@lcv-ideas-software/cross-review 4.2.1 → 4.2.3

package/CHANGELOG.md

@@ -7,6 +7,48 @@ standard `v00.00.00`; npm package versions remain SemVer.
 
 ## [Unreleased]
 
+## [v04.02.03] — 2026-06-03
+
+**Patch — Gemini replacement pin and rate-card refresh.** This release follows
+Google's deprecation schedule for Gemini 2.5 Pro by making Gemini 3.1 Pro Preview
+the active canonical Gemini pin.
+
+### Changed
+
+- Promoted the Google/Gemini canonical default from `gemini-2.5-pro` to
+  `gemini-3.1-pro-preview` after Google's deprecation schedule listed the
+  former for shutdown on 2026-10-16.
+- Updated the active local Gemini rate card from Gemini 2.5 Pro pricing to
+  Gemini 3.1 Pro Preview pricing, including the >200K extended tier and
+  cached-input rates.
+
+## [v04.02.02] — 2026-06-02
+
+**Patch — provider-doc refresh and Perplexity probe repair.** This release
+updates the maintained provider pins and rate-card guidance after a
+cross-review audit of the current v4.2.1 session corpus.
+
+### Fixed
+
+- Raised the Perplexity `sonar-reasoning-pro` health probe to `max_tokens=16`,
+  matching the provider's current minimum and preventing false unavailable
+  capability snapshots while still keeping `disable_search=true`.
+- Added `provider-refresh-smoke` coverage for the Perplexity probe minimum and
+  for the current Claude/Grok canonical model pins.
+
+### Changed
+
+- Promoted the Anthropic canonical/default model from `claude-opus-4-7` to
+  `claude-opus-4-8`.
+- Promoted the Grok canonical/default model from the alias `grok-4-latest` to
+  the concrete `grok-4.3` pin while keeping alias behavior documented.
+- Refreshed provider rate-card documentation for GPT-5.5, Claude Opus 4.8,
+  Gemini 2.5 Pro, DeepSeek V4 Pro, Grok 4.3, and Perplexity Sonar Reasoning
+  Pro.
+- Updated the active local runtime config at
+  `C:\Users\leona\.cross-review\data\config.json` with current cached-input,
+  extended-tier, and DeepSeek base rates.
+
 ## [v04.02.01] — 2026-05-21
 
 **Patch — publish the workspace hard-gate cleanup as a package release.** The

package/README.md

@@ -5,7 +5,7 @@
 # cross-review
 
 > MCP server orchestrating API-first cross-review between Claude, ChatGPT Codex,
-> Gemini, DeepSeek, and Grok with unanimous convergence gates.
+> Gemini, DeepSeek, Grok, and Perplexity with unanimous convergence gates.
 
 [![status: stable](https://img.shields.io/badge/status-stable-brightgreen.svg)](#status)
 [![release](https://img.shields.io/github/v/release/LCV-Ideas-Software/cross-review?sort=semver)](https://github.com/LCV-Ideas-Software/cross-review/releases)
@@ -24,7 +24,7 @@ npm install -g @lcv-ideas-software/cross-review
 npm install -g @lcv-ideas-software/cross-review --registry=https://npm.pkg.github.com
 ```
 
-**Status.** Stable. Current release: **v04.02.01** (npm package `4.2.1`). See [CHANGELOG.md](./CHANGELOG.md) for the full release history.
+**Status.** Stable. Current release: **v04.02.03** (npm package `4.2.3`). See [CHANGELOG.md](./CHANGELOG.md) for the full release history.
 
 > **Project renamed 2026-05-15.** This project was previously published as
 > [`@lcv-ideas-software/cross-review-v2`](https://www.npmjs.com/package/@lcv-ideas-software/cross-review-v2)
@@ -38,6 +38,8 @@ The version history at a glance:
 
 | Release              | Scope                                                                                                                                                                                                              |
 | -------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| **`v04.02.03`**      | Patch — promote the Gemini canonical default to `gemini-3.1-pro-preview` and refresh the active local Gemini rate card.                                                                                            |
+| **`v04.02.02`**      | Patch — provider-doc refresh, Perplexity probe repair, current model pins, and rate-card guidance.                                                                                                                 |
 | **`v04.02.01`**      | Patch — publish the workspace hard-gate cleanup as a package release.                                                                                                                                              |
 | **`v04.02.00`**      | Minor — bounded MCP session listing and cancellation semantics cleanup.                                                                                                                                            |
 | **`v04.01.01`**      | Patch — release the hard-gate cleanup as a published package.                                                                                                                                                      |
@@ -119,8 +121,8 @@ The version history at a glance:
 
 `cross-review` is the stable API-first implementation of the cross-review
 pattern. It orchestrates provider API clients (OpenAI/Codex, Anthropic/Claude,
-Google Gemini, DeepSeek, and xAI/Grok) and provides an MCP-compatible server
-surface.
+Google Gemini, DeepSeek, xAI/Grok, and Perplexity Sonar) and provides an
+MCP-compatible server surface.
 
 Runtime calls are real provider calls by default. Stubs exist only for smoke
 tests and CI when `CROSS_REVIEW_STUB=1`.
@@ -130,6 +132,7 @@ tests and CI when `CROSS_REVIEW_STUB=1`.
 - Google Gen AI client library for Gemini.
 - OpenAI-compatible DeepSeek API through the OpenAI client library.
 - OpenAI-compatible xAI Grok API through the OpenAI client library.
+- OpenAI-compatible Perplexity Sonar API through the OpenAI client library.
 
 ## Quick Start
 
@@ -171,11 +174,12 @@ variables. Example overrides (PowerShell):
 [Environment]::SetEnvironmentVariable("CROSS_REVIEW_GROK_REASONING_EFFORT", "xhigh", "User")
 ```
 
-For Grok, `GROK_API_KEY` is canonical. `grok-4-latest`, `grok-4.3`,
-`grok-4.20`, and `grok-4.20-reasoning` use xAI automatic reasoning without an explicit
-`reasoning.effort` field. `grok-4.20-multi-agent` accepts explicit
-`reasoning.effort`; `low`/`medium` select 4 agents and `high`/`xhigh` select
-16 agents.
+For Grok, `GROK_API_KEY` is canonical. The default pin is `grok-4.3`, which
+accepts explicit `reasoning.effort` through `high`; the adapter clamps the
+shared effort scale before sending it. `grok-4-latest`, `grok-4.20`, and
+`grok-4.20-reasoning` use xAI automatic reasoning in this runtime.
+`grok-4.20-multi-agent` remains available as an explicit override for the
+multi-agent variant.
 
 Financial and budget controls are required for paid provider calls. Configure
 these environment variables before running real sessions (example):

package/dist/scripts/provider-refresh-smoke.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/scripts/provider-refresh-smoke.js

@@ -0,0 +1,56 @@
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+import { loadConfig } from "../src/core/config.js";
+import { selectFromCandidates } from "../src/peers/model-selection.js";
+import { PerplexityAdapter } from "../src/peers/perplexity.js";
+process.env.CROSS_REVIEW_STUB = "1";
+process.env.CROSS_REVIEW_STUB_CONFIRMED = "1";
+process.env.PERPLEXITY_API_KEY = "test-perplexity-key";
+process.env.CROSS_REVIEW_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "cross-review-provider-refresh-"));
+const config = loadConfig();
+{
+    const adapter = new PerplexityAdapter(config);
+    let capturedPayload;
+    adapter.client = async () => ({
+        chat: {
+            completions: {
+                create: async (payload) => {
+                    capturedPayload = payload;
+                },
+            },
+        },
+    });
+    const probe = await adapter.probe();
+    assert.equal(probe.available, true);
+    assert.equal(capturedPayload?.disable_search, true);
+    assert.ok(typeof capturedPayload?.max_tokens === "number" && capturedPayload.max_tokens >= 16, "Perplexity probe must request at least 16 max_tokens for sonar-reasoning-pro.");
+}
+{
+    const claude = selectFromCandidates("claude", [{ id: "claude-opus-4-8", source: "api" }], "claude-opus-4-8");
+    assert.equal(claude.selected, "claude-opus-4-8");
+    assert.equal(claude.confidence, "verified");
+}
+{
+    const gemini = selectFromCandidates("gemini", [{ id: "gemini-3.1-pro-preview", source: "api" }], "gemini-3.1-pro-preview");
+    assert.equal(gemini.selected, "gemini-3.1-pro-preview");
+    assert.equal(gemini.confidence, "verified");
+}
+{
+    const grok = selectFromCandidates("grok", [{ id: "grok-4.3", source: "api" }], "grok-4.3");
+    assert.equal(grok.selected, "grok-4.3");
+    assert.equal(grok.confidence, "verified");
+}
+{
+    const configSource = fs.readFileSync("src/core/config.ts", "utf8");
+    const modelSelectionSource = fs.readFileSync("src/peers/model-selection.ts", "utf8");
+    assert.ok(configSource.includes('claude: envValue("CROSS_REVIEW_ANTHROPIC_MODEL") || "claude-opus-4-8"'));
+    assert.ok(configSource.includes('gemini: envValue("CROSS_REVIEW_GEMINI_MODEL") || "gemini-3.1-pro-preview"'));
+    assert.ok(configSource.includes('grok: envValue("CROSS_REVIEW_GROK_MODEL") || "grok-4.3"'));
+    assert.ok(modelSelectionSource.includes('claude: ["claude-opus-4-8"]'));
+    assert.ok(modelSelectionSource.includes('gemini: ["gemini-3.1-pro-preview"]'));
+    assert.ok(modelSelectionSource.includes('grok: ["grok-4.3"]'));
+}
+console.log("[provider-refresh-smoke] PASS");
+//# sourceMappingURL=provider-refresh-smoke.js.map

package/dist/scripts/provider-refresh-smoke.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"provider-refresh-smoke.js","sourceRoot":"","sources":["../../scripts/provider-refresh-smoke.ts"],"names":[],"mappings":"AAAA,OAAO,MAAM,MAAM,oBAAoB,CAAC;AACxC,OAAO,EAAE,MAAM,SAAS,CAAC;AACzB,OAAO,EAAE,MAAM,SAAS,CAAC;AACzB,OAAO,IAAI,MAAM,WAAW,CAAC;AAE7B,OAAO,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AACnD,OAAO,EAAE,oBAAoB,EAAE,MAAM,iCAAiC,CAAC;AACvE,OAAO,EAAE,iBAAiB,EAAE,MAAM,4BAA4B,CAAC;AAE/D,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,GAAG,CAAC;AACpC,OAAO,CAAC,GAAG,CAAC,2BAA2B,GAAG,GAAG,CAAC;AAC9C,OAAO,CAAC,GAAG,CAAC,kBAAkB,GAAG,qBAAqB,CAAC;AACvD,OAAO,CAAC,GAAG,CAAC,qBAAqB,GAAG,EAAE,CAAC,WAAW,CAChD,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,gCAAgC,CAAC,CACzD,CAAC;AAEF,MAAM,MAAM,GAAG,UAAU,EAAE,CAAC;AAE5B,CAAC;IACC,MAAM,OAAO,GAAG,IAAI,iBAAiB,CAAC,MAAM,CAAC,CAAC;IAC9C,IAAI,eAA8E,CAAC;IAEjF,OASD,CAAC,MAAM,GAAG,KAAK,IAAI,EAAE,CAAC,CAAC;QACtB,IAAI,EAAE;YACJ,WAAW,EAAE;gBACX,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;oBACxB,eAAe,GAAG,OAAO,CAAC;gBAC5B,CAAC;aACF;SACF;KACF,CAAC,CAAC;IAEH,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;IACpC,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;IACpC,MAAM,CAAC,KAAK,CAAC,eAAe,EAAE,cAAc,EAAE,IAAI,CAAC,CAAC;IACpD,MAAM,CAAC,EAAE,CACP,OAAO,eAAe,EAAE,UAAU,KAAK,QAAQ,IAAI,eAAe,CAAC,UAAU,IAAI,EAAE,EACnF,+EAA+E,CAChF,CAAC;AACJ,CAAC;AAED,CAAC;IACC,MAAM,MAAM,GAAG,oBAAoB,CACjC,QAAQ,EACR,CAAC,EAAE,EAAE,EAAE,iBAAiB,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC,EAC1C,iBAAiB,CAClB,CAAC;IACF,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,EAAE,iBAAiB,CAAC,CAAC;IACjD,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;AAC9C,CAAC;AAED,CAAC;IACC,MAAM,MAAM,GAAG,oBAAoB,CACjC,QAAQ,EACR,CAAC,EAAE,EAAE,EAAE,wBAAwB,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC,EACjD,wBAAwB,CACzB,CAAC;IACF,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,EAAE,wBAAwB,CAAC,CAAC;IACxD,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;AAC9C,CAAC;AAED,CAAC;IACC,MAAM,IAAI,GAAG,oBAAoB,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,EAAE,UAAU,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC,EAAE,UAAU,CAAC,CAAC;IAC3F,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;IACxC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;AAC5C,CAAC;AAED,CAAC;IACC,MAAM,YAAY,GAAG,EAAE,CAAC,YAAY,CAAC,oBAAoB,EAAE,MAAM,CAAC,CAAC;IACnE,MAAM,oBAAoB,GAAG,EAAE,CAAC,YAAY,CAAC,8BAA8B,EAAE,MAAM,CAAC,CAAC;IACrF,MAAM,CAAC,EAAE,CACP,YAAY,CAAC,QAAQ,CAAC,uEAAuE,CAAC,CAC/F,CAAC;IACF,MAAM,CAAC,EAAE,CACP,YAAY,CAAC,QAAQ,CACnB,2EAA2E,CAC5E,CACF,CAAC;IACF,MAAM,CAAC,EAAE,CAAC,YAAY,CAAC,QAAQ,CAAC,yDAAyD,CAAC,CAAC,CAAC;IAC5F,MAAM,CAAC,EAAE,CAAC,oBAAoB,CAAC,QAAQ,CAAC,6BAA6B,CAAC,CAAC,CAAC;IACxE,MAAM,CAAC,EAAE,CAAC,oBAAoB,CAAC,QAAQ,CAAC,oCAAoC,CAAC,CAAC,CAAC;IAC/E,MAAM,CAAC,EAAE,CAAC,oBAAoB,CAAC,QAAQ,CAAC,oBAAoB,CAAC,CAAC,CAAC;AACjE,CAAC;AAED,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC"}

package/dist/scripts/smoke.js

@@ -209,6 +209,7 @@ for (const { file, field } of adapterExpectations) {
 const modelSelectionSource = fs.readFileSync("src/peers/model-selection.ts", "utf8");
 for (const deprecatedOrWeakModel of [
     "claude-haiku-4-5",
+    "gemini-2.5-pro",
     "gemini-3-pro-preview",
     "deepseek-reasoner",
     "deepseek-chat",
@@ -220,16 +221,16 @@ for (const deprecatedOrWeakModel of [
 // "must remain" list is therefore exactly the 6 lone canonical pins.
 for (const canonicalPin of [
     "gpt-5.5",
-    "claude-opus-4-7",
-    "gemini-2.5-pro",
+    "claude-opus-4-8",
+    "gemini-3.1-pro-preview",
     "deepseek-v4-pro",
-    "grok-4-latest",
+    "grok-4.3",
     "sonar-reasoning-pro",
 ]) {
     assert.ok(modelSelectionSource.includes(`"${canonicalPin}"`), `${canonicalPin} must remain the lone canonical PRIORITY pin`);
 }
-const noWeakDowngrade = selectFromCandidates("claude", [{ id: "claude-haiku-4-5-20251001", source: "api" }], "claude-opus-4-7");
-assert.equal(noWeakDowngrade.selected, "claude-opus-4-7");
+const noWeakDowngrade = selectFromCandidates("claude", [{ id: "claude-haiku-4-5-20251001", source: "api" }], "claude-opus-4-8");
+assert.equal(noWeakDowngrade.selected, "claude-opus-4-8");
 assert.equal(noWeakDowngrade.confidence, "unknown");
 assert.match(noWeakDowngrade.reason, /silently downgrading/);
 const pemMarker = (side, label) => ["-----", side, " ", label, "-----"].join("");
@@ -1347,6 +1348,35 @@ assert.equal(Object.hasOwn(metrics.decision_quality, "undefined"), false);
     assert.ok(/evidence_sources/.test(instruction), "statusInstruction must direct detail to evidence_sources (v2.5.0)");
     console.log("[smoke] session_contract_directives_test: PASS");
 }
+// v4.2.2 — verified_requires_evidence_sources_test. Peer JSON may still
+// declare READY, but a `confidence:"verified"` verdict without concrete
+// evidence_sources must not be classified as a clean machine decision.
+{
+    const statusModule = await import("../src/core/status.js");
+    const parseStatusForTruth = statusModule.parsePeerStatus;
+    const statusInstruction = statusModule.statusInstruction;
+    const ungrounded = parseStatusForTruth(JSON.stringify({
+        status: "READY",
+        summary: "Looks correct.",
+        confidence: "verified",
+        evidence_sources: [],
+        caller_requests: [],
+        follow_ups: [],
+    }));
+    assert.ok(ungrounded.parser_warnings.includes("verified_without_evidence_sources"), "v4.2.2 / truthfulness_guardrails: confidence=verified with empty evidence_sources must emit verified_without_evidence_sources");
+    assert.equal(ungrounded.structured?.confidence, "verified", "v4.2.2 / truthfulness_guardrails: parser warning must not silently rewrite peer confidence");
+    const grounded = parseStatusForTruth(JSON.stringify({
+        status: "READY",
+        summary: "Runtime claim matches the raw source.",
+        confidence: "verified",
+        evidence_sources: ['server_info: {"version":"4.2.1","release_date":"2026-05-21"}'],
+        caller_requests: [],
+        follow_ups: [],
+    }));
+    assert.ok(!grounded.parser_warnings.includes("verified_without_evidence_sources"), "v4.2.2 / truthfulness_guardrails: concrete evidence_sources must satisfy verified confidence");
+    assert.ok(/confidence.*verified[\s\S]+evidence_sources/i.test(statusInstruction()), "v4.2.2 / truthfulness_guardrails: statusInstruction must tie verified confidence to concrete evidence_sources");
+    console.log("[smoke] verified_requires_evidence_sources_test: PASS");
+}
 // v2.5.0: CROSS_REVIEW_DEFAULT_MAX_ROUNDS env override is honored.
 {
     const { loadConfig: reload } = await import("../src/core/config.js");
@@ -3644,14 +3674,12 @@ assert.equal(Object.hasOwn(metrics.decision_quality, "undefined"), false);
     // v3.0.0: PEERS now has 6 entries (perplexity added).
     assert.equal(PEERS.length, 6, "PEERS must have 6 entries (codex/claude/gemini/deepseek/grok/perplexity)");
     const cfg = loadConfig();
-    // v3.7.2 (AUDIT-3 + operator directive 2026-05-14): default grok model
-    // is `grok-4-latest` — the operator's chosen canonical pin for
-    // cross-review, superseding the v2.14.1 `grok-4.20-multi-agent`
-    // default. `grok-4.20-multi-agent` remains a valid env-override
-    // (CROSS_REVIEW_GROK_MODEL) for explicit reasoning.effort control — the
-    // adapter still handles it; the modelAcceptsReasoningEffort /
-    // clampEffortForModel tests below continue to pin that capability.
-    assert.equal(cfg.models.grok, "grok-4-latest", "default grok model must be grok-4-latest (v3.7.2, operator directive)");
+    // v4.2.2 provider-doc refresh: default grok model is the concrete
+    // `grok-4.3` pin. `grok-4-latest` remains a valid xAI alias and
+    // `grok-4.20-multi-agent` remains a valid env-override for explicit
+    // multi-agent reasoning behavior; the adapter tests below continue to
+    // pin those capabilities.
+    assert.equal(cfg.models.grok, "grok-4.3", "default grok model must be grok-4.3 (v4.2.2 provider-doc refresh)");
     assert.ok("grok" in cfg.fallback_models, "fallback_models must have grok entry");
     assert.equal(cfg.peer_enabled.grok, true, "grok must be enabled by default");
     assert.ok(cfg.cost_rates.grok, "grok cost rates must be configured (env-set in smoke setup)");
@@ -5210,6 +5238,19 @@ assert.equal(Object.hasOwn(metrics.decision_quality, "undefined"), false);
     });
     assert.ok(netNewAssertionWithDraft.fabricated === true &&
         netNewAssertionWithDraft.suspicious_assertion_count >= 2, `v3.7.4 / fabrication_lock: operational assertions NET-NEW vs {provenance ∪ priorDraft} — invented by the relator even though a prior draft exists — MUST still trip fabricated=true (got count=${netNewAssertionWithDraft.suspicious_assertion_count}, fabricated=${netNewAssertionWithDraft.fabricated})`);
+    const inventedWorkflowDispatch = detectFabricatedEvidence("Refazendo agora. Workflow launched in background. Task ID: wllbll9am. Run ID: wf_e7c69578-e23.", {
+        provenanceCorpus: "",
+        priorDraftCorpus: "The user challenged the report and did not authorize a redo.",
+        narrativeCorpus: "Analyze why Claude lied about the prior v4.2.0 audit.",
+    });
+    assert.ok(inventedWorkflowDispatch.fabricated === true &&
+        inventedWorkflowDispatch.suspicious_assertion_count >= 2, `v4.2.2 / truthfulness_guardrails: invented workflow dispatch claims MUST trip fabricated=true (got count=${inventedWorkflowDispatch.suspicious_assertion_count}, fabricated=${inventedWorkflowDispatch.fabricated})`);
+    const genericConfirmation = detectFabricatedEvidence("The reviewer confirmed the model-selection rationale is clear.", {
+        provenanceCorpus: "",
+        priorDraftCorpus: "",
+        narrativeCorpus: "",
+    });
+    assert.equal(genericConfirmation.fabricated, false, "v4.2.2 / truthfulness_guardrails: generic 'confirmed' prose without a dispatch/authorization claim must not trip fabrication detection");
     // Source-level: threshold constants pinned at the documented values.
     assert.ok(/FABRICATED_NET_NEW_HEX_THRESHOLD\s*=\s*3/.test(orchSrc), "v2.24.0 / fabrication_lock: net-new hex threshold pinned at 3");
     assert.ok(/FABRICATED_SUSPICIOUS_ASSERTION_THRESHOLD\s*=\s*2/.test(orchSrc), "v2.24.0 / fabrication_lock: suspicious assertion threshold pinned at 2");
@@ -5675,6 +5716,75 @@ assert.equal(Object.hasOwn(metrics.decision_quality, "undefined"), false);
     assert.ok(/boolEnv\("CROSS_REVIEW_EVIDENCE_PREFLIGHT", true\)/.test(configSrcPf), "v3.5.0 / evidence_preflight: CROSS_REVIEW_EVIDENCE_PREFLIGHT env var must default ON");
     console.log("[smoke] evidence_preflight_test: PASS");
 }
+// v4.2.2 — truthfulness_preflight_test. Pins the guard added after the
+// Claude Code Opus 4.8 incident where a report asserted
+// "v4.2.0 current production" despite live server_info showing
+// v4.2.1. The old evidence preflight only checked completed-work
+// claims (tests/diff/build) and did not reject current-runtime
+// contradictions or unsupported historical timing narratives.
+{
+    const { truthfulnessPreflight } = await import("../src/core/orchestrator.js");
+    const runtimeFacts = {
+        runtime_version: "4.2.1",
+        release_date: "2026-05-21",
+        model_pins: {
+            claude: "claude-opus-4-8",
+            grok: "grok-4.3",
+        },
+    };
+    const contradictedByRuntime = truthfulnessPreflight({
+        task: "Audit all sessions generated with the current cross-review version.",
+        initialDraft: 'Live server_info: {"version":"4.2.1","release_date":"2026-05-21"}\nAudit report for cross-review v4.2.0 current production, released 2026-05-17.',
+        runtimeFacts,
+        attachmentsPresent: false,
+    });
+    assert.equal(contradictedByRuntime.pass, false, "v4.2.2 / truthfulness_preflight: current-runtime version claim contradicting runtime facts must trip even when server_info text is present");
+    assert.ok(contradictedByRuntime.contradictions.some((item) => item.includes("4.2.0")), "v4.2.2 / truthfulness_preflight: mismatch diagnostics must include the contradicted version token");
+    const backedByRuntime = truthfulnessPreflight({
+        task: "Audit all sessions generated with the current cross-review version.",
+        initialDraft: 'Live server_info: {"version":"4.2.1","release_date":"2026-05-21"}\nAudit report for cross-review v4.2.1 current production, released 2026-05-21.',
+        runtimeFacts,
+        attachmentsPresent: false,
+    });
+    assert.equal(backedByRuntime.pass, true, "v4.2.2 / truthfulness_preflight: current-runtime claim matching runtime facts must pass");
+    const unsupportedCurrentState = truthfulnessPreflight({
+        task: "Audit all sessions generated with the current cross-review version.",
+        initialDraft: "Audit report for cross-review v4.2.1 current production.",
+        runtimeFacts: {},
+        attachmentsPresent: false,
+    });
+    assert.equal(unsupportedCurrentState.pass, false, "v4.2.2 / truthfulness_preflight: current-runtime claim without runtime facts or source evidence must trip");
+    const historicalChangelog = truthfulnessPreflight({
+        task: "Review this changelog text.",
+        initialDraft: "v4.2.0 was released on 2026-05-17. v4.2.1 was released on 2026-05-21.",
+        runtimeFacts,
+        attachmentsPresent: false,
+    });
+    assert.equal(historicalChangelog.pass, true, "v4.2.2 / truthfulness_preflight: historical version text without current/timing claims must not trip");
+    const fabricatedTiming = truthfulnessPreflight({
+        task: "Explain why the report said v4.2.0.",
+        initialDraft: "When the workflow began, cross-review was running v4.2.0. It was bumped to v4.2.1 between R1 and R3.",
+        runtimeFacts,
+        attachmentsPresent: false,
+    });
+    assert.equal(fabricatedTiming.pass, false, "v4.2.2 / truthfulness_preflight: historical runtime timing narrative without snapshot evidence must trip");
+    const withStructuredEvidence = truthfulnessPreflight({
+        task: "Explain why the report said v4.2.0.",
+        initialDraft: "When the workflow began, cross-review was running v4.2.0. It was bumped to v4.2.1 between R1 and R3.",
+        runtimeFacts,
+        structuredEvidence: "Historical runtime snapshot from events.ndjson: workflow_start server_info version=4.2.0; later reload server_info version=4.2.1.",
+        attachmentsPresent: false,
+    });
+    assert.equal(withStructuredEvidence.pass, true, "v4.2.2 / truthfulness_preflight: structured evidence can satisfy historical timing claims");
+    const orchSrcTruth = fs.readFileSync(new URL("../src/core/orchestrator.ts", import.meta.url), "utf8");
+    const configSrcTruth = fs.readFileSync(new URL("../src/core/config.ts", import.meta.url), "utf8");
+    assert.ok(/export function truthfulnessPreflight\b/.test(orchSrcTruth), "v4.2.2 / truthfulness_preflight: truthfulnessPreflight must be exported");
+    assert.ok(/truthfulness_preflight_enabled/.test(orchSrcTruth) &&
+        /askPeers[\s\S]+truthfulnessPreflight/.test(orchSrcTruth) &&
+        /runUntilUnanimous[\s\S]+truthfulnessPreflight/.test(orchSrcTruth), "v4.2.2 / truthfulness_preflight: both askPeers and runUntilUnanimous must gate on config.truthfulness_preflight_enabled");
+    assert.ok(/boolEnv\("CROSS_REVIEW_TRUTHFULNESS_PREFLIGHT", true\)/.test(configSrcTruth), "v4.2.2 / truthfulness_preflight: CROSS_REVIEW_TRUTHFULNESS_PREFLIGHT env var must default ON");
+    console.log("[smoke] truthfulness_preflight_test: PASS");
+}
 // v3.5.0 (CRV2-1 + CRV2-6) — budget + max_rounds traceability.
 //
 // setSessionTraceability persists requested-vs-effective max_rounds and
@@ -6020,15 +6130,15 @@ assert.equal(Object.hasOwn(metrics.decision_quality, "undefined"), false);
     // SINGLE canonical pin. Negative pins (off-policy models that must never
     // appear) + positive pins (the exact lone-entry shape per peer).
     const a3ModelSrc = fs.readFileSync(new URL("../src/peers/model-selection.ts", import.meta.url), "utf8");
-    for (const offPolicyModel of ["deepseek-v4-flash", "gemini-3.1-pro-preview"]) {
+    for (const offPolicyModel of ["deepseek-v4-flash", "gemini-2.5-pro"]) {
         assert.ok(!a3ModelSrc.includes(`"${offPolicyModel}"`), `v3.7.2 / AUDIT-3: ${offPolicyModel} must not appear in the PRIORITY lists`);
     }
     for (const [peer, pin] of [
         ["codex", "gpt-5.5"],
-        ["claude", "claude-opus-4-7"],
-        ["gemini", "gemini-2.5-pro"],
+        ["claude", "claude-opus-4-8"],
+        ["gemini", "gemini-3.1-pro-preview"],
         ["deepseek", "deepseek-v4-pro"],
-        ["grok", "grok-4-latest"],
+        ["grok", "grok-4.3"],
         ["perplexity", "sonar-reasoning-pro"],
     ]) {
         assert.ok(new RegExp(`${peer}: \\["${pin}"\\]`).test(a3ModelSrc), `v3.7.2 / AUDIT-3: ${peer} PRIORITY must be the lone canonical pin ["${pin}"] (no fallback)`);