MaxGhenis · Oct 31, 2026 · Oct 27, 2026 · Oct 27, 2026 · Oct 27, 2026 · Oct 27, 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/annotations/us_full_run_20260612_policyengine_4_16_1_populace/us_wording-amendments.json b/annotations/us_full_run_20260612_policyengine_4_16_1_populace/us_wording-amendments.json
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
@@ -92,8 +92,8 @@ export default function ExclusionNote({
             {exclusion.alternativeReading} Applying that rule gives{" "}
             <span className="font-[family-name:var(--font-mono)]">
               {formatValue(exclusion.alternativeValue, isBinary, currencySymbol)}
-            </span>{" "}
-            rather than the frozen{" "}
+            </span>
+            ; the frozen value is{" "}
             <span className="font-[family-name:var(--font-mono)]">
               {formatValue(exclusion.frozenValue, isBinary, currencySymbol)}
             </span>
@@ -114,8 +114,8 @@ export default function ExclusionNote({
             {exclusion.alternativeReading} Under that reading the reference is{" "}
             <span className="font-[family-name:var(--font-mono)]">
               {formatValue(exclusion.alternativeValue, isBinary, currencySymbol)}
-            </span>{" "}
-            rather than the frozen{" "}
+            </span>
+            ; the frozen value is{" "}
             <span className="font-[family-name:var(--font-mono)]">
               {formatValue(exclusion.frozenValue, isBinary, currencySymbol)}
             </span>

diff --git a/app/src/components/FailureModes.tsx b/app/src/components/FailureModes.tsx
@@ -138,7 +138,7 @@ function ErrorReadPatterns({
                     </div>
                     <p className="mt-1 text-sm leading-relaxed text-text-secondary">
                       {explainer?.summary ??
-                        "This target combines multiple policy rules, and errors usually come from positive cases rather than zero cases."}
+                        "This target combines multiple policy rules, and errors usually come from positive cases."}
                     </p>
                   </div>
                 </div>

diff --git a/app/src/components/Methodology.tsx b/app/src/components/Methodology.tsx
@@ -185,9 +185,9 @@ export default function Methodology({
               PolicyEngine: Medicaid and CHIP eligibility are asked &ldquo;under
               PolicyEngine rules,&rdquo; and the two school-meal outputs ask
               whether &ldquo;PolicyEngine returns&rdquo; support. PolicyBench
-              aims to measure knowledge of tax and benefit law, not of
-              PolicyEngine&apos;s implementation, and is revising these
-              definitions.{" "}
+              aims to measure knowledge of tax and benefit law. Knowledge of
+              PolicyEngine&apos;s implementation is outside that aim, so
+              PolicyBench is revising these definitions.{" "}
             </>
           ) : null}
           The answer instructions follow each row&apos;s transport: a
@@ -237,7 +237,7 @@ export default function Methodology({
           facts. It excludes intermediate tax bases, payroll subcomponents, and
           outputs that mainly require unavailable history, restricted local
           market data, or program take-up assignment. WIC is scored as
-          person-level eligibility, not as a dollar amount. Local income tax is
+          person-level eligibility, with no dollar amount. Local income tax is
           retained as a displayed requested output, but currently receives zero
           default population-impact weight because the full populace dataset has
           no positive modeled local-income-tax records.
@@ -281,8 +281,8 @@ export default function Methodology({
           Each frozen row uses its recorded per-call cost: reconstructed from
           token counts at the list price configured at request time, or the
           provider-reported charge where no reconstruction was available.
-          List-price overrides apply at
-          request time, not retroactively to recorded costs. Models without
+          A list-price override applies to
+          requests made after it is set and leaves recorded costs unchanged. Models without
           per-call costs use the frozen release-metadata cost.
         </SectionCard>
 
@@ -314,8 +314,7 @@ export default function Methodology({
           Binary coverage flags have 0/1 labels, but a 0/1 label is not their
           economic impact. Their leaderboard weights therefore come from
           PolicyEngine value proxies where available, such as estimated health
-          coverage or nutrition-program value, rather than from the binary label
-          itself. When every reference impact in a household is zero, the
+          coverage or nutrition-program value. When every reference impact in a household is zero, the
           household falls back to equal output weights.
         </SectionCard>
       </div>

diff --git a/app/src/data.artifact.json b/app/src/data.artifact.json
@@ -1,9 +1,9 @@
 {
   "asset": "dashboard-data.json",
-  "bytes": 135840397,
+  "bytes": 138702498,
   "repo": "PolicyEngine/policybench",
-  "sha256": "22db9a9c052545371687f6afa3551896336abbd15c76f767ebb777a77f6d2268",
-  "tag": "dashboard-data-20261014",
-  "url": "https://github.com/PolicyEngine/policybench/releases/download/dashboard-data-20261014/dashboard-data.json",
+  "sha256": "67cb0d14ca93cefaeaf2678ef6e6f81aea1858bb7b23b1b61d24c46fe378021b",
+  "tag": "dashboard-data-20261027",
+  "url": "https://github.com/PolicyEngine/policybench/releases/download/dashboard-data-20261027/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 2.41.0 (the 49 excluded outputs keep the values they were decided on: 29 from policyengine-us 1.755.4, 18 from 2.15.17, 2 from 2.38.6, and the 12 that move on 2.41.0 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 - 50 models",
+      "description": "Scored reference outputs from policyengine-us 2.60.0 (the 49 excluded outputs keep the values they were decided on: 29 from policyengine-us 1.755.4, 18 from 2.15.17, 2 from 2.38.6, and the 16 that move on 2.60.0 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 - 51 models",
       "id": "1.1",
       "label": "1.1",
-      "snapshotLabel": "Snapshot 2026-10-14"
+      "snapshotLabel": "Snapshot 2026-10-27"
     },
     {
       "artifact": {

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: "2.41.0",
-  rechecked: 12,
+  engineVersion: "2.60.0",
+  rechecked: 16,
 };
 
 function compareVersions(a: string, b: string): number {

diff --git a/app/src/lib/servingSensitivity.ts b/app/src/lib/servingSensitivity.ts
@@ -38,30 +38,30 @@ const FORCED_TOOL =
 
 export const SERVING_SENSITIVITY: Record<string, ServingSensitivity> = {
   "claude-fable-5": {
-    autoExact: 91.451,
+    autoExact: 91.381,
     boardTreatment: `${FORCED_TOOL}, one output per request`,
     autoTreatment:
       "re-run with tool_choice: auto and the whole household in one request",
     thinkingSuppressedOnBoard: true,
     noteHref: SENSITIVITY_DOC_HREF,
   },
   "claude-opus-5": {
-    autoExact: 90.109,
+    autoExact: 90.039,
     boardTreatment: FORCED_TOOL,
     autoTreatment: "re-run with tool_choice: auto",
     thinkingSuppressedOnBoard: true,
     noteHref: SENSITIVITY_DOC_HREF,
   },
   "claude-sonnet-5": {
-    autoExact: 84.574,
+    autoExact: 84.504,
     boardTreatment: `${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: 91.748,
+    autoExact: 91.679,
     boardTreatment:
       "rejects forced tool calls, so its row answers as a JSON object and reasons at the provider default",
     autoTreatment:
@@ -70,7 +70,7 @@ export const SERVING_SENSITIVITY: Record<string, ServingSensitivity> = {
     noteHref: "/notes/2026-09-01-claude-fable-5-1-added",
   },
   "claude-haiku-5.5": {
-    autoExact: 88.82,
+    autoExact: 88.75,
     boardTreatment: FORCED_TOOL,
     autoTreatment: "re-run with tool_choice: auto",
     thinkingSuppressedOnBoard: true,

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": 21
+    "run_state": 22
   },
   "models": {
     "claude-fable-5": {
@@ -1297,6 +1297,63 @@
       "shared_completion_budget_tokens": 16384,
       "tool_choice": "forced"
     },
+    "gpt-6.1-terra": {
+      "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": "0f0bc249927d0567e93d41afb01613f5e475955a5eba8c0bba75e3db203ef411",
+        "kind": "run_state",
+        "rows_match": true,
+        "run": "terra-runs/terra",
+        "source_predictions_sha256": "6de90940fb3774b4d3c96722e11de33e47b375da1e2bc1a4d5b53dff253b69cb",
+        "source_rows_sha256": "0f0bc249927d0567e93d41afb01613f5e475955a5eba8c0bba75e3db203ef411",
+        "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": "gpt-6.1-terra",
+          "prompt_contract_version": "2026-08-09-v2-scoring-contract",
+          "request_timeout_seconds": 300,
+          "thinking": {
+            "mode": "provider_default"
+          },
+          "tool_choice_mode": "forced"
+        }
+      },
+      "provider_id": "gpt-6.1-terra",
+      "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"
+    },
     "grok-4.3": {
       "answer_contract": "tool",
       "evidence": {
@@ -1756,7 +1813,7 @@
       "tool_choice": null
     }
   },
-  "registry_commit": "236134f5c3b4400a14ae7de24604c988f6f602eb",
+  "registry_commit": "2677b1164761d23356f73f6674007f5bc79966ee",
   "sources": [
     "policybench.config.MODELS",
     "policybench.model_cards",

diff --git a/app/src/modelMeta.ts b/app/src/modelMeta.ts
@@ -24,6 +24,7 @@ export const MODEL_ORDER = [
   "gpt-6-astra",
   "gpt-6-sol",
   "gpt-6.1-sol",
+  "gpt-6.1-terra",
   "gpt-6-luna",
   "gpt-5.6-sol",
   "gpt-5.6-terra",
@@ -78,6 +79,7 @@ export const MODEL_LABELS: Record<string, string> = {
   "gpt-6-astra": "GPT-6 Astra",
   "gpt-6-sol": "GPT-6 Sol",
   "gpt-6.1-sol": "GPT-6.1 Sol",
+  "gpt-6.1-terra": "GPT-6.1 Terra",
   "gpt-6-luna": "GPT-6 Luna",
   "gpt-5.6-sol": "GPT-5.6 Sol",
   "gpt-5.6-terra": "GPT-5.6 Terra",

diff --git a/app/src/notes/2026-10-31-gpt-6-1-terra-joins-the-board.json b/app/src/notes/2026-10-31-gpt-6-1-terra-joins-the-board.json
@@ -0,0 +1,78 @@
+{
+  "slug": "2026-10-31-gpt-6-1-terra-joins-the-board",
+  "date": "2026-10-31",
+  "title": "OpenAI's GPT-6.1 Terra joins the board at #3 as the references move to policyengine-us 2.60.0",
+  "paragraphs": [
+    "PolicyBench added OpenAI's GPT-6.1 Terra to its {snapshotDate} snapshot, the day after OpenAI released it, bringing the board to {nModels} models. It gets {terraExact}% of answers within $1, weighted by household impact, #{terraRank} of {nModels}. Only GPT-6 Sol ({solExact}%) and Claude Opus 5.5 ({opusExact}%) score higher, and GPT-5.6 Sol, which also rounds to {sol56Exact}%, is less than a tenth of a point behind. On the bounded score, which gives partial credit for amounts close to the reference, GPT-6.1 Terra ranks #{terraBoundedRank}. GPT-6.1 Sol, which OpenAI released on September 29, gets {sol61Exact}% (#{sol61Rank}).",
+    "PolicyBench runs GPT-6.1 Terra as it runs GPT-6.1 Sol, on OpenAI's Responses API with the board's forced answer tool at OpenAI's default reasoning effort, and it returned reasoning tokens on all {households} households. At OpenAI's price of $1 per million input tokens and $5 per million output tokens, the {households} households cost ${terraCost}, against ${sol61Cost} for GPT-6.1 Sol and ${solCost} for GPT-6 Sol. GPT-5.6 Terra, the Terra model before it, gets {terra56Exact}% (#{terra56Rank}) and cost ${terra56Cost}.",
+    "The release also moves PolicyBench's references from policyengine-us {previousEngine} to {engineVersion}, the newest release when PolicyBench began recomputing them on October 23. The move changes {changed:words} scored references, each for law in force before the references were frozen. An elderly Californian who lives alone, whose SNAP was $0, now gets the ${snapMonthly} monthly minimum, ${snap031} a year: California's categorical eligibility for CalFresh applies no net income test, even to a household with an elderly member, and the previous version applied one (policyengine-us #10036). Three households in Pennsylvania and Arkansas have too much income for free or reduced-price school lunch, but those states give students at covered schools a free breakfast, so the output asking whether PolicyEngine returns free school meal support is now yes for each (policyengine-us #10238). The new version models both breakfast laws for a student at a covered school and leaves out their conditions: Pennsylvania reimburses schools for them to the extent funds are available, and Arkansas serves them on request.",
+    "Nothing is newly excluded, and no excluded output returns to scoring, so every model is still scored on {scoredOutputs} of its {totalOutputs} requested outputs. When PolicyBench checked PyPI on {checkedOn}, the newest policyengine-us release was {checkedEngine}, and it gives the same value as {engineVersion} for all {scoredOutputs}; it moves {excludedMoved:words} excluded output, which keeps the value it was decided on.",
+    "Against the previous release, the exact rates of {fallers:words} of the {incumbents} earlier models fall, by {fallMin} to {fallMax} points, and {risers:words} rise by {rise}. No model answers yes for any of the three households' school meals, and only {snapMatch:words} give the Californian's new SNAP: GPT-5.5, GPT-6 Astra and GPT-6.1 Sol, the three that rise. Claude Sonnet 5.5 falls from #{sonnetThen} to #{sonnetNow}, behind GPT-6.1 Terra and GPT-6 Astra."
+  ],
+  "facts": {
+    "snapshotDate": "2026-10-27",
+    "nModels": 51,
+    "terraExact": 93.1,
+    "terraRank": 3,
+    "solExact": 95.2,
+    "opusExact": 94.1,
+    "sol56Exact": 93.1,
+    "terraBoundedRank": 5,
+    "sol61Exact": 90.2,
+    "sol61Rank": 10,
+    "households": 100,
+    "terraCost": "1.25",
+    "sol61Cost": "2.30",
+    "solCost": "2.71",
+    "terra56Exact": 87.6,
+    "terra56Rank": 18,
+    "terra56Cost": "3.89",
+    "previousEngine": "2.41.0",
+    "engineVersion": "2.60.0",
+    "changed": 4,
+    "snapMonthly": 24,
+    "snap031": 288,
+    "scoredOutputs": 1935,
+    "totalOutputs": 1984,
+    "checkedOn": "2026-10-29",
+    "checkedEngine": "2.78.0",
+    "excludedMoved": 1,
+    "incumbents": 50,
+    "fallers": 47,
+    "fallMin": 0.03,
+    "fallMax": 0.07,
+    "risers": 3,
+    "rise": 0.02,
+    "snapMatch": 3,
+    "sonnetThen": 4,
+    "sonnetNow": 6
+  },
+  "data": [
+    {
+      "label": "Dashboard data release",
+      "href": "https://github.com/PolicyEngine/policybench/releases/tag/dashboard-data-20261027"
+    },
+    {
+      "label": "Release design note",
+      "href": "https://github.com/PolicyEngine/policybench/blob/main/docs/terra/design.md"
+    },
+    {
+      "label": "Engine upgrade record",
+      "href": "https://github.com/PolicyEngine/policybench/tree/main/reference_audit/2026-10-22-engine-upgrade"
+    },
+    {
+      "label": "Publication check against policyengine-us 2.78.0",
+      "href": "https://github.com/PolicyEngine/policybench/blob/main/reference_audit/2026-10-22-engine-upgrade/verification/publication_check_2_78_0.csv"
+    },
+    {
+      "label": "GPT-6.1 Terra model page",
+      "href": "/model/gpt-6.1-terra"
+    },
+    {
+      "label": "Previous release note (October 14)",
+      "href": "/notes/2026-10-14-gemini-4-argon-joins-the-board"
+    }
+  ],
+  "boardSnapshot": "2026-10-27",
+  "release": "dashboard-data-20261027"
+}
diff --git a/app/src/notes/index.ts b/app/src/notes/index.ts
@@ -1,3 +1,4 @@
+import gpt61TerraJoinsTheBoard from "./2026-10-31-gpt-6-1-terra-joins-the-board.json";
 import givenAPolicyEngineToolThreeModelsScoreAbove99 from "./2026-10-20-given-a-policyengine-tool-three-models-score-above-99.json";
 import americaGovScoresLevelWithGpt55 from "./2026-10-19-america-gov-scores-level-with-gpt-5-5.json";
 import gemini4ArgonJoinsTheBoard from "./2026-10-14-gemini-4-argon-joins-the-board.json";
@@ -34,6 +35,7 @@ export type PolicyBenchNote = {
 };
 
 export const notes = [
+  gpt61TerraJoinsTheBoard,
   givenAPolicyEngineToolThreeModelsScoreAbove99,
   americaGovScoresLevelWithGpt55,
   gemini4ArgonJoinsTheBoard,

diff --git a/app/src/paperSnapshot.json b/app/src/paperSnapshot.json
@@ -1,6 +1,6 @@
 {
-  "pdfVersion": "41443b877bdb",
-  "responseWindow": "June 12 and October 13, 2026",
-  "snapshotDate": "2026-10-14",
-  "webVersion": "bf2ec3dedcca"
+  "pdfVersion": "501fb9038a2f",
+  "responseWindow": "June 12 and October 26, 2026",
+  "snapshotDate": "2026-10-27",
+  "webVersion": "f9d806df25d9"
 }
diff --git a/app/tests/auditUniverse.test.ts b/app/tests/auditUniverse.test.ts
@@ -18,17 +18,17 @@ describe("audit universe", () => {
     expect(AUDIT_SELECTION_RULE).toBe(
       "rows whose legacy threshold score is below 1",
     );
-    // Release 20261014: the counts tests/test_disclosures.py recomputes
-    // from the frozen annotations and payload (50 models; Gemini 4 Argon's
-    // and Muse Spark 1.3's rows join on release 20261013's references and
-    // exclusions).
+    // Release 20261027: the counts tests/test_disclosures.py recomputes
+    // from the frozen annotations and payload (51 models; GPT-6.1 Terra's
+    // rows join, and the move to policyengine-us 2.60.0 changes four scored
+    // references; release 20261014: 8,765 annotated, 2,329 unannotated).
     expect(summarizeAuditUniverse(us)).toEqual({
-      annotatedRowCount: 8_765,
-      legacyThresholdRowCount: 8_765,
-      exactMissCount: 8_761,
-      annotatedExactMissCount: 8_761,
+      annotatedRowCount: 9_050,
+      legacyThresholdRowCount: 9_050,
+      exactMissCount: 9_046,
+      annotatedExactMissCount: 9_046,
       annotatedExactHitCount: 4,
-      unannotatedBelowFullBoundedScoreCount: 2_329,
+      unannotatedBelowFullBoundedScoreCount: 2_407,
     });
   });
 });
diff --git a/app/tests/expandPage.test.ts b/app/tests/expandPage.test.ts
@@ -71,7 +71,7 @@ test("expand page derives its exact-score headline from the live summary", async
   ).text();
 
   expect(leader.model).toBe("gpt-6-sol");
-  expect(leader.exact).toBeCloseTo(95.225, 3);
+  expect(leader.exact).toBeCloseTo(95.156, 3);
   expect(pageSource).toContain("const leader = headlineExactLeader(dashboard);");
   expect(pageSource).toContain(
     "const leaderLabel = MODEL_LABELS[leader.model] ?? leader.model;",