MaxGhenis · Oct 10, 2026 · Oct 8, 2026 · Oct 8, 2026 · Oct 8, 2026 · Oct 8, 2026
diff --git a/annotations/us_full_run_20260612_policyengine_4_16_1_populace/us_adjudications.json b/annotations/us_full_run_20260612_policyengine_4_16_1_populace/us_adjudications.json
diff --git a/annotations/us_full_run_20260612_policyengine_4_16_1_populace/us_audit_row_annotations.csv b/annotations/us_full_run_20260612_policyengine_4_16_1_populace/us_audit_row_annotations.csv
diff --git a/annotations/us_full_run_20260612_policyengine_4_16_1_populace/us_case_notes.csv b/annotations/us_full_run_20260612_policyengine_4_16_1_populace/us_case_notes.csv
diff --git a/...ions/us_full_run_20260612_policyengine_4_16_1_populace/us_case_reference_explanations.csv b/...ions/us_full_run_20260612_policyengine_4_16_1_populace/us_case_reference_explanations.csv
diff --git a/app/public/paper/policybench.pdf b/app/public/paper/policybench.pdf
diff --git a/app/public/paper/web/figures/positive_zero_scatter.png b/app/public/paper/web/figures/positive_zero_scatter.png
diff --git a/app/public/paper/web/index.html b/app/public/paper/web/index.html
diff --git a/app/src/components/ExclusionNote.tsx b/app/src/components/ExclusionNote.tsx
@@ -76,7 +76,7 @@ export default function ExclusionNote({
             <span className="font-[family-name:var(--font-mono)]">
               {formatValue(exclusion.frozenValue, isBinary, currencySymbol)}
             </span>{" "}
-            used the engine&rsquo;s projection.
+            used the engine&rsquo;s own figure.
           </p>
           <p className="mt-2 text-xs text-text-muted leading-relaxed">
             {exclusion.law ? `Law: ${exclusion.law}. ` : ""}

diff --git a/app/src/data.artifact.json b/app/src/data.artifact.json
@@ -1,9 +1,9 @@
 {
   "asset": "dashboard-data.json",
-  "bytes": 127225036,
+  "bytes": 128756688,
   "repo": "PolicyEngine/policybench",
-  "sha256": "aa34e5c9ea926848dc7af460a98f17021bc42dae54a8015c9988a85e79a5d462",
-  "tag": "dashboard-data-20261006",
-  "url": "https://github.com/PolicyEngine/policybench/releases/download/dashboard-data-20261006/dashboard-data.json",
+  "sha256": "f834478e65191f8426ef479506f9f6fe32c2e7107f30fdbf4e7ba0e6acba6410",
+  "tag": "dashboard-data-20261010",
+  "url": "https://github.com/PolicyEngine/policybench/releases/download/dashboard-data-20261010/dashboard-data.json",
   "version": 1
 }
diff --git a/app/src/data.versions.json b/app/src/data.versions.json
@@ -6,10 +6,10 @@
       "artifact": {
         "pointer": "live"
       },
-      "description": "Scored reference outputs from policyengine-us 2.15.17 (the 64 excluded outputs keep the values they were decided on: 52 from policyengine-us 1.755.4, 12 from 2.15.17, and the 19 that move on 2.15.17 were re-reviewed and stay excluded); same household facts and requested outputs for every model; answer transport and request shape per model as recorded in the serving-configuration table - 46 models",
+      "description": "Scored reference outputs from policyengine-us 2.38.6 (the 58 excluded outputs keep the values they were decided on: 38 from policyengine-us 1.755.4, 18 from 2.15.17, 2 from 2.38.6, and the 17 that move on 2.38.6 were re-reviewed and stay excluded); same household facts and requested outputs for every model; answer transport and request shape per model as recorded in the serving-configuration table - 47 models",
       "id": "1.1",
       "label": "1.1",
-      "snapshotLabel": "Snapshot 2026-09-30"
+      "snapshotLabel": "Snapshot 2026-10-10"
     },
     {
       "artifact": {

diff --git a/app/src/lib/predictionStatus.ts b/app/src/lib/predictionStatus.ts
@@ -84,7 +84,7 @@ const EXCLUDED_REFERENCE_BASIS: Record<string, string> = {
     "which assumes one reading of the unlisted input described above",
   reference_engine_defect: "which carries the engine defect described above",
   reference_law_published_after_freeze:
-    "which uses the engine's projection described above",
+    "which uses the engine's own 2026 figure described above",
 };
 
 /**

diff --git a/app/src/lib/referenceEngine.ts b/app/src/lib/referenceEngine.ts
@@ -15,8 +15,8 @@ import type { ReferenceExclusion } from "../types";
  * the frozen sidecar.
  */
 export const ENGINE_UPGRADE_RECHECK = {
-  engineVersion: "2.15.17",
-  rechecked: 19,
+  engineVersion: "2.38.6",
+  rechecked: 17,
 };
 
 function compareVersions(a: string, b: string): number {

diff --git a/app/src/lib/servingSensitivity.ts b/app/src/lib/servingSensitivity.ts
@@ -6,7 +6,8 @@
  * regardless); Claude Fable 5.1 rejects forced tool calls and answers as JSON,
  * reasoning in both runs, so its comparison is one of transport. Claude Opus
  * 5.5 and Claude Sonnet 5.5 reject forced calls the same way and have no auto
- * re-run. Scores are
+ * re-run. Claude Haiku 5.5 accepts forced calls, so its board row is a
+ * forced-tool row like Claude Opus 5's. Scores are
  * the pinned three-decimal measurements from sensitivity/data/*.json; round
  * only for display. Each entry's "would rank" is derived from the live board
  * rows at render time, never typed by hand.
@@ -35,37 +36,44 @@ const FORCED_TOOL =
 
 export const SERVING_SENSITIVITY: Record<string, ServingSensitivity> = {
   "claude-fable-5": {
-    autoExact: 92.817,
+    autoExact: 91.746,
     boardTreatment: `${FORCED_TOOL}, one output per request`,
     autoTreatment:
       "re-run with tool_choice: auto and the whole household in one request",
     thinkingSuppressedOnBoard: true,
     noteHref: SENSITIVITY_DOC_HREF,
   },
   "claude-opus-5": {
-    autoExact: 91.252,
+    autoExact: 90.456,
     boardTreatment: FORCED_TOOL,
     autoTreatment: "re-run with tool_choice: auto",
     thinkingSuppressedOnBoard: true,
     noteHref: SENSITIVITY_DOC_HREF,
   },
   "claude-sonnet-5": {
-    autoExact: 86.076,
+    autoExact: 84.953,
     boardTreatment: `${FORCED_TOOL}, one output per request`,
     autoTreatment:
       "re-run with tool_choice: auto and the whole household in one request",
     thinkingSuppressedOnBoard: true,
     noteHref: SENSITIVITY_DOC_HREF,
   },
   "claude-fable-5.1": {
-    autoExact: 93.037,
+    autoExact: 92.044,
     boardTreatment:
       "rejects forced tool calls, so its row answers as a JSON object and reasons at the provider default",
     autoTreatment:
       "re-run with the answer tool declared under tool_choice: auto (it reasons in both runs, so this compares transports)",
     thinkingSuppressedOnBoard: false,
     noteHref: "/notes/2026-09-01-claude-fable-5-1-added",
   },
+  "claude-haiku-5.5": {
+    autoExact: 89.239,
+    boardTreatment: FORCED_TOOL,
+    autoTreatment: "re-run with tool_choice: auto",
+    thinkingSuppressedOnBoard: true,
+    noteHref: SENSITIVITY_DOC_HREF,
+  },
 };
 
 export function servingSensitivityFor(

diff --git a/app/src/model-serving-config.json b/app/src/model-serving-config.json
@@ -13,7 +13,7 @@
   },
   "evidence_summary": {
     "registry": 29,
-    "run_state": 17
+    "run_state": 18
   },
   "models": {
     "claude-fable-5": {
@@ -108,6 +108,63 @@
       "shared_completion_budget_tokens": null,
       "tool_choice": "forced"
     },
+    "claude-haiku-5.5": {
+      "answer_contract": "tool",
+      "evidence": {
+        "compared_columns": [
+          "scenario_id",
+          "variable",
+          "prediction",
+          "explanation",
+          "raw_response",
+          "provider_resolved_model",
+          "prompt_tokens",
+          "completion_tokens",
+          "error"
+        ],
+        "fields": [
+          "answer_contract",
+          "chunk_size",
+          "completion_budget_ceiling",
+          "fingerprint_version",
+          "initial_completion_budget_tokens",
+          "max_repair_rounds",
+          "model_id",
+          "prompt_contract_version",
+          "request_timeout_seconds",
+          "thinking",
+          "tool_choice_mode"
+        ],
+        "frozen_rows_sha256": "4a7373791a5f9df6e4c775b8aade37ecd58df87bd8b99f238889f1afe744621c",
+        "kind": "run_state",
+        "rows_match": true,
+        "run": "haiku55-runs/haiku55",
+        "source_predictions_sha256": "27a5e48f44ad875f6d0901571b91c06c67045a7a6841f95a5ab56ecf0cfc8344",
+        "source_rows_sha256": "4a7373791a5f9df6e4c775b8aade37ecd58df87bd8b99f238889f1afe744621c",
+        "treatment_fingerprint": {
+          "answer_contract": "tool",
+          "chunk_size": null,
+          "completion_budget_ceiling": 128000,
+          "fingerprint_version": 3,
+          "initial_completion_budget_tokens": 16384,
+          "max_repair_rounds": 2,
+          "model_id": "claude-haiku-5-5",
+          "prompt_contract_version": "2026-08-09-v2-scoring-contract",
+          "request_timeout_seconds": 300,
+          "thinking": {
+            "mode": "provider_default"
+          },
+          "tool_choice_mode": "forced"
+        }
+      },
+      "provider_id": "claude-haiku-5-5",
+      "reasoning_setup": "provider default; 16,384-token shared budget",
+      "registry_derived": [],
+      "request_shape": "whole scenario",
+      "request_timeout_seconds": 300,
+      "shared_completion_budget_tokens": 16384,
+      "tool_choice": "forced"
+    },
     "claude-opus-4.7": {
       "answer_contract": "tool",
       "evidence": {
@@ -1528,7 +1585,7 @@
       "tool_choice": null
     }
   },
-  "registry_commit": "cddd01812ad6d6311502ef997f93634f70c28fda",
+  "registry_commit": "0562d77a9d263c14bb9c9fa93ba3c765fcff95a9",
   "sources": [
     "policybench.config.MODELS",
     "policybench.model_cards",

diff --git a/app/src/modelMeta.ts b/app/src/modelMeta.ts
@@ -13,6 +13,7 @@ export const MODEL_ORDER = [
   "claude-sonnet-5.5",
   "claude-sonnet-5",
   "claude-sonnet-4.6",
+  "claude-haiku-5.5",
   "claude-haiku-4.5",
   "grok-4.3",
   "grok-4.5",
@@ -60,6 +61,7 @@ export const MODEL_LABELS: Record<string, string> = {
   "claude-opus-5": "Claude Opus 5",
   "claude-opus-4.8": "Claude Opus 4.8",
   "claude-opus-4.7": "Claude Opus 4.7",
+  "claude-haiku-5.5": "Claude Haiku 5.5",
   "claude-haiku-4.5": "Claude Haiku 4.5",
   "claude-sonnet-5.5": "Claude Sonnet 5.5",
   "claude-sonnet-5": "Claude Sonnet 5",

diff --git a/app/src/notes/2026-10-10-claude-haiku-5-5-joins-the-board.json b/app/src/notes/2026-10-10-claude-haiku-5-5-joins-the-board.json
@@ -0,0 +1,90 @@
+{
+  "slug": "2026-10-10-claude-haiku-5-5-joins-the-board",
+  "date": "2026-10-10",
+  "title": "Claude Haiku 5.5 joins the board as the references move to policyengine-us 2.38.6",
+  "paragraphs": [
+    "Claude Haiku 5.5 joined the board on {releaseDate}, bringing it to {nModels} models. It gets {haikuExact}% of answers within $1, weighted by household impact, #{haikuRank} of {nModels} and {haikuGain} points above Claude Haiku 4.5 ({haiku45Exact}%, #{haiku45Rank}). GPT-6 Sol still leads at {solExact}%, ahead of Claude Opus 5.5 ({opusExact}%) and GPT-5.6 Sol ({sol56Exact}%).",
+    "Claude Haiku 5.5 cost ${haikuCost} a household, the least of any row with a cost above zero; Claude Haiku 4.5 cost ${haiku45Cost}. Unlike Claude Opus 5.5 and Claude Sonnet 5.5, its API accepts a forced tool call, so its row takes the board's forced answer tool. A probe with that request returned no thinking block, where a request without the forced tool did, so the row answers without extended thinking, as Claude Opus 5's does. Re-run with tool_choice: auto, it gets {haikuAutoExact}%, which would rank #{haikuAutoRank}.",
+    "The release also moves PolicyBench's references from policyengine-us {previousEngine} to {engineVersion}, a release that fixes the engine defects behind {restored:words} outputs. The previous release excluded {newlyScored:words} of them, and they return to scoring at the new version's values. The other {restoredRuled:words} are the October 6 defects below, which stay scored at the new version's values. Each of the {restored:words} lands within $1 of the corrected value PolicyBench's audits computed for it. Between them they take the upstream fixes for {restoredFixes}. PolicyBench still builds each reference from the stated facts and from law published before it froze the references on 2026-07-03, under the same publication conventions.",
+    "Two rulings of October 6 apply too. PolicyBench's reference adversary flags the scored outputs where many models, or several of the strongest, agree on an answer the reference does not give, and a judge that cannot yet see the engine's derivation works each one from primary law. Its findings led to rulings to exclude {adversaryOutputs:words} outputs. Four were engine defects, in Arizona, Ohio, Colorado and New York; the new version fixes them, so PolicyBench keeps scoring those four instead. The other four are the federal and state income tax of a Pennsylvania and a Missouri household whose dependent earns $45,000 of wages and must file their own return. The output definitions do not say whether the household's income tax includes that return, so PolicyBench stops scoring them. It also stops scoring {louisianaOutputs:words} Louisiana households' state income tax, whose reference rests on a 2026 standard deduction computed from published price indexes; Louisiana published the amount itself on September 28, after the freeze.",
+    "The new version counts Indiana county income tax in local income tax, and the prompt names no county, so PolicyBench stops scoring the local income tax of {indianaOutputs:words} Indiana households. It also counts Idaho's $10 permanent building fund tax in state income tax, which moves one household's reference from ${idahoBefore} to ${idahoAfter}. Every model is now scored on {scoredOutputs} of its {totalOutputs} requested outputs; PolicyBench excludes the other {excluded}.",
+    "Against the previous release, the exact rate of every one of the {incumbents} earlier models falls, by {fallMin} to {fallMax} points. The cause is the {newlyScored} restored outputs that the previous release did not score, which few models get right: half of them are answered within $1 by at most {newlyScoredHalfAtMost:words} of the {nModels} models, and {newlyScoredUnmatched:words} by none. Had they stayed excluded, every earlier model's rate would have risen, by {riseWithoutMin} to {riseWithoutMax} points."
+  ],
+  "facts": {
+    "releaseDate": "2026-10-10",
+    "nModels": 47,
+    "haikuExact": 79.6,
+    "haikuRank": 37,
+    "haiku45Exact": 74.8,
+    "haiku45Rank": 43,
+    "haikuGain": 4.8,
+    "solExact": 95.6,
+    "opusExact": 94.4,
+    "sol56Exact": 93.5,
+    "haikuCost": 0.0013,
+    "haiku45Cost": 0.0554,
+    "haikuAutoExact": 89.2,
+    "haikuAutoRank": 10,
+    "previousEngine": "2.15.17",
+    "engineVersion": "2.38.6",
+    "restored": 18,
+    "restoredRuled": 4,
+    "restoredFixes": "the IRA deduction's compensation limit and a dependent's contributions (policyengine-us #10032, #9801; three outputs), the IRA deduction's active-participant phase-out and California's itemized deduction conformity (policyengine-us #10032, #10034), New Jersey's worker unemployment and workforce contributions (policyengine-us #10031), Arizona's standard deduction indexing (policyengine-us #9928), California's itemized deduction conformity (policyengine-us #10034), Ohio's medical deduction for health insurance premiums (policyengine-us #10020), estate income in gross income (policyengine-us #10027, #9633; two outputs), Colorado's 2026 sales tax refund (policyengine-us #9946), the IRA deduction's active-participant phase-out (policyengine-us #10032; four outputs), New York's 2026 child and dependent care credit (policyengine-us #9948) and the IRA deduction's active-participant phase-out and estate income in gross income (policyengine-us #10032, #10027; two outputs)",
+    "adversaryOutputs": 8,
+    "louisianaOutputs": 2,
+    "indianaOutputs": 2,
+    "idahoBefore": "6,818.34",
+    "idahoAfter": "6,828.34",
+    "scoredOutputs": 1926,
+    "totalOutputs": 1984,
+    "excluded": 58,
+    "incumbents": 46,
+    "fallMin": 0.24,
+    "fallMax": 1.37,
+    "newlyScored": 14,
+    "newlyScoredHalfAtMost": 1,
+    "newlyScoredUnmatched": 4,
+    "riseWithoutMin": 0.39,
+    "riseWithoutMax": 0.71
+  },
+  "data": [
+    {
+      "label": "Dashboard data release",
+      "href": "https://github.com/PolicyEngine/policybench/releases/tag/dashboard-data-20261010"
+    },
+    {
+      "label": "Engine upgrade record",
+      "href": "https://github.com/PolicyEngine/policybench/blob/main/reference_audit/2026-10-09-engine-upgrade"
+    },
+    {
+      "label": "Reference sidecar with every change",
+      "href": "https://github.com/PolicyEngine/policybench/blob/main/paper/snapshot/20260501/runs/us_full_run_20260612_policyengine_4_16_1_populace/reference_outputs.csv.meta.json"
+    },
+    {
+      "label": "Exclusion record",
+      "href": "https://github.com/PolicyEngine/policybench/blob/main/paper/snapshot/20260501/runs/us_full_run_20260612_policyengine_4_16_1_populace/reference_exclusions.json"
+    },
+    {
+      "label": "Reference adversary's record",
+      "href": "https://github.com/PolicyEngine/policybench/blob/main/reference_audit/2026-10-05-reference-adversary/README.md"
+    },
+    {
+      "label": "Louisiana standard deduction audit",
+      "href": "https://github.com/PolicyEngine/policybench/blob/main/reference_audit/2026-10-05-louisiana/README.md"
+    },
+    {
+      "label": "Serving sensitivity note",
+      "href": "https://github.com/PolicyEngine/policybench/blob/main/sensitivity/claude-thinking-2026-08.md"
+    },
+    {
+      "label": "Claude Haiku 5.5 model page",
+      "href": "/model/claude-haiku-5.5"
+    },
+    {
+      "label": "Previous release note (October 6)",
+      "href": "/notes/2026-10-06-policybench-stops-scoring-eight-tax-outputs"
+    }
+  ],
+  "boardSnapshot": "2026-10-10",
+  "release": "dashboard-data-20261010"
+}
diff --git a/app/src/notes/index.ts b/app/src/notes/index.ts
@@ -1,3 +1,4 @@
+import claudeHaiku55JoinsTheBoard from "./2026-10-10-claude-haiku-5-5-joins-the-board.json";
 import claudeFable51Added from "./2026-09-01-claude-fable-5-1-added.json";
 import claudeSonnet55DebutsFifth from "./2026-09-29-claude-sonnet-5-5-debuts-fifth.json";
 import fiveSnapHouseholdsBbce from "./2026-10-05-five-snap-households-bbce.json";
@@ -29,6 +30,7 @@ export type PolicyBenchNote = {
 };
 
 export const notes = [
+  claudeHaiku55JoinsTheBoard,
   stopsScoringEightTaxOutputs,
   fiveSnapHouseholdsBbce,
   claudeSonnet55DebutsFifth,

diff --git a/app/src/paperSnapshot.json b/app/src/paperSnapshot.json
@@ -1,6 +1,6 @@
 {
-  "pdfVersion": "2cffc69cffde",
-  "responseWindow": "June 12 and September 29, 2026",
-  "snapshotDate": "2026-09-30",
-  "webVersion": "c6dfba0c5718"
+  "pdfVersion": "1607c71abaf3",
+  "responseWindow": "June 12 and October 8, 2026",
+  "snapshotDate": "2026-10-10",
+  "webVersion": "31cd4141373a"
 }
diff --git a/app/tests/auditUniverse.test.ts b/app/tests/auditUniverse.test.ts
@@ -18,16 +18,17 @@ describe("audit universe", () => {
     expect(AUDIT_SELECTION_RULE).toBe(
       "rows whose legacy threshold score is below 1",
     );
-    // Release 20261006: the counts tests/test_disclosures.py recomputes
-    // from the frozen annotations and payload (release 20260930's less the
-    // 368 scored rows on the eight outputs it excludes).
+    // Release 20261010: the counts tests/test_disclosures.py recomputes
+    // from the frozen annotations and payload (47 models; the ten outputs
+    // ruled on 2026-10-06 and the two Indiana county outputs leave scoring,
+    // the outputs the engine upgrade regenerates return to it).
     expect(summarizeAuditUniverse(us)).toEqual({
-      annotatedRowCount: 7_527,
-      legacyThresholdRowCount: 7_527,
-      exactMissCount: 7_523,
-      annotatedExactMissCount: 7_523,
+      annotatedRowCount: 8_001,
+      legacyThresholdRowCount: 8_001,
+      exactMissCount: 7_997,
+      annotatedExactMissCount: 7_997,
       annotatedExactHitCount: 4,
-      unannotatedBelowFullBoundedScoreCount: 2_072,
+      unannotatedBelowFullBoundedScoreCount: 2_138,
     });
   });
 });
diff --git a/app/tests/boardScope.test.ts b/app/tests/boardScope.test.ts
@@ -131,10 +131,26 @@ describe("dataset switching", () => {
       byEngine.set(version, (byEngine.get(version) ?? 0) + 1);
     }
     const engine = data.policyengineBundles?.us?.model_version as string;
-    const [older] = [...byEngine.keys()].filter((version) => version !== engine);
+    // Oldest engine first, as the sentence lists them.
+    const groups = [...byEngine.entries()]
+      .sort(([a], [b]) => {
+        const left = a.split(".").map(Number);
+        const right = b.split(".").map(Number);
+        for (let i = 0; i < Math.max(left.length, right.length); i += 1) {
+          const diff = (left[i] ?? 0) - (right[i] ?? 0);
+          if (diff !== 0) return diff;
+        }
+        return 0;
+      })
+      .map(([version, count], index) =>
+        index === 0
+          ? `${count} computed with policyengine-us ${version}`
+          : `${count} with ${version}`,
+      );
+    expect(groups.length).toBeGreaterThanOrEqual(2);
     const html = render("1.1").replaceAll("&#x27;", "'");
     expect(html).toContain(
-      `The ${exclusions.length} excluded outputs keep the values they were decided on (${byEngine.get(older)} computed with policyengine-us ${older}, ${byEngine.get(engine)} with ${engine})`,
+      `The ${exclusions.length} excluded outputs keep the values they were decided on (${groups.join(", ")})`,
     );
     expect(html).toContain(`that move on ${engine}; all`);
     expect(render("1.0")).not.toContain("excluded outputs keep the values");

diff --git a/app/tests/dataVersions.test.ts b/app/tests/dataVersions.test.ts
@@ -1,6 +1,7 @@
 import { describe, expect, test } from "bun:test";
 
 import registryJson from "../src/data.versions.json";
+import paperSnapshot from "../src/paperSnapshot.json";
 import {
   isLivePointerRef,
   parseDataVersionRegistry,
@@ -93,7 +94,7 @@ describe("parseDataVersionRegistry", () => {
     expect(
       registry.versions.find((version) => version.id === registry.default)
         ?.snapshotLabel,
-    ).toBe("Snapshot 2026-09-30");
+    ).toBe(`Snapshot ${paperSnapshot.snapshotDate}`);
   });
 });