Alert Triage: environment #2, built end to end by the pipeline

The first environment shipped through .claude/workflows/new-environment.js:
specification, three adversarial reviews (all 'fixable', none fatal), the
Python environment, the TypeScript port, captured rollouts, and the demo page.
Eleven agents, no errors.

The proof that the platform scales is one line long. Alert Triage has a
completely different shape from Word Five — JSON actions, priced lookups, an
analyst screen instead of a grid — and the only change under
src/components/demo/ is a comment edit, because the isolation lint refused the
word "wordle" there. Zero shell code changed. 415 contract checks now pass
against two demos, up from 206 against one.

The environment is honest by construction. Every alert is synthetic, generated
from the seed, and the banner saying so sits inside the board surface. Two of
the eleven scenario templates are hidden-suspicious: generated by the same code
as their benign twin with the signal overlaid only in lookup data, so the free
screen is identically distributed and a screen-only policy STRUCTURALLY cannot
tell them apart. The probe ladder measures it: `fast` catches 0.0 of hidden
seeds. That is the counterweight made real rather than asserted.

Twelve policies, thirteen ladder assertions, a genuine three-way trade:

  fast      0.846   wins hours (0.85), misses every hidden case
  targeted  0.894   wins the shipped total
  thorough  0.820   wins evidence (1.00), spends 2.9 hours

None dominates. 92 Python tests, 35 TypeScript tests, 65 fixtures replaying at
delta 0, and conformance gated on world + scorer + protocol so the browser shows
the same alert for ?seed= that Python generated.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019mt6sHQHEnEYrJZvoMCJSB

envs/build_manifest.py

@@ -13,11 +13,18 @@ from pathlib import Path
 
 TRACES = Path(__file__).parent.parent / "public" / "traces"
 
+# Arm ids are shared across tasksets where they mean the same thing (the two
+# model arms); the generated arms are per-taskset and simply do not collide.
 ARMS = {
     "base-off": ("Out of the box", "recorded"),
     "base-on": ("Allowed to think", "intervened"),
+    # wordle
     "solver": ("Best-known play", "generated"),
     "cautious": ("Never wastes a guess", "generated"),
+    # alert-triage: the shipped scripted analysts from alert_triage/policies.py
+    "fast": ("Reads the screen", "generated"),
+    "targeted": ("Checks the hidden tells", "generated"),
+    "thorough": ("Runs the full procedure", "generated"),
 }
 
 # What was done to the run, for arms that had something done to them. Required
@@ -27,7 +34,7 @@ INTERVENTIONS = {
     "base-on": "Same model, same seeds, sampled with thinking enabled. No training, no fine-tuning.",
 }
 
-ORDER = ["base-off", "base-on", "solver", "cautious"]
+ORDER = ["base-off", "base-on", "solver", "cautious", "fast", "targeted", "thorough"]
 
 
 def main() -> int:
@@ -67,11 +74,13 @@ def main() -> int:
             by_arm.setdefault(r["id"].rsplit("-s", 1)[0], []).append(r)
         print(f"{slug}: {len(runs)} runs")
         for arm, group in by_arm.items():
-            solved = 0
+            outcomes: dict[str, int] = {}
             for r in group:
                 data = json.loads((TRACES.parent / r["path"].lstrip("/")).read_text())
-                solved += 1 if data["outcome"] == "solved" else 0
-            print(f"  {arm:<10} {len(group)} runs, solved {solved}/{len(group)}")
+                outcomes[data["outcome"]] = outcomes.get(data["outcome"], 0) + 1
+            solved = outcomes.get("solved", 0)
+            rest = ", ".join(f"{k} {v}" for k, v in sorted(outcomes.items()) if k != "solved")
+            print(f"  {arm:<10} {len(group)} runs, solved {solved}/{len(group)}" + (f"  ({rest})" if rest else ""))
     return 0