@lcv-ideas-software/cross-review 4.2.3 → 4.2.5

package/dist/src/core/config.d.ts

@@ -1,6 +1,6 @@
 import type { AppConfig, PeerId } from "./types.js";
-export declare const VERSION = "4.2.3";
-export declare const RELEASE_DATE = "2026-06-03";
+export declare const VERSION = "4.2.5";
+export declare const RELEASE_DATE = "2026-06-05";
 export declare const DEFAULT_MAX_OUTPUT_TOKENS = 20000;
 export declare function getLastFileConfigResult(): import("./file-config.js").ApplyFileConfigResult | undefined;
 export declare function loadConfig(): AppConfig;

package/dist/src/core/config.js

@@ -17,8 +17,8 @@ function expandHome(rawPath) {
     }
     return rawPath;
 }
-export const VERSION = "4.2.3";
-export const RELEASE_DATE = "2026-06-03";
+export const VERSION = "4.2.5";
+export const RELEASE_DATE = "2026-06-05";
 export const DEFAULT_MAX_OUTPUT_TOKENS = 20_000;
 const COST_RATE_ENV_PREFIX = {
     codex: "CROSS_REVIEW_OPENAI",

package/dist/src/core/orchestrator.d.ts

@@ -112,6 +112,7 @@ export interface TruthfulnessRuntimeFacts {
 export interface TruthfulnessPreflightResult {
     pass: boolean;
     reason: string;
+    issue_classes: TruthfulnessIssueClass[];
     current_state_claim_matched: boolean;
     historical_state_claim_matched: boolean;
     contradictions: string[];
@@ -121,6 +122,7 @@ export interface TruthfulnessPreflightResult {
     source_marker_found: boolean;
     runtime_facts_available: boolean;
 }
+export type TruthfulnessIssueClass = "runtime_contradiction" | "unsupported_current_state_claim" | "unsupported_historical_claim" | "fabrication_pattern";
 export declare function truthfulnessPreflight(params: {
     task: string;
     initialDraft?: string | undefined;

package/dist/src/core/orchestrator.js

@@ -323,6 +323,14 @@ const FABRICATED_ASSERTION_PATTERNS = [
     { pattern: /cargo\s+test\b/g, label: "cargo_test_assertion" },
     { pattern: /npm\s+run\s+(?:build|test|typecheck)\b/g, label: "npm_run_assertion" },
     { pattern: /index\s+[a-f0-9]{6,}\.{2}[a-f0-9]{6,}/g, label: "git_diff_index_hash" },
+    {
+        pattern: /\b[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}\b/gi,
+        label: "session_id_reference",
+    },
+    {
+        pattern: /https:\/\/github\.com\/[^\s)\]}>"']+/gi,
+        label: "github_url_reference",
+    },
     {
         pattern: /\b(?:workflow\s+(?:launched|started|dispatched|created)|(?:launched|started|dispatched)\s+(?:a\s+)?workflow)\b/gi,
         label: "workflow_dispatch_claim",
@@ -501,7 +509,12 @@ const VERSION_TOKEN_PATTERN = /\bv?(\d+\.\d+\.\d+(?:[-._a-z0-9]+)?)\b/gi;
 const ISO_DATE_TOKEN_PATTERN = /\b20\d{2}-\d{2}-\d{2}\b/g;
 const CURRENT_STATE_CLAIM_PATTERN = /\b(?:current|currently|actual|atual|runtime|production|prod|loaded|carregad[ao]s?|(?:is|are|est[aã]o?|esta|está)\s+(?:running|rodando))\b/i;
 const HISTORICAL_RUNTIME_TIMING_PATTERN = /\b(?:when\s+(?:the\s+)?(?:workflow|run|audit|session)\s+began|at\s+(?:workflow|run|audit|session)\s+start|between\s+r\d+\s+and\s+r\d+|bump(?:ed)?|started\s+on|was\s+running|quando\s+(?:o\s+)?(?:workflow|run|auditoria|sess[aã]o)\s+come[cç]ou|no\s+in[ií]cio\s+(?:do|da)\s+(?:workflow|run|auditoria|sess[aã]o)|estava\s+rodando)\b/i;
-const TRUTHFULNESS_SOURCE_MARKER_PATTERN = /\b(?:server_info|runtime_capabilities|probe_peers|capability_snapshot|session_read|session_events|provider docs|provider api)\b|https?:\/\/|\b[\w./-]+\.\w+:\d+\b|```/i;
+const TRUTHFULNESS_SOURCE_MARKER_PATTERN = /\b(?:server_info|runtime_capabilities|probe_peers|capability_snapshot|session_read|session_events|provider docs|provider api)\b|https?:\/\/|\b[\w./-]+\.\w+:\d+\b|\bevidence[\\/][\w./-]+\b|\bAttachment:\s*\S|\bL\d{2,}\b|```/i;
+const FABRICATION_PRONE_OPERATIONAL_CLAIM_PATTERN = /\b(?:triggered|dispatched|started|ran|launched|executei|rodei|disparei)\s+(?:the\s+|o\s+|a\s+)?(?:workflow|dispatch|deployment|deploy|ci|github actions?|pipeline)\b|\boperator authorization\b|\bautorizad[ao]\s+pelo\s+operador\b|\bconfirmed\s+(?:the\s+)?(?:remote\s+)?deployment\s+(?:succeeded|success)\b|\bconfirmei\s+(?:que\s+)?(?:o\s+)?deploy\b/i;
+function addIssueClass(issueClasses, issueClass) {
+    if (!issueClasses.includes(issueClass))
+        issueClasses.push(issueClass);
+}
 function normalizeVersionToken(value) {
     return value.trim().replace(/^v/i, "").toLowerCase();
 }
@@ -533,9 +546,17 @@ export function truthfulnessPreflight(params) {
     const runtimeFactsAvailable = Boolean(runtimeVersion || releaseDate);
     const contradictions = [];
     const unsupportedClaims = [];
+    const issueClasses = [];
     let currentStateClaimMatched = false;
     let historicalStateClaimMatched = false;
     for (const line of lines) {
+        if (FABRICATION_PRONE_OPERATIONAL_CLAIM_PATTERN.test(line) &&
+            !structuredEvidenceSupplied &&
+            !params.attachmentsPresent &&
+            !TRUTHFULNESS_SOURCE_MARKER_PATTERN.test(line)) {
+            addIssueClass(issueClasses, "fabrication_pattern");
+            unsupportedClaims.push(`fabrication-prone operational claim lacks provenance evidence: ${line.slice(0, 240)}`);
+        }
         const versions = uniqueMatches(VERSION_TOKEN_PATTERN, line);
         const dates = uniqueMatches(ISO_DATE_TOKEN_PATTERN, line);
         if (!versions.length && !dates.length)
@@ -546,6 +567,7 @@ export function truthfulnessPreflight(params) {
                 const expected = normalizeVersionToken(runtimeVersion);
                 for (const version of versions) {
                     if (normalizeVersionToken(version) !== expected) {
+                        addIssueClass(issueClasses, "runtime_contradiction");
                         contradictions.push(`current-state version claim ${version} contradicts runtime_version ${runtimeVersion}`);
                     }
                 }
@@ -553,29 +575,39 @@ export function truthfulnessPreflight(params) {
             if (releaseDate) {
                 for (const date of dates) {
                     if (date !== releaseDate) {
+                        addIssueClass(issueClasses, "runtime_contradiction");
                         contradictions.push(`current-state release_date claim ${date} contradicts runtime release_date ${releaseDate}`);
                     }
                 }
             }
             if (!runtimeFactsAvailable && !sourceMarkerFound && !params.attachmentsPresent) {
+                addIssueClass(issueClasses, "unsupported_current_state_claim");
                 unsupportedClaims.push(`current-state claim lacks runtime facts or source marker: ${line.slice(0, 240)}`);
             }
         }
         if (HISTORICAL_RUNTIME_TIMING_PATTERN.test(line)) {
             historicalStateClaimMatched = true;
             if (!structuredEvidenceSupplied && !params.attachmentsPresent) {
+                addIssueClass(issueClasses, "unsupported_historical_claim");
                 unsupportedClaims.push(`historical runtime timing claim lacks snapshot evidence: ${line.slice(0, 240)}`);
             }
         }
     }
     const pass = contradictions.length === 0 && unsupportedClaims.length === 0;
+    const detail = [...contradictions, ...unsupportedClaims].join("; ");
+    const evidenceState = `attachments_present=${params.attachmentsPresent}; ` +
+        `structured_evidence_supplied=${structuredEvidenceSupplied}; ` +
+        `source_marker_found=${sourceMarkerFound}; ` +
+        `runtime_facts_available=${runtimeFactsAvailable}`;
+    const remediation = "attach raw snapshot evidence with session_attach_evidence or pass a structured evidence field, then retry the truthfulness preflight";
     return {
         pass,
         reason: pass
             ? currentStateClaimMatched || historicalStateClaimMatched
                 ? "high-risk runtime truthfulness claims are consistent with runtime facts or backed by evidence"
                 : "no high-risk runtime truthfulness claim detected"
-            : [...contradictions, ...unsupportedClaims].join("; "),
+            : `${detail}. ${evidenceState}. Remediation: ${remediation}.`,
+        issue_classes: issueClasses,
         current_state_claim_matched: currentStateClaimMatched,
         historical_state_claim_matched: historicalStateClaimMatched,
         contradictions,
@@ -610,7 +642,7 @@ function leadShipModeDirective() {
         // relator is free to synthesize ANALYSIS (interpretation, design
         // rationale, prose) but MUST refuse to invent operational facts.
         "## Evidence Provenance Lock (HARD)",
-        "Operational evidence — git SHAs, content hashes, build outputs, test counts (e.g. `147 passed`), diff hunks, `git diff --check passed` style assertions, vite asset filenames with hex suffixes, `cargo test`/`npm run build`/`npm run typecheck` result lines, `git rev-parse HEAD` output, timestamps, file paths — has a PROVENANCE level. Two levels exist:",
+        "Operational evidence — git SHAs, content hashes, build outputs, test counts (e.g. `147 passed`), diff hunks, `git diff --check passed` style assertions, vite asset filenames with hex suffixes, `cargo test`/`npm run build`/`npm run typecheck` result lines, `git rev-parse HEAD` output, session IDs, GitHub URLs, timestamps, file paths — has a PROVENANCE level. Two levels exist:",
         "  - PROVENANCE-GRADE: raw command/tool output persisted via `session_attach_evidence` (visible to you below as `## Attached Evidence`), or a verbatim file slice with explicit path:line refs.",
         "  - NARRATIVE: the caller's natural-language summary in the task or in a prior draft (e.g. `I ran cargo test, 147 passed`).",
         "NARRATIVE is NOT evidence. The caller's claim that a command produced a specific result is unverified until the raw output is attached. You MUST NOT quote NARRATIVE operational claims as if they were verified evidence. You MAY summarize that the caller claims X; you MUST NOT assert that X happened.",
@@ -900,7 +932,7 @@ function budgetPreflightFailure(peer, provider, model, message) {
         latency_ms: 0,
     };
 }
-function truthfulnessPreflightFailure(peer, provider, model, message) {
+function truthfulnessPreflightFailure(peer, provider, model, message, issueClasses = []) {
     return {
         peer,
         provider,
@@ -910,6 +942,7 @@ function truthfulnessPreflightFailure(peer, provider, model, message) {
         retryable: false,
         attempts: 0,
         latency_ms: 0,
+        preflight_issue_classes: issueClasses,
     };
 }
 function financialControlsMissingMessage(missingVars) {
@@ -1975,7 +2008,7 @@ export class CrossReviewOrchestrator {
             if (!truthfulness.pass) {
                 const message = `Truthfulness preflight failed before any paid peer call: ${truthfulness.reason}`;
                 const promptFile = this.store.savePrompt(session.session_id, roundNumber, `# Cross Review - Truthfulness Preflight Block\n\n${message}`);
-                const rejected = selectAdapters(adapters, selectedPeers).map((adapter) => truthfulnessPreflightFailure(adapter.id, adapter.provider, adapter.model, message));
+                const rejected = selectAdapters(adapters, selectedPeers).map((adapter) => truthfulnessPreflightFailure(adapter.id, adapter.provider, adapter.model, message, truthfulness.issue_classes));
                 for (const failure of rejected) {
                     await this.store.savePeerFailure(session.session_id, roundNumber, failure);
                 }
@@ -2002,6 +2035,8 @@ export class CrossReviewOrchestrator {
                         historical_state_claim_matched: truthfulness.historical_state_claim_matched,
                         contradictions: truthfulness.contradictions,
                         unsupported_claims: truthfulness.unsupported_claims,
+                        issue_classes: truthfulness.issue_classes,
+                        structured_evidence_supplied: truthfulness.structured_evidence_supplied,
                         source_marker_found: truthfulness.source_marker_found,
                         runtime_facts_available: truthfulness.runtime_facts_available,
                         attachments_present: truthfulness.attachments_present,
@@ -3210,17 +3245,24 @@ export class CrossReviewOrchestrator {
                 runtimeFacts: runtimeTruthFacts(this.config),
             });
             if (!truthfulness.pass) {
+                const message = `Truthfulness preflight failed before any paid peer call: ${truthfulness.reason}`;
+                const rejected = selectAdapters(adapters, reviewerPeers.length ? reviewerPeers : selectedPeers).map((adapter) => truthfulnessPreflightFailure(adapter.id, adapter.provider, adapter.model, message, truthfulness.issue_classes));
+                for (const failure of rejected) {
+                    await this.store.savePeerFailure(session.session_id, 0, failure);
+                }
+                await this.store.recordPreflightFailure(session.session_id, rejected);
                 await this.store.finalize(session.session_id, "aborted", "needs_truthfulness_preflight");
                 this.emit({
                     type: "session.truthfulness_preflight_failed",
                     session_id: session.session_id,
-                    message: `Truthfulness preflight failed before any paid peer call: ${truthfulness.reason}`,
+                    message,
                     data: {
                         reason: truthfulness.reason,
                         current_state_claim_matched: truthfulness.current_state_claim_matched,
                         historical_state_claim_matched: truthfulness.historical_state_claim_matched,
                         contradictions: truthfulness.contradictions,
                         unsupported_claims: truthfulness.unsupported_claims,
+                        issue_classes: truthfulness.issue_classes,
                         structured_evidence_supplied: truthfulness.structured_evidence_supplied,
                         source_marker_found: truthfulness.source_marker_found,
                         runtime_facts_available: truthfulness.runtime_facts_available,
@@ -3330,6 +3372,52 @@ export class CrossReviewOrchestrator {
                 caller: callerForLottery,
             });
             await this.store.saveGeneration(session.session_id, 0, generation, "initial-draft");
+            if (this.config.truthfulness_preflight_enabled) {
+                const attachmentsPresent = this.store.readEvidenceAttachments(session.session_id, this.config.prompt.max_attached_evidence_chars).length > 0;
+                const truthfulness = truthfulnessPreflight({
+                    task: input.task,
+                    initialDraft: generation.text,
+                    structuredEvidence: input.evidence,
+                    attachmentsPresent,
+                    runtimeFacts: runtimeTruthFacts(this.config),
+                });
+                if (!truthfulness.pass) {
+                    const message = `Truthfulness preflight failed on lead-generated initial draft before reviewer peer calls: ${truthfulness.reason}`;
+                    const rejected = selectAdapters(adapters, reviewerPeers.length ? reviewerPeers : selectedPeers).map((adapter) => truthfulnessPreflightFailure(adapter.id, adapter.provider, adapter.model, message, truthfulness.issue_classes));
+                    for (const failure of rejected) {
+                        await this.store.savePeerFailure(session.session_id, 0, failure);
+                    }
+                    await this.store.recordPreflightFailure(session.session_id, rejected);
+                    await this.store.finalize(session.session_id, "aborted", "needs_truthfulness_preflight");
+                    this.emit({
+                        type: "session.truthfulness_preflight_failed",
+                        session_id: session.session_id,
+                        round: 0,
+                        peer: leadPeer,
+                        message,
+                        data: {
+                            reason: truthfulness.reason,
+                            current_state_claim_matched: truthfulness.current_state_claim_matched,
+                            historical_state_claim_matched: truthfulness.historical_state_claim_matched,
+                            contradictions: truthfulness.contradictions,
+                            unsupported_claims: truthfulness.unsupported_claims,
+                            issue_classes: truthfulness.issue_classes,
+                            structured_evidence_supplied: truthfulness.structured_evidence_supplied,
+                            source_marker_found: truthfulness.source_marker_found,
+                            runtime_facts_available: truthfulness.runtime_facts_available,
+                            attachments_present: truthfulness.attachments_present,
+                            lead_peer: leadPeer,
+                            round_kind: "initial-draft",
+                        },
+                    });
+                    return {
+                        session: this.store.read(session.session_id),
+                        final_text: undefined,
+                        converged: false,
+                        rounds: 0,
+                    };
+                }
+            }
             // v2.13.0: drift detection on initial-draft path. There is no
             // prior draft to fall back to here, so a drifted initial generation
             // aborts immediately. Only fires in `ship` mode — in `review` mode
@@ -3457,6 +3545,52 @@ export class CrossReviewOrchestrator {
                     caller: callerForLottery,
                 });
                 await this.store.saveGeneration(session.session_id, round, generation, "revision");
+                if (this.config.truthfulness_preflight_enabled) {
+                    const attachmentsPresent = this.store.readEvidenceAttachments(session.session_id, this.config.prompt.max_attached_evidence_chars).length > 0;
+                    const truthfulness = truthfulnessPreflight({
+                        task: input.task,
+                        initialDraft: generation.text,
+                        structuredEvidence: input.evidence,
+                        attachmentsPresent,
+                        runtimeFacts: runtimeTruthFacts(this.config),
+                    });
+                    if (!truthfulness.pass) {
+                        const message = `Truthfulness preflight failed on lead-generated revision before reviewer peer calls: ${truthfulness.reason}`;
+                        const rejected = selectAdapters(adapters, reviewerPeers.length ? reviewerPeers : selectedPeers).map((adapter) => truthfulnessPreflightFailure(adapter.id, adapter.provider, adapter.model, message, truthfulness.issue_classes));
+                        for (const failure of rejected) {
+                            await this.store.savePeerFailure(session.session_id, round + 1, failure);
+                        }
+                        await this.store.recordPreflightFailure(session.session_id, rejected, round + 1);
+                        await this.store.finalize(session.session_id, "aborted", "needs_truthfulness_preflight");
+                        this.emit({
+                            type: "session.truthfulness_preflight_failed",
+                            session_id: session.session_id,
+                            round: round + 1,
+                            peer: leadPeer,
+                            message,
+                            data: {
+                                reason: truthfulness.reason,
+                                current_state_claim_matched: truthfulness.current_state_claim_matched,
+                                historical_state_claim_matched: truthfulness.historical_state_claim_matched,
+                                contradictions: truthfulness.contradictions,
+                                unsupported_claims: truthfulness.unsupported_claims,
+                                issue_classes: truthfulness.issue_classes,
+                                structured_evidence_supplied: truthfulness.structured_evidence_supplied,
+                                source_marker_found: truthfulness.source_marker_found,
+                                runtime_facts_available: truthfulness.runtime_facts_available,
+                                attachments_present: truthfulness.attachments_present,
+                                lead_peer: leadPeer,
+                                round_kind: "revision",
+                            },
+                        });
+                        return {
+                            session: this.store.read(session.session_id),
+                            final_text: draft,
+                            converged: false,
+                            rounds: round,
+                        };
+                    }
+                }
                 // v2.23.0: empty-text degeneracy detection. Provider-side parser
                 // diagnostics (e.g. Anthropic extended-thinking returning only
                 // `thinking`/`redacted_thinking` blocks with no final `text` block,