test: add multi-finding batching regression test (periodic tier)

Adds a periodic-tier E2E that catches the May 2026 transcript bug shape the existing single-finding gate-tier floor test cannot detect: a model that fires one AskUserQuestion and then batches the remaining findings into a single "## Decisions to confirm" plan write + ExitPlanMode. Why a separate test from skill-e2e-plan-eng-finding-floor: the gate-tier floor (runPlanSkillFloorCheck) exits on the first AUQ render and returns success, so a once-then-batch model would pass it trivially. This test uses runPlanSkillCounting at periodic tier with N-AUQ tracking and asserts >= 3 distinct review-phase AUQs on a 4-finding seeded plan. - test/fixtures/forcing-finding-seeds.ts: FORCING_BATCHING_ENG fixture (4 distinct non-trivial findings spread across Architecture, Code Quality, Tests, Performance — mirrors the D1-D4 transcript shape) - test/skill-e2e-plan-eng-multi-finding-batching.test.ts: new test - test/helpers/touchfiles.ts: registered in BOTH E2E_TOUCHFILES and E2E_TIERS (touchfiles.test.ts asserts exact equality) Test will fail on baseline today because today's model uses the preamble fallback to batch findings; passes after the architectural fix lands in a follow-up commit. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-06-19 16:20:09 +02:00 · 2026-05-08 15:25:17 -07:00
parent 443bde054c
commit 4c2bcf5c17
3 changed files with 143 additions and 0 deletions
@@ -0,0 +1,96 @@
+/**
+ * /plan-eng-review multi-finding batching regression (periodic, paid, real-PTY).
+ *
+ * Catches the specific shape of the May 2026 transcript bug that the
+ * single-finding gate-tier floor test cannot detect: a model that fires
+ * one AskUserQuestion and then batches the remaining findings into a
+ * single "## Decisions to confirm" plan write + ExitPlanMode.
+ *
+ * Why a separate test from skill-e2e-plan-eng-finding-floor:
+ *   - The gate-tier floor (runPlanSkillFloorCheck) exits on the first AUQ
+ *     render and returns success. A model that fires once-then-batches
+ *     would pass that test trivially.
+ *   - This test uses runPlanSkillCounting at periodic tier (~25 min budget,
+ *     N-AUQ tracking, ceiling-bounded retries) to actually count distinct
+ *     review-phase AUQs and assert the model fires one per finding.
+ *
+ * Why a separate test from skill-e2e-plan-eng-finding-count (the existing
+ * 5-finding count test):
+ *   - The fixture here mirrors the D1-D4 transcript shape (4 findings) and
+ *     the floor matches that exact threshold (3, the [N-1] tolerance band).
+ *     This is the tightest regression test for the original bug class —
+ *     not a band-around-N test, but a "did the agent batch?" test.
+ *
+ * Tier: periodic (~25 min, ~$5/run). Sequential by default.
+ */
+
+import { describe, test } from 'bun:test';
+import * as fs from 'node:fs';
+import {
+  runPlanSkillCounting,
+  engStep0Boundary,
+} from './helpers/claude-pty-runner';
+import { FORCING_BATCHING_ENG } from './fixtures/forcing-finding-seeds';
+
+const shouldRun = !!process.env.EVALS && process.env.EVALS_TIER === 'periodic';
+const describeE2E = shouldRun ? describe : describe.skip;
+
+const N = 4;
+const FLOOR = N - 1; // 3 — agent must fire at least one AUQ per non-batched finding
+
+const PLAN_PATH = '/tmp/gstack-test-plan-eng-batching.md';
+
+describeE2E('/plan-eng-review multi-finding batching regression (periodic)', () => {
+  test(
+    `4-finding plan emits >= ${FLOOR} review-phase AskUserQuestions (no batching)`,
+    async () => {
+      try {
+        fs.rmSync(PLAN_PATH, { force: true });
+      } catch {
+        /* best-effort */
+      }
+
+      const obs = await runPlanSkillCounting({
+        skillName: 'plan-eng-review',
+        slashCommand: '/plan-eng-review',
+        followUpPrompt: FORCING_BATCHING_ENG,
+        isLastStep0AUQ: engStep0Boundary,
+        reviewCountCeiling: N + 3, // hard cap above floor + tolerance
+        cwd: process.cwd(),
+        timeoutMs: 1_500_000, // 25 min
+        env: { QUESTION_TUNING: 'false', EXPLAIN_LEVEL: 'default' },
+      });
+
+      try {
+        if (!['plan_ready', 'completion_summary', 'ceiling_reached'].includes(obs.outcome)) {
+          throw new Error(
+            `multi-finding batching test FAILED: outcome=${obs.outcome}\n` +
+              `step0=${obs.step0Count} review=${obs.reviewCount} elapsed=${obs.elapsedMs}ms\n` +
+              `--- evidence (last 3KB) ---\n${obs.evidence}`,
+          );
+        }
+        if (obs.reviewCount < FLOOR) {
+          throw new Error(
+            `BATCHING REGRESSION: reviewCount=${obs.reviewCount} < FLOOR=${FLOOR}.\n` +
+              `Agent surfaced fewer review-phase AUQs than findings — this is the\n` +
+              `May 2026 transcript bug shape: model batched multiple findings into\n` +
+              `a single plan write + ExitPlanMode instead of asking one per finding.\n` +
+              `Review-phase fingerprints:\n` +
+              obs.fingerprints
+                .filter((f) => !f.preReview)
+                .map((f) => `  - "${f.promptSnippet.slice(0, 80)}"`)
+                .join('\n') +
+              `\n--- evidence (last 3KB) ---\n${obs.evidence}`,
+          );
+        }
+      } finally {
+        try {
+          fs.rmSync(PLAN_PATH, { force: true });
+        } catch {
+          /* best-effort */
+        }
+      }
+    },
+    1_700_000,
+  );
+});