@inflexa-ai/harness 0.35.7 → 0.36.0

package/dist/workflows/target-assessment/assemblers/orchestrator.js

@@ -1,945 +0,0 @@
-/**
- * Phase-4 deterministic assemblers.
- *
- * Each assembler maps Phase-1 / Phase-2 / Phase-3 outputs onto the
- * corresponding dossier section schema. Coverage is honored end-to-end: an
- * upstream that came back empty or was never consulted yields
- * `queried_no_data` / `not_loaded` rather than a fabricated section.
- *
- * Where a section applies a threshold of its own, the decision runs through
- * `coverageFromFilteredRows` over both counts — what survived and what the
- * threshold discarded — so a section our own filter emptied reports
- * `filtered` rather than an `available` section holding nothing. Deciding
- * coverage first and filtering the payload afterwards is the shape that loses
- * that distinction.
- *
- * Aggregate rows preserve their contributing evidence under `evidence: [...]`
- * arrays.
- */
-import { classifyClinicalEvidenceTrial, } from "../lib/clinical-evidence-attribution.js";
-import { inferTherapeuticArea, getBenchmarks, getDatasetAttribution } from "../../../tools/lib/clinical-benchmarks-client.js";
-import { inferModalityFromFamily } from "../../../tools/lib/protein-family-modality.js";
-import { resolveFamilySiblingUniprots, resolveOnTargetChemblIds } from "../lib/target-identity-filter.js";
-import { getDrugPrimaryTargetUniprots } from "../../../tools/lib/chembl-client.js";
-import { annotateOffTargetPanel } from "../lib/clinical-consequence-annotator.js";
-import { coverageFromFilteredRows, coverageFromRows } from "../coverage.js";
-import { toOrganSystems } from "../lib/impc-organ-map.js";
-import { fetchRegulatoryActions } from "../lib/regulatory-actions.js";
-import { HIGH_EXPRESSION_TPM_THRESHOLD } from "../lib/expression-constants.js";
-export { HIGH_EXPRESSION_TPM_THRESHOLD };
-import { resolveModulatorMoleculeType } from "../lib/dedup-modulators.js";
-import { searchFailedTrialsForDrugNames } from "../../../tools/lib/clinical-trials-client.js";
-import { aggregateFaersAcrossModulators, aggregateTrialAes, aggregateOffTargetPanel, aggregateClassPrecedent, buildOrganRollup, assembleDrugInteractions, assembleLiabilitySummary, claimSupportEvidence, meetsTpmFloor, isSafetyRelevant, resolveTissueOrganSystem, } from "./safety.js";
-import { assemblePreclinicalLiterature, assembleKoSupportingLiterature, tallyInternalReferences, assembleKeyPapers, assembleEvidenceTimeline, assembleTranslationalChain, assembleMolecularInteractions, assembleBiomarkerPotential, assembleResistanceEvidence, assembleCombinationEvidence, assembleEvidenceConflicts, assembleAdditionalEvidence, isSelfReference, isClinicalMeasurement, indicationCompositeScore, partitionTrialsByAttribution, classifyClinicalTrialConfidence, assembleDiscoveryTrials, dedupePpiPartners, } from "./literature.js";
-import { buildTrialOutcomeFilter, aggregateTrialOutcomes } from "./trials.js";
-import { buildFailureCategoryDiscriminated } from "./literature.js";
-const NOT_LOADED_PHASE5 = "Phase 5 synthesis not yet implemented";
-/** What emptied `clinical_development.trials` when nothing survived. */
-const ACTIVE_TRIALS_FILTER = "text match confidence below medium, attribution not eligible for toxicology aggregation, or attributed to a related family receptor";
-/** What emptied the two failed-trial sections when nothing survived. */
-const FAILED_TRIALS_FILTER = "attribution not eligible for toxicology aggregation, or attributed to a related family receptor";
-/** What emptied the two association sections when nothing survived. */
-const INDICATIONS_FILTER = "association restates the target itself, names a clinical measurement rather than a disease, or carries no scored evidence source";
-/** Sections the workflow body stamps between assembly and synthesis. */
-const NOT_LOADED_POST_ASSEMBLY = "assembled after Phase 4, from the segmented regulatory label signals";
-// Order matches evidence-priority rank: genetic > known_drug > animal_model > somatic > literature.
-function makeAssociationEvidence(a) {
-    const out = [];
-    if (a.geneticAssociationScore != null) {
-        out.push({
-            source: "opentargets:genetic_association",
-            score: a.geneticAssociationScore,
-            predicate: "associated_with",
-            is_human: true,
-            metadata: { disease_id: a.diseaseId },
-        });
-    }
-    if (a.knownDrugScore != null) {
-        out.push({
-            source: "opentargets:known_drug",
-            score: a.knownDrugScore,
-            predicate: "therapeutic_target",
-            is_human: true,
-            is_clinical: true,
-            metadata: { disease_id: a.diseaseId },
-        });
-    }
-    if (a.animalModelScore != null) {
-        out.push({
-            source: "opentargets:animal_model",
-            score: a.animalModelScore,
-            predicate: "validated_in_animal_model",
-            metadata: { disease_id: a.diseaseId },
-        });
-    }
-    if (a.somaticMutationScore != null) {
-        out.push({
-            source: "opentargets:somatic_mutation",
-            score: a.somaticMutationScore,
-            predicate: "somatic_mutation_associated",
-            is_human: true,
-            metadata: { disease_id: a.diseaseId },
-        });
-    }
-    if (a.literatureScore != null) {
-        out.push({
-            source: "opentargets:literature",
-            score: a.literatureScore,
-            predicate: "literature_co_occurrence",
-            metadata: { disease_id: a.diseaseId },
-        });
-    }
-    return out;
-}
-// ── Section assemblers ───────────────────────────────────────────────
-export function assembleEntity(resolved) {
-    return {
-        canonicalId: resolved.canonicalId,
-        symbol: resolved.geneSymbol,
-        displayName: resolved.approvedName,
-        entityType: "gene",
-        ontology: resolved.canonicalOntology === "hgnc" ? "HGNC" : "Ensembl",
-        identifiers: {
-            hgnc: resolved.ids.hgnc ?? undefined,
-            ensembl: resolved.ids.ensembl ?? undefined,
-            uniprot: resolved.ids.uniprot ?? undefined,
-            chembl: resolved.ids.chembl ?? undefined,
-            ncbiGene: resolved.ids.entrez ?? undefined,
-        },
-        synonyms: resolved.synonyms,
-        proteinFamily: resolved.proteinFamily ?? undefined,
-    };
-}
-// Maps ChEMBL molecule_type (lowercase) to the tractability modality bucket.
-const MOLECULE_TYPE_TO_MODALITY = {
-    "small molecule": "small_molecule",
-    antibody: "antibody",
-    "antibody-drug conjugate": "antibody",
-    protein: "other_clinical",
-    peptide: "other_clinical",
-    enzyme: "other_clinical",
-};
-function normalizeDrugMoleculeType(drug) {
-    const name = (drug.drug_name ?? "").trim().toUpperCase();
-    // SMC021 is oral salmon calcitonin, not a non-peptide small molecule. Some
-    // source paths can describe the formulation as oral and accidentally leave a
-    // small-molecule annotation downstream; keep the therapeutic substance class.
-    if (name === "SMC021" || name.includes("SALMON CALCITONIN") || name.includes("CALCITONIN SALMON")) {
-        return "Peptide";
-    }
-    return drug.molecule_type;
-}
-/**
- * Pure helper that flips `has_approved_drug=true` and populates
- * `approved_drug_ids` on any modality that has a max_phase=4 drug in
- * the provided list. Unrecognised molecule_types are silently ignored.
- */
-export function enrichTractabilityModalitiesWithApprovals(modalities, drugs) {
-    const approvedByModality = new Map();
-    for (const d of drugs) {
-        if ((d.max_phase ?? 0) < 4)
-            continue;
-        const mod = MOLECULE_TYPE_TO_MODALITY[(normalizeDrugMoleculeType(d) ?? "").toLowerCase()];
-        if (!mod)
-            continue;
-        const arr = approvedByModality.get(mod) ?? [];
-        arr.push(d.drug_id);
-        approvedByModality.set(mod, arr);
-    }
-    return modalities.map((m) => {
-        const ids = approvedByModality.get(m.modality);
-        if (!ids || ids.length === 0)
-            return m;
-        return { ...m, has_approved_drug: true, approved_drug_ids: [...new Set(ids)].sort() };
-    });
-}
-export function assembleTractability(resolved, ot, drugs) {
-    const familyFallback = () => {
-        const fallback = inferModalityFromFamily(resolved.proteinFamily);
-        return {
-            coverage: "available",
-            data: {
-                modalities: [
-                    {
-                        modality: "small_molecule",
-                        levels: [],
-                        has_approved_drug: false,
-                        has_clinical_stage: false,
-                        is_inferred_from_family: true,
-                    },
-                ],
-                preferred_modality: fallback.preferred_modality,
-                drug_molecule_types_present: [],
-            },
-            inference_path: `family-fallback (${fallback.source}: ${fallback.family ?? "default"}) — ${fallback.rationale}`,
-        };
-    };
-    if (ot.coverage !== "available")
-        return familyFallback();
-    const t = ot.data.tractability;
-    if (!t)
-        return familyFallback();
-    const modalities = [
-        {
-            modality: "small_molecule",
-            levels: t.smallMolecule ? ["validated"] : [],
-            has_approved_drug: false,
-            has_clinical_stage: !!t.smallMolecule,
-            is_inferred_from_family: false,
-        },
-        {
-            modality: "antibody",
-            levels: t.antibody ? ["validated"] : [],
-            has_approved_drug: false,
-            has_clinical_stage: !!t.antibody,
-            is_inferred_from_family: false,
-        },
-        {
-            modality: "other_clinical",
-            levels: t.otherModalities ? ["validated"] : [],
-            has_approved_drug: false,
-            has_clinical_stage: !!t.otherModalities,
-            is_inferred_from_family: false,
-        },
-    ];
-    const preferred = t.smallMolecule ? "small_molecule" : t.antibody ? "antibody" : t.otherModalities ? "other_clinical" : null;
-    const modalitiesEnriched = enrichTractabilityModalitiesWithApprovals(modalities, drugs);
-    // Collect molecule types from approved/clinical drugs for the tractability field.
-    const drugMoleculeTypes = [...new Set(drugs.map((d) => d.molecule_type).filter((m) => !!m))];
-    return {
-        coverage: "available",
-        data: {
-            modalities: modalitiesEnriched,
-            preferred_modality: preferred,
-            drug_molecule_types_present: drugMoleculeTypes,
-        },
-    };
-}
-// ── Top-level dossier assembly ──────────────────────────────────────
-/**
- * Build a body-only dossier from Phase-2 (and optionally Phase-3)
- * outputs. Synthesis-dependent fields stay `not_loaded` until Phase 5
- * stamps them in `phase5-persist.ts`. The derived sub-tree is NOT
- * attached here — phase5-persist computes it after synthesis is stamped
- * so the final derived values reflect liability bullets and safety flags.
- */
-export async function assembleDossier(pool, phase2, phase3, annotatorDeps, logger) {
-    const fanout = phase3?.fanout;
-    const faersAgg = aggregateFaersAcrossModulators(fanout);
-    const trialAesAgg = aggregateTrialAes(fanout);
-    // Off-target panel needs the assessment UniProt and gene symbol: UniProt to
-    // recognise alternate ChEMBL IDs as self-hits, gene symbol to detect
-    // obligate-cofactor heterodimers (e.g., CALCR/RAMP3 = AMY3 for CALCR).
-    const assessmentUniprot = phase2.phase1.resolved.ids.uniprot ?? "";
-    const geneSymbol = phase2.phase1.resolved.geneSymbol ?? "";
-    const familyComplexesCollector = phase2.phase1.collectors.familyComplexes;
-    const familyComplexesBundle = familyComplexesCollector.coverage === "available" ? familyComplexesCollector.data : null;
-    const [onTargetChemblIds, familySiblingUniprots] = await Promise.all([
-        resolveOnTargetChemblIds(assessmentUniprot, logger),
-        resolveFamilySiblingUniprots(assessmentUniprot || geneSymbol, logger),
-    ]);
-    const offTargetPanel = aggregateOffTargetPanel(phase2, fanout, assessmentUniprot, geneSymbol, familyComplexesBundle, onTargetChemblIds);
-    if (offTargetPanel) {
-        await annotateOffTargetPanel(pool, offTargetPanel, geneSymbol, annotatorDeps);
-    }
-    const offTargetRows = offTargetPanel?.rows ?? null;
-    const classP = aggregateClassPrecedent(phase2, fanout);
-    // Enrich modulator metadata before any downstream assembly so that
-    // tractability, drug_interactions, and class-precedent all see the same
-    // PubChem-refined molecule_type for entries that ChEMBL labelled "Unknown".
-    const chemblModsCollector = phase2.phase1.collectors.chemblModulators;
-    const enrichedModulators = chemblModsCollector.coverage === "available" ? await resolveModulatorMoleculeType(chemblModsCollector.data.modulators) : [];
-    // drugInteractionRows must be computed before buildOrganRollup so we can
-    // derive the drug IDs for regulatory_actions lookup and pass them in.
-    const drugInteractionRows = assembleDrugInteractions(phase2, enrichedModulators);
-    const regulatoryActionRows = await fetchRegulatoryActions((drugInteractionRows ?? []).map((d) => ({ chemblId: d.drug_id, name: d.drug_name })));
-    const organRollupRows = buildOrganRollup(faersAgg, trialAesAgg, offTargetRows, classP, regulatoryActionRows.length > 0 ? regulatoryActionRows : null);
-    const preclinicalLitData = assemblePreclinicalLiterature(phase2);
-    const koSupportingLiterature = assembleKoSupportingLiterature(phase2);
-    const internalRefCounts = tallyInternalReferences([
-        koSupportingLiterature,
-        ...(organRollupRows ?? []).map((r) => r.evidence),
-        ...(offTargetRows ?? []).map((r) => claimSupportEvidence(r.support)),
-    ]);
-    const keyPapersRows = assembleKeyPapers(phase2, internalRefCounts);
-    const evidenceTimeline = assembleEvidenceTimeline(phase2);
-    const translationalChain = assembleTranslationalChain(phase2);
-    const attrCtx = {
-        assessmentUniprot,
-        assessmentSymbol: phase2.phase1.resolved.geneSymbol,
-        familyUniprots: familySiblingUniprots,
-        drugTargetResolver: getDrugPrimaryTargetUniprots,
-    };
-    // Derive known class drug names from ChEMBL modulators (uppercase) for
-    // drug_in_class_match classification inside assembleDiscoveryTrials.
-    const knownClassDrugNames = new Set((drugInteractionRows ?? []).map((r) => r.drug_name.toUpperCase()));
-    // Build the attribution filter for aggregateTrialOutcomes by partitioning
-    // all ctgov trials (active + failed) with the same attribution logic used
-    // for clinical_development.trials. This drops erenumab/paralog trials from
-    // clinical_development.outcomes when their ChEMBL IDs resolve to a related
-    // receptor rather than the assessment target.
-    const trialOutcomeFilter = await buildTrialOutcomeFilter(phase2, attrCtx, knownClassDrugNames, phase2.phase1.resolved.geneSymbol);
-    const trialOutcomesRows = aggregateTrialOutcomes(fanout, trialOutcomeFilter);
-    const resolved = phase2.phase1.resolved;
-    const ot = phase2.phase1.collectors.opentargets;
-    const ctgov = phase2.phase1.collectors.ctgov;
-    const clinvar = phase2.phase1.collectors.clinvar;
-    const cbioportal = phase2.phase1.collectors.cbioportal;
-    const expression = phase2.phase1.collectors.expressionHuman;
-    const expressionMulti = phase2.phase1.collectors.expressionMultiSpecies;
-    const impc = phase2.phase1.collectors.impc;
-    const pubmed = phase2.phase1.collectors.pubmedIndex;
-    const pathways = phase2.phase1.collectors.pathways;
-    const ppi = phase2.phase1.collectors.stringPpi;
-    // Every trial row carries its attribution from assembly onward: the
-    // evidence role decides whether a row supports the target or is context.
-    const knownClassDrugs = await Promise.all((drugInteractionRows ?? []).map(async (d) => {
-        const resolvedTargets = d.drug_id ? await getDrugPrimaryTargetUniprots(d.drug_id).catch(() => []) : [];
-        return {
-            name: d.drug_name,
-            moleculeChemblId: d.drug_id,
-            targetUniprots: resolvedTargets.length > 0 ? resolvedTargets : assessmentUniprot ? [assessmentUniprot] : [],
-        };
-    }));
-    const programsCollector = phase2.phase1.collectors.therapeuticPrograms;
-    const therapeuticPrograms = programsCollector?.coverage === "available"
-        ? programsCollector.data.programs.map((program) => ({
-            programId: program.programId,
-            name: program.name,
-            targetSymbol: program.targetSymbol,
-            targetUniprot: program.targetUniprot,
-            modality: program.modality,
-            nctIds: program.nctIds,
-        }))
-        : [];
-    const evidenceAttribution = {
-        assessmentSymbol: geneSymbol,
-        assessmentUniprot,
-        familyUniprots: familySiblingUniprots,
-        knownClassDrugs,
-        therapeuticPrograms,
-    };
-    const trialSourceByNct = new Map();
-    if (ctgov.coverage === "available") {
-        for (const t of [...ctgov.data.active, ...ctgov.data.failed])
-            trialSourceByNct.set(t.nctId, t);
-    }
-    const attachTrialAttribution = (row) => {
-        const source = trialSourceByNct.get(row.nct_id);
-        const classified = source
-            ? classifyClinicalEvidenceTrial(source, evidenceAttribution)
-            : {
-                attribution: {
-                    relationship: "unknown",
-                    evidence_role: "excluded",
-                    basis: [{ kind: "text_match", source: "ctgov" }],
-                    resolved_interventions: [],
-                    exclusion_reason: "no CT.gov record was available to attribute this trial to the assessed target",
-                },
-                eligible_for_toxicology_aggregation: false,
-            };
-        return {
-            ...row,
-            attribution: classified.attribution,
-            eligible_for_toxicology_aggregation: classified.eligible_for_toxicology_aggregation,
-        };
-    };
-    const discoveryTrials = await assembleDiscoveryTrials(phase2, phase2.phase1.resolved.geneSymbol, attrCtx, knownClassDrugNames, attachTrialAttribution);
-    const indicationsRaw = ot.coverage === "available" ? ot.data.associations : [];
-    const filteredIndications = indicationsRaw.filter((a) => !isSelfReference(resolved.canonicalId, a.diseaseId, resolved.geneSymbol, a.diseaseName) && !isClinicalMeasurement(a.diseaseName));
-    const ta = inferTherapeuticArea(filteredIndications.map((a) => a.diseaseName));
-    const benchmarks = getBenchmarks(ta);
-    const datasetAttr = getDatasetAttribution();
-    const offTissueCandidates = expression.coverage === "available"
-        ? expression.data.tissues.filter((t) => meetsTpmFloor(t.tissueLabel, t.value ?? 0) && isSafetyRelevant(t.tissueLabel, t.organSystem))
-        : [];
-    // A tissue whose anatomy does not resolve onto an organ system is dropped
-    // rather than filed under a neighbouring organ; the count is reported so
-    // the section does not overstate its own completeness.
-    const offTissueRows = offTissueCandidates.flatMap((t) => {
-        const organ = resolveTissueOrganSystem(t.tissueLabel, t.organSystem);
-        if (!organ)
-            return [];
-        const support = {
-            state: "unknown",
-            reason: `${t.tissueLabel} expression is a measured value from the expression atlas, not a citable per-tissue record`,
-        };
-        return [{ tissue: t.tissueLabel, organ, tpm: t.value ?? 0, support }];
-    });
-    const offTissueDroppedCount = offTissueCandidates.length - offTissueRows.length;
-    // Candidates that all failed to resolve leave the section empty for a
-    // reason of our own making, not because the atlas held nothing.
-    const offTissueSection = expression.coverage !== "available"
-        ? { coverage: "queried_no_data" }
-        : offTissueRows.length > 0
-            ? {
-                coverage: "available",
-                data: { rows: offTissueRows },
-                ...(offTissueDroppedCount > 0 ? { dropped_count: offTissueDroppedCount } : {}),
-            }
-            : offTissueDroppedCount > 0
-                ? {
-                    coverage: "filtered",
-                    filter: "tissue anatomy did not resolve onto an organ system",
-                    dropped_count: offTissueDroppedCount,
-                }
-                : { coverage: "queried_no_data" };
-    const drugsForTractability = enrichedModulators.map((m) => ({
-        drug_id: m.moleculeChemblId,
-        drug_name: m.preferredName,
-        molecule_type: m.moleculeType,
-        max_phase: m.maxPhase,
-    }));
-    const tractabilitySection = assembleTractability(resolved, ot, drugsForTractability);
-    const preferredModality = tractabilitySection.coverage === "available" ? tractabilitySection.data.preferred_modality : null;
-    const liabilitySummary = assembleLiabilitySummary(phase2, {
-        offTargetRows,
-        preferredModality,
-    });
-    const allRowCandidates = filteredIndications.slice(0, 50).map((a) => {
-        const sources = [
-            a.geneticAssociationScore != null ? "ot_genetics" : null,
-            a.knownDrugScore != null ? "ot_known_drug" : null,
-            a.animalModelScore != null ? "ot_animal_model" : null,
-            a.somaticMutationScore != null ? "ot_somatic_mutation" : null,
-            a.literatureScore != null ? "literature" : null,
-        ].filter((s) => s !== null);
-        const evidence = makeAssociationEvidence(a);
-        return { a, sources, evidence };
-    });
-    const supportedCandidates = allRowCandidates.filter(({ sources, evidence }) => sources.length > 0 || evidence.length > 0);
-    const unsupportedCandidates = allRowCandidates.filter(({ sources, evidence }) => sources.length === 0 && evidence.length === 0);
-    const indicationRows = supportedCandidates.map(({ a, sources, evidence }) => {
-        const uniquePaperCount = a.literaturePmids?.length ?? 0;
-        const { composite, breakdown } = indicationCompositeScore(a.score, sources, uniquePaperCount);
-        return {
-            disease_id: a.diseaseId,
-            disease_name: a.diseaseName,
-            composite_score: composite,
-            composite_score_breakdown: breakdown,
-            evidence_score: a.score,
-            source_count: sources.length,
-            unique_paper_count: uniquePaperCount,
-            sources,
-            evidence,
-        };
-    });
-    // Run partitionTrialsByAttribution on active trials first to detect off-target
-    // before building clinical_development.trials and discovery_trials sections.
-    // Since ctgov interventions are plain strings (no ChEMBL IDs), the resolver
-    // cannot fire and all trials land in primary — off-target detection activates
-    // only when ChEMBL IDs are available downstream.
-    const activeTrialsPartitioned = ctgov.coverage === "available" ? await partitionTrialsByAttribution(ctgov.data.active, attrCtx) : null;
-    // Everything CT.gov offered before any threshold of ours ran — the baseline
-    // the trials section measures its own drops against.
-    const activeTrialCandidateCount = (activeTrialsPartitioned?.primary.length ?? 0) + (activeTrialsPartitioned?.related.length ?? 0);
-    // Map ctgov active trials to clinical-trial row shape with text-based confidence,
-    // then remove rows identified as off-target by the attribution filter.
-    // Drug-in-class matches (calcitonin drugs for CALCR, GLP-1 drugs for GLP1R, etc.)
-    // are promoted to "medium" via classifyClinicalTrialConfidence so they survive
-    // the ≥ medium threshold even when the gene symbol is absent from intervention names.
-    const clinicalTrialRows = activeTrialsPartitioned !== null
-        ? activeTrialsPartitioned.primary
-            .map((t) => {
-            const text_confidence = classifyClinicalTrialConfidence(t, resolved.geneSymbol, knownClassDrugNames);
-            return {
-                nct_id: t.nctId,
-                title: t.title,
-                phase: t.phase,
-                status: t.status,
-                conditions: t.conditions,
-                start_date: t.startDate,
-                completion_date: t.primaryCompletionDate,
-                match_confidence: text_confidence,
-            };
-        })
-            .filter((r) => r.match_confidence !== "low")
-        : null;
-    // Build failed-trial rows with off-target partitioning. The gene-symbol
-    // query in collectCtgov misses terminated trials whose intervention is
-    // recorded only by sponsor code or brand name (e.g., SMC021 for oral
-    // salmon calcitonin), so we supplement with a drug-name query against
-    // CT.gov keyed by the assembled drug interactions. The function self-
-    // throttles via withHost("ctgov", ...) on its per-drug call, so the
-    // assembler does not need its own outer wrapper. Per-drug failures
-    // inside searchFailedTrialsForDrugNames are isolated.
-    const drugNameFailedTrials = drugInteractionRows && drugInteractionRows.length > 0
-        ? await searchFailedTrialsForDrugNames(drugInteractionRows.map((d) => d.drug_name).filter(Boolean))
-            .then((r) => r.trials)
-            .catch(() => [])
-        : [];
-    const failedById = new Map();
-    if (ctgov.coverage === "available") {
-        for (const t of ctgov.data.failed) {
-            failedById.set(t.nctId, {
-                nctId: t.nctId,
-                title: t.title,
-                interventions: t.interventions,
-                conditions: t.conditions,
-            });
-        }
-    }
-    for (const t of drugNameFailedTrials) {
-        if (!failedById.has(t.nctId)) {
-            failedById.set(t.nctId, {
-                nctId: t.nctId,
-                title: t.title,
-                interventions: t.interventions,
-                conditions: t.conditions,
-            });
-        }
-    }
-    const rawFailedTrials = [...failedById.values()].map((t) => {
-        const fanoutItem = fanout?.perTrialAes.results.find((r) => r.coverage === "available" && r.data.nctId === t.nctId);
-        const whyStopped = fanoutItem?.coverage === "available" ? (fanoutItem.data.whyStopped ?? "") : "";
-        return {
-            nctId: t.nctId,
-            title: t.title,
-            interventions: t.interventions,
-            conditions: t.conditions,
-            why_stopped: whyStopped,
-            failure_category: buildFailureCategoryDiscriminated(whyStopped),
-            classifier: "rules",
-        };
-    });
-    const failedTrialsPartitioned = rawFailedTrials.length > 0 ? await partitionTrialsByAttribution(rawFailedTrials, attrCtx) : null;
-    const failedTrialRows = failedTrialsPartitioned?.primary.map((t) => ({
-        nct_id: t.nctId,
-        title: t.title,
-        why_stopped: t.why_stopped,
-        failure_category: t.failure_category,
-        classifier: t.classifier,
-    })) ?? [];
-    const failedRelatedRows = failedTrialsPartitioned?.related.map((t) => ({
-        nct_id: t.nctId,
-        title: t.title,
-        why_stopped: t.why_stopped,
-        failure_category: t.failure_category,
-        classifier: t.classifier,
-    })) ?? [];
-    const attributedFailedRows = failedTrialRows.map(attachTrialAttribution);
-    const attributedFailedRelatedRows = failedRelatedRows.map(attachTrialAttribution);
-    // The same partition the clinical-development section reports, shared with the
-    // safety lens: a related-receptor failure is context for both, and it must be
-    // told apart from a target-attributed one in both.
-    const eligibleFailedRows = attributedFailedRows.filter((r) => r.eligible_for_toxicology_aggregation);
-    const excludedFailedRows = [...attributedFailedRows.filter((r) => !r.eligible_for_toxicology_aggregation), ...attributedFailedRelatedRows];
-    // `clinical_development.failed_trials` and `safety_profile.failed_trials_safety_lens`
-    // report the same partition under two headings, so the section is decided once.
-    // Two constructions of it would let the same condition report two coverages.
-    const failedTrialsSection = ctgov.coverage === "available"
-        ? coverageFromFilteredRows({
-            data: {
-                rows: eligibleFailedRows,
-                excluded_rows: excludedFailedRows,
-            },
-            retainedCount: eligibleFailedRows.length,
-            droppedCount: excludedFailedRows.length,
-            filter: FAILED_TRIALS_FILTER,
-            emptyReason: "ClinicalTrials.gov returned no terminated or withdrawn trials for the assessment target",
-        })
-        : { coverage: "queried_no_data", error: { message: "ClinicalTrials.gov unavailable" } };
-    // Typed explicitly so literal string coverage values narrow correctly.
-    // Phase-5 persist attaches the derived sub-tree after synthesis is stamped.
-    const dossierBody = {
-        generated_at: new Date().toISOString(),
-        entity: assembleEntity(resolved),
-        liability_summary: liabilitySummary,
-        tractability: tractabilitySection,
-        indications: ot.coverage === "available"
-            ? coverageFromFilteredRows({
-                data: {
-                    rows: indicationRows,
-                    excluded_unsupported_count: unsupportedCandidates.length,
-                    ...(unsupportedCandidates.length > 0
-                        ? {
-                            unsupported_associations: unsupportedCandidates.map(({ a }) => ({
-                                disease_id: a.diseaseId,
-                                disease_name: a.diseaseName,
-                            })),
-                        }
-                        : {}),
-                },
-                retainedCount: indicationRows.length,
-                // The head cap is a carrying limit rather than a judgement, so
-                // only the two content filters count against completeness.
-                droppedCount: indicationsRaw.length - filteredIndications.length + unsupportedCandidates.length,
-                filter: INDICATIONS_FILTER,
-                emptyReason: "Open Targets returned no disease associations for the target",
-            })
-            : {
-                coverage: "queried_no_data",
-                error: {
-                    message: ot.coverage === "queried_no_data" && ot.error?.message ? ot.error.message : "Open Targets unavailable",
-                },
-            },
-        drug_interactions: drugInteractionRows
-            ? { coverage: "available", data: { rows: drugInteractionRows } }
-            : { coverage: "queried_no_data", error: { message: "no ChEMBL modulators available" } },
-        clinical_development: {
-            trials: clinicalTrialRows === null
-                ? { coverage: "queried_no_data", error: { message: "ClinicalTrials.gov unavailable" } }
-                : (() => {
-                    const attributed = clinicalTrialRows.map(attachTrialAttribution);
-                    const relatedTargetTrials = (activeTrialsPartitioned?.related ?? []).map((t) => attachTrialAttribution({
-                        nct_id: t.nctId,
-                        title: t.title,
-                        phase: t.phase,
-                        status: t.status,
-                        conditions: t.conditions,
-                        start_date: t.startDate,
-                        completion_date: t.primaryCompletionDate,
-                        match_confidence: "off_target",
-                    }));
-                    const rows = attributed.filter((r) => r.eligible_for_toxicology_aggregation);
-                    return coverageFromFilteredRows({
-                        data: {
-                            rows,
-                            excluded_rows: [...attributed.filter((r) => !r.eligible_for_toxicology_aggregation), ...relatedTargetTrials],
-                            selection_criteria: {
-                                derived_from: "analytics.discovery_trials",
-                                min_confidence: "medium",
-                                excluded_off_target_count: activeTrialsPartitioned?.related.length ?? 0,
-                            },
-                        },
-                        retainedCount: rows.length,
-                        // Everything CT.gov returned for the target, against what survived:
-                        // the low-confidence cut, the attribution cut, and the related-receptor
-                        // partition are all filters of ours.
-                        droppedCount: activeTrialCandidateCount - rows.length,
-                        filter: ACTIVE_TRIALS_FILTER,
-                        emptyReason: "ClinicalTrials.gov returned no trials for the assessment target",
-                    });
-                })(),
-            outcomes: trialOutcomesRows
-                ? { coverage: "available", data: { rows: trialOutcomesRows.map(attachTrialAttribution) } }
-                : phase3
-                    ? { coverage: "queried_no_data", error: { message: "no per-trial outcomes available" } }
-                    : { coverage: "not_loaded", reason: "Phase-3 fan-out not run" },
-            failed_trials: failedTrialsSection,
-            benchmarks: {
-                therapeutic_area: benchmarks.therapeutic_area,
-                fallback_to_all_areas: benchmarks.source === "fallback",
-                phase_transitions: benchmarks.transitions,
-                likelihood_of_approval: benchmarks.transitions.phase1_to_approval,
-                source_attribution: `${datasetAttr.source} (${datasetAttr.data_version}, ${datasetAttr.data_window})`,
-            },
-        },
-        safety_profile: {
-            organ_rollup: organRollupRows
-                ? { coverage: "available", data: { rows: organRollupRows } }
-                : phase3
-                    ? { coverage: "queried_no_data", error: { message: "no organ-level signals" } }
-                    : { coverage: "not_loaded", reason: "Phase-3 fan-out not run" },
-            faers: (() => {
-                const classPrecedentSection = classP
-                    ? { coverage: "available", data: classP }
-                    : phase3
-                        ? { coverage: "queried_no_data", error: { message: "no class precedent data" } }
-                        : { coverage: "not_loaded", reason: "Phase-3 fan-out not run" };
-                const rawFaers = faersAgg
-                    ? { coverage: "available", data: faersAgg }
-                    : phase3
-                        ? { coverage: "queried_no_data", error: { message: "FAERS returned no reports" } }
-                        : { coverage: "not_loaded", reason: "Phase-3 fan-out not run" };
-                return reconcileFaersCoverage({ faers: rawFaers, class_precedent: classPrecedentSection });
-            })(),
-            trial_aes: trialAesAgg
-                ? { coverage: "available", data: trialAesAgg }
-                : phase3
-                    ? { coverage: "queried_no_data", error: { message: "no trial AE data" } }
-                    : { coverage: "not_loaded", reason: "Phase-3 fan-out not run" },
-            off_target_panel: offTargetPanel
-                ? { coverage: "available", data: offTargetPanel }
-                : phase3
-                    ? { coverage: "queried_no_data", error: { message: "no off-target data" } }
-                    : { coverage: "not_loaded", reason: "Phase-3 fan-out not run" },
-            // The safety lens carries the same two buckets as
-            // `clinical_development.failed_trials`. Merging them would put a
-            // related-receptor termination in the same list as a target-attributed
-            // one, and the reader has no other way to tell them apart.
-            failed_trials_safety_lens: failedTrialsSection,
-            class_precedent: classP
-                ? { coverage: "available", data: classP }
-                : phase3
-                    ? { coverage: "queried_no_data", error: { message: "no class precedent data" } }
-                    : { coverage: "not_loaded", reason: "Phase-3 fan-out not run" },
-            // Stamped by Phase 5 after the safety-flags-trail agent runs.
-            target_organ_liabilities: { coverage: "not_loaded", reason: NOT_LOADED_PHASE5 },
-            regulatory_actions: regulatoryActionRows.length > 0
-                ? { coverage: "available", data: { rows: regulatoryActionRows } }
-                : { coverage: "queried_no_data" },
-        },
-        // The corroboration fold reads the FDA label signals segmented after
-        // Phase 4, and the claim investigation reads the fold, so both are
-        // stamped by the workflow body before synthesis runs.
-        safety_corroboration: { coverage: "not_loaded", reason: NOT_LOADED_POST_ASSEMBLY },
-        claim_investigation: { coverage: "not_loaded", reason: "the claim investigation runs after Phase 4, over the corroborated organ claims" },
-        off_tissue_risk: offTissueSection,
-        off_target_panel: offTargetPanel
-            ? { coverage: "available", data: offTargetPanel }
-            : phase3
-                ? { coverage: "queried_no_data", error: { message: "no off-target data" } }
-                : { coverage: "not_loaded", reason: "Phase-3 fan-out not run" },
-        reference_biology: {
-            therapeutic_area_associations: ot.coverage === "available"
-                ? coverageFromFilteredRows({
-                    data: {
-                        rows: filteredIndications.slice(0, 200).map((a) => {
-                            const evidence = makeAssociationEvidence(a);
-                            return {
-                                partner_id: a.diseaseId,
-                                partner_name: a.diseaseName,
-                                predicate: a.geneticAssociationScore != null
-                                    ? "associated_with"
-                                    : a.knownDrugScore != null
-                                        ? "therapeutic_target"
-                                        : a.animalModelScore != null
-                                            ? "validated_in_animal_model"
-                                            : "literature_co_occurrence",
-                                best_score: a.score,
-                                source_count: [
-                                    a.geneticAssociationScore,
-                                    a.knownDrugScore,
-                                    a.animalModelScore,
-                                    a.somaticMutationScore,
-                                    a.literatureScore,
-                                ].filter((s) => s != null).length,
-                                paper_count: a.literaturePmids?.length ?? 0,
-                                evidence,
-                            };
-                        }),
-                    },
-                    retainedCount: Math.min(filteredIndications.length, 200),
-                    droppedCount: indicationsRaw.length - filteredIndications.length,
-                    filter: INDICATIONS_FILTER,
-                    emptyReason: "Open Targets returned no disease associations for the target",
-                })
-                : { coverage: "queried_no_data" },
-            molecular_interactions: (() => {
-                const rows = assembleMolecularInteractions(phase2, fanout);
-                return rows ? { coverage: "available", data: { rows } } : { coverage: "queried_no_data" };
-            })(),
-            biomarker_potential: (() => {
-                const rows = assembleBiomarkerPotential(phase2);
-                return rows ? { coverage: "available", data: { rows } } : { coverage: "queried_no_data" };
-            })(),
-            genetic_alterations: {
-                somatic: cbioportal.coverage === "available"
-                    ? coverageFromRows(cbioportal.data.rows.map((r) => ({
-                        cancer_type: r.cancerTypeName,
-                        cohort: r.studies.join(", "),
-                        mutation_count: r.mutatedSamples,
-                        total_samples: r.totalSamples,
-                        frequency: r.frequency,
-                        source: "cbioportal",
-                    })), { reason: "cBioPortal returned no somatic mutation cohorts for this gene" })
-                    : { coverage: "queried_no_data" },
-                clinvar: clinvar.coverage === "available"
-                    ? coverageFromRows(clinvar.data.variants.map((v) => ({
-                        variant_id: v.variationId,
-                        hgvs: v.title,
-                        classification: v.clinicalSignificance,
-                        condition: v.conditions[0] ?? "",
-                        review_status: v.reviewStatus,
-                    })), { reason: "ClinVar returned 0 variants for this gene" })
-                    : { coverage: "queried_no_data" },
-            },
-            resistance_evidence: (() => {
-                const rows = assembleResistanceEvidence(phase2);
-                return rows ? { coverage: "available", data: { rows } } : { coverage: "queried_no_data" };
-            })(),
-            combination_evidence: (() => {
-                const rows = assembleCombinationEvidence(phase2);
-                return rows ? { coverage: "available", data: { rows } } : { coverage: "queried_no_data" };
-            })(),
-            pathway_context: pathways.coverage === "available"
-                ? (() => {
-                    const presentDbs = new Set(pathways.data.pathways.map((p) => p.source));
-                    const ALL_DBS = ["reactome", "kegg", "wikipathways", "msigdb"];
-                    const QUERIED = ["reactome", "kegg"];
-                    return {
-                        coverage: "available",
-                        data: {
-                            rows: pathways.data.pathways.map((p) => ({
-                                pathway_id: p.id,
-                                pathway_name: p.name,
-                                database: p.source,
-                                evidence_score: 1,
-                            })),
-                            databases_queried: QUERIED.filter((db) => presentDbs.has(db)).length === QUERIED.length
-                                ? QUERIED
-                                : QUERIED.filter((db) => presentDbs.has(db)),
-                            databases_skipped: ALL_DBS.filter((db) => !presentDbs.has(db)),
-                        },
-                    };
-                })()
-                : { coverage: "queried_no_data" },
-            ppi_network: ppi.coverage === "available"
-                ? {
-                    coverage: "available",
-                    data: { partners: dedupePpiPartners(ppi.data.partners) },
-                }
-                : { coverage: "queried_no_data" },
-            normal_tissue_expression: expression.coverage === "available"
-                ? {
-                    coverage: "available",
-                    data: {
-                        source: expression.data.source,
-                        unit: expression.data.unit,
-                        normalization_notes: expression.data.normalization_notes,
-                        rows: expression.data.tissues
-                            .filter((t) => t.value != null)
-                            .sort((a, b) => (b.value ?? 0) - (a.value ?? 0))
-                            .map((t) => ({
-                            tissue: t.tissueLabel,
-                            value: t.value ?? 0,
-                        })),
-                    },
-                }
-                : { coverage: "queried_no_data" },
-            preclinical: {
-                ko_phenotype: impc.coverage === "available"
-                    ? {
-                        coverage: "available",
-                        data: {
-                            marker_symbol: impc.data.mouseMarkerSymbol,
-                            viability: impc.data.viability,
-                            sex_dimorphism: impc.data.sexDimorphic,
-                            organ_systems_with_phenotype: toOrganSystems(impc.data.organSystems),
-                            top_mp_terms: impc.data.mpTerms.slice(0, 25).map((m) => m.term),
-                            total_phenotype_count: impc.data.phenotypeCount,
-                            pre_weaning_lethal: impc.data.viability === "lethal_pre_weaning",
-                            supporting_literature: koSupportingLiterature,
-                        },
-                    }
-                    : { coverage: "queried_no_data" },
-                expression_heatmap: expressionMulti.coverage === "available"
-                    ? {
-                        coverage: "available",
-                        data: {
-                            cells: expressionMulti.data.bySpecies.flatMap((s) => {
-                                const speciesKey = s.species === "homo_sapiens"
-                                    ? "human"
-                                    : s.species === "mus_musculus"
-                                        ? "mouse"
-                                        : s.species === "rattus_norvegicus"
-                                            ? "rat"
-                                            : s.species === "macaca_mulatta"
-                                                ? "macaque"
-                                                : "dog";
-                                return s.tissues.map((t) => ({
-                                    tissue: t.tissue,
-                                    species: speciesKey,
-                                    rank: t.rank,
-                                }));
-                            }),
-                            per_species_coverage: Object.fromEntries(expressionMulti.data.bySpecies.map((s) => [
-                                s.species,
-                                s.tissues.length > 0 ? "available" : "queried_no_data",
-                            ])),
-                        },
-                    }
-                    : { coverage: "queried_no_data" },
-                translational_commentary: {
-                    coverage: "not_loaded",
-                    reason: NOT_LOADED_PHASE5,
-                },
-                preclinical_literature: preclinicalLitData
-                    ? { coverage: "available", data: preclinicalLitData }
-                    : { coverage: "queried_no_data", error: { message: "no species-tagged literature" } },
-                data_coverage: {
-                    ko: impc.coverage === "available" ? "available" : "queried_no_data",
-                    expression: expressionMulti.coverage,
-                    literature: pubmed.coverage,
-                    per_species: expressionMulti.coverage === "available"
-                        ? Object.fromEntries(expressionMulti.data.bySpecies.map((s) => [s.species, s.tissues.length > 0]))
-                        : {},
-                },
-            },
-            key_papers: keyPapersRows
-                ? { coverage: "available", data: { rows: keyPapersRows } }
-                : { coverage: "queried_no_data", error: { message: "no PubMed results" } },
-        },
-        analytics: {
-            evidence_conflicts: (() => {
-                const rows = assembleEvidenceConflicts(phase2);
-                return rows ? { coverage: "available", data: { rows } } : { coverage: "queried_no_data" };
-            })(),
-            evidence_timeline: evidenceTimeline
-                ? { coverage: "available", data: evidenceTimeline }
-                : pubmed.coverage === "not_loaded"
-                    ? { coverage: "not_loaded", reason: "PubMed not queried" }
-                    : { coverage: "queried_no_data", error: { message: "no PubMed years available" } },
-            translational_chain: translationalChain
-                ? { coverage: "available", data: translationalChain }
-                : pubmed.coverage === "not_loaded" && ctgov.coverage === "not_loaded" && impc.coverage === "not_loaded"
-                    ? { coverage: "not_loaded", reason: "no upstream tier evidence" }
-                    : { coverage: "queried_no_data", error: { message: "no tier evidence" } },
-            additional_evidence: (() => {
-                const data = assembleAdditionalEvidence(phase2);
-                return data ? { coverage: "available", data } : { coverage: "queried_no_data" };
-            })(),
-            discovery_trials: discoveryTrials,
-            // Stamped by Phase 5 after synthesis agents run; not available at assembly time.
-            synthesis_diagnostics: { coverage: "not_loaded", reason: "Phase-5 synthesis not yet run" },
-            quality_gates: { coverage: "not_loaded", reason: "Phase-5 synthesis not yet run" },
-        },
-        // Stamped by Phase 5 after the dossier-recommendation agent runs.
-        executive_recommendation: { coverage: "not_loaded", reason: "Phase-5 synthesis not yet run" },
-    };
-    return dossierBody;
-}
-/**
- * Reconciles target-level FAERS with class_precedent aggregates.
- *
- * When target-level FAERS has data (coverage: "available"), pass through
- * unchanged. When class_precedent has FAERS-derived reports in per_organ,
- * fold them into a class-scoped FAERS summary payload with
- * coverage: "available" and inference_path tagged — rather than leaving
- * the section as queried_no_data, which creates an evidence conflict
- * that the conflict-detector flags. When neither has data, return
- * faers unchanged.
- */
-export function reconcileFaersCoverage(input) {
-    const { faers, class_precedent: classPrecedent } = input;
-    // Target-level FAERS already has data — identity pass-through.
-    if (faers.coverage === "available")
-        return faers;
-    if (!classPrecedent || classPrecedent.coverage === "queried_no_data" || classPrecedent.coverage === "not_loaded")
-        return faers;
-    const perOrgan = classPrecedent.data?.per_organ ?? [];
-    const aes = perOrgan.flatMap((po) => (po.top_aes ?? []).map((ae) => ({
-        meddra_term: ae.term ?? "",
-        organ: po.organ,
-        report_count: ae.report_count ?? 0,
-    })));
-    // No class-level AEs either — leave the section as-is.
-    if (aes.length === 0)
-        return faers;
-    // Each AE-term count in top_aes counts reports that have that specific AE.
-    // A report with multiple AEs (e.g. NAUSEA + DIZZINESS) contributes to each
-    // term's count, so this sum is an upper bound on unique reports, not the
-    // true unique count. The class_precedent structure does not preserve unique
-    // report counts — this is the best approximation available.
-    const total = aes.reduce((s, ae) => s + ae.report_count, 0);
-    return {
-        coverage: "available",
-        data: {
-            total_reports: total,
-            seriousness: { coverage: "queried_no_data", total_reports: total },
-            top_signals: aes
-                .slice()
-                .sort((a, b) => b.report_count - a.report_count)
-                .slice(0, 20),
-            // class_precedent.per_organ aggregates across drugs without preserving
-            // per-drug report counts. Emitting per_modulator entries with
-            // report_count: 0 would look like real measured zeros to any consumer
-            // summing them to cross-check total_reports. Returning an empty array
-            // is the honest representation — the per-drug breakdown is not known
-            // at this layer.
-            per_modulator: [],
-        },
-        inference_path: "class_precedent.per_organ → FAERS",
-    };
-}