 / plan-04 token in plan-file notation
+    // ({padded_phase}-{NN}-PLAN.md). Both readings are valid productions; the
+    // token alone cannot tell a resolver which subsystem is asking. This is a
+    // genuine ambiguity but a different class than the collapse above.
+  });
+});
+
+/**
+ * Property generalization of the collapse (requested by the PR-2181 review:
+ * "covers the collapse with literal examples only; no fast-check property test
+ * generalizing the collision class"). The literal cases above assert two points
+ * on the class; this proves the collapse is the CLASS behavior, not cherry-
+ * picked examples: for EVERY fully-specified 4-dimension M-NN token
+ * `${m}-${pp}.${ss}-${ll}` (m: 1-99 integer, no leading zeros; pp, ss, ll:
+ * zero-padded 2-digit 01-99), normalizePhaseName collapses it to the bare
+ * zero-padded milestone `String(m).padStart(2, '0')`.
+ *
+ * Still GREEN-ONLY: it characterizes CURRENT behavior exactly (same contract as
+ * the literal cases). It is expected to be REPLACED, not merely kept passing,
+ * when PR-1 lands the bracket grammar (parsePhaseId/renderPhaseId/toDir) behind
+ * `phase_id_convention: 'bracket'`, at which point a full bracket token parses
+ * to exactly one tuple and the collapse no longer occurs on the bracket path.
+ *
+ * numRuns: the house default for the fast-check block in tests/phase-id.test.cjs
+ * is the implicit 100; this sets 1000 for denser sampling of the ~96M-combination
+ * class (99^4). The property was additionally verified EXHAUSTIVELY offline over
+ * all 96,059,601 tokens of the domain (0 failures) — it holds totally, not just
+ * on the sampled draws, so no domain narrowing was needed.
+ */
+describe('ADR-612 PR-0 — M-NN collapse is the class, not the examples (fast-check property)', () => {
+  const pad2 = (n) => String(n).padStart(2, '0');
+
+  test('every 4-tuple `${m}-${pp}.${ss}-${ll}` collapses to bare `String(m).padStart(2,"0")`', () => {
+    fc.assert(
+      fc.property(
+        fc.integer({ min: 1, max: 99 }), // milestone m — leading integer, no leading zeros
+        fc.integer({ min: 1, max: 99 }), // phase pp     — rendered zero-padded 2-digit
+        fc.integer({ min: 1, max: 99 }), // subphase ss  — rendered zero-padded 2-digit
+        fc.integer({ min: 1, max: 99 }), // plan ll      — rendered zero-padded 2-digit
+        (m, pp, ss, ll) => {
+          // The trailing plan hyphen after the subphase dot kills the anchored
+          // milestone regex (src/phase-id.cts:71); the unanchored fallback
+          // (src/phase-id.cts:79) keeps only the leading integer, zero-padded.
+          const token = `${m}-${pad2(pp)}.${pad2(ss)}-${pad2(ll)}`;
+          return normalizePhaseName(token) === String(m).padStart(2, '0');
+        },
+      ),
+      { numRuns: 1000 },
+    );
+  });
+});