@inflexa-ai/harness 0.28.1 → 0.29.0

package/dist/contracts/profile-vocabulary.js

@@ -0,0 +1,449 @@
+/**
+ * The controlled vocabulary the profiler draws from, as data.
+ *
+ * One source, two readers: submit validation derives its enums from these lists and
+ * prompt assembly renders from the same entries, so the catalogue the agent is shown
+ * and the catalogue the validator enforces cannot diverge. Growth is an edit to these
+ * lists, measured by how often `other` is reached for.
+ *
+ * It lives in `contracts/` because the persisted profile carries these ids: a consumer
+ * rendering a stored group needs the category's meaning without importing the ledger,
+ * the schema module, or zod. Hosts do not extend it — the package owns the catalogue and
+ * versions it with the harness.
+ *
+ * Every category carries its nearest neighbour and the rule that separates them, because
+ * a category list without one is a list of labels an author picks by vibe. Every
+ * dimension category additionally carries a default treatment: whether a downstream step
+ * would TYPICALLY consume one value's files as a different substrate than another's.
+ * The agent follows the default and deviates only with a stated reason.
+ */
+// ── Group roles ────────────────────────────────────────────────────────────
+export const GROUP_ROLE_IDS = ["data", "metadata", "index", "documentation", "reference"];
+/**
+ * Companions (`.bai`/`.tbi`/per-file `.md5`) take no role — they attach to members. A
+ * checksum or inventory file spanning MANY members is not a companion but a member of
+ * its own, category `manifest`; `index` is for tables that exist to be joined against,
+ * never for binary sidecars.
+ */
+export const GROUP_ROLES = [
+    { id: "data", definition: "The measurements themselves — variant calls, a count matrix, images." },
+    { id: "metadata", definition: "Describes other members' meaning or provenance — a sample sheet, a pipeline manifest, config files." },
+    { id: "index", definition: "Standalone lookup or join tables — a variable-code dictionary, a file inventory table." },
+    { id: "documentation", definition: "Prose for humans — README, dataset card, methods paper, licences." },
+    {
+        id: "reference",
+        definition: "External or canonical inputs the analysis consumes but did not produce — genome build, annotation GTF, compound library, pathway sets.",
+    },
+];
+// ── Group categories ───────────────────────────────────────────────────────
+export const GROUP_CATEGORY_IDS = [
+    "sequencing-reads",
+    "alignment",
+    "variant-calls",
+    "copy-number",
+    "structural-variants",
+    "association-results",
+    "regions-signal",
+    "expression-matrix",
+    "single-cell-matrix",
+    "methylation",
+    "proteomics",
+    "metabolomics",
+    "cytometry",
+    "chemical-structures",
+    "imaging",
+    "clinical-table",
+    "sample-annotation",
+    "code",
+    "document",
+    "manifest",
+    "reference-data",
+    "other",
+];
+/**
+ * Coarse on purpose — a free `subtype` refines below the category. Content beats shape:
+ * a beta-value matrix is methylation whatever its layout, and a MAF is variant-calls
+ * whatever its extension. Grain and meaning decide.
+ */
+export const GROUP_CATEGORIES = [
+    {
+        id: "sequencing-reads",
+        definition: "Reads or raw instrument signal — FASTQ, unaligned BAM, FAST5/POD5.",
+        note: "vs alignment: mapped-ness decides — an unaligned BAM is reads.",
+    },
+    {
+        id: "alignment",
+        definition: "Reads mapped to a reference — BAM/CRAM carrying mapped records.",
+        note: "vs sequencing-reads: an unaligned BAM is reads, whatever the container.",
+    },
+    {
+        id: "variant-calls",
+        definition: "Per-sample small-variant calls (SNV/indel) — VCF with genotypes, MAF.",
+        note: "vs copy-number/structural-variants: those products get their own category whatever the container. vs association-results: per-sample genotypes rather than cohort statistics. vs reference-data: a sites-only population or annotation VCF is reference-data, role reference.",
+    },
+    {
+        id: "copy-number",
+        definition: "Copy-number products — segment tables, ratio tracks, CNV calls.",
+        note: "vs variant-calls: small variants stay variant-calls.",
+    },
+    {
+        id: "structural-variants",
+        definition: "Structural-variant products — breakpoints, SV calls.",
+        note: "vs variant-calls: an SV VCF is structural-variants.",
+    },
+    {
+        id: "association-results",
+        definition: "Per-variant or per-feature summary statistics without per-sample data — GWAS/QTL sumstats, PLINK output.",
+        note: "vs regions-signal: test statistics rather than positions or coverage.",
+    },
+    {
+        id: "regions-signal",
+        definition: "Genomic intervals and coverage tracks — BED/narrowPeak/broadPeak, bigWig/bedGraph.",
+        note: "vs association-results: positions and coverage rather than test statistics.",
+    },
+    {
+        id: "expression-matrix",
+        definition: "Bulk feature × sample abundance tables.",
+        note: "vs single-cell-matrix: barcoded per-cell products are single-cell. vs methylation: content beats shape — a beta-value matrix is methylation.",
+    },
+    {
+        id: "single-cell-matrix",
+        definition: "Barcoded per-cell products — MEX triplets, h5ad. Spatial omics is a subtype.",
+        note: "vs expression-matrix: bulk gene × sample tables are expression.",
+    },
+    {
+        id: "methylation",
+        definition: "Methylation measurements — beta/M-value matrices, per-CpG calls.",
+        note: "vs expression-matrix: the measured substrate decides, not the table's shape.",
+    },
+    {
+        id: "proteomics",
+        definition: "Protein-level measurements. MS rawness is a subtype.",
+        note: "vs imaging: an imaging mass-spec run is proteomics with subtype 'imaging MS' — assay substrate wins.",
+    },
+    {
+        id: "metabolomics",
+        definition: "Metabolite-level measurements. MS rawness is a subtype.",
+        note: "vs proteomics: by measured substrate.",
+    },
+    {
+        id: "cytometry",
+        definition: "Flow or mass cytometry — FCS and its derivatives.",
+        note: "vs imaging: per-event measurements rather than pixels; vs proteomics: neither.",
+    },
+    {
+        id: "chemical-structures",
+        definition: "Compound structures and libraries — SDF, SMILES, MOL.",
+        note: "vs reference-data: a screening library the analysis measures against is reference-data, role reference.",
+    },
+    {
+        id: "imaging",
+        definition: "Microscopy, radiology, and other pixel data — DICOM, TIFF stacks.",
+        note: "vs cytometry: pixels rather than per-event measurements.",
+    },
+    {
+        id: "clinical-table",
+        definition: "Facts about subjects over time — diagnoses, medications, outcomes, visits.",
+        note: "vs sample-annotation: subject grain rather than specimen grain. When one table does both, prefer sample-annotation only if its grain is the sample.",
+    },
+    {
+        id: "sample-annotation",
+        definition: "Specimen or measurement attribute maps — sample sheets, characteristic tables.",
+        note: "vs clinical-table: specimen attributes rather than subject history.",
+    },
+    {
+        id: "code",
+        definition: "Scripts, notebooks, and workflow definitions arriving as inputs.",
+        note: "vs document: executable rather than prose — a notebook is code even when narrated.",
+    },
+    {
+        id: "document",
+        definition: "Prose for humans — README, dataset card, methods paper, licence.",
+        note: "vs manifest: a dataset card in Markdown is a document; its JSON twin is a manifest.",
+    },
+    {
+        id: "manifest",
+        definition: "Machine-readable inventories of these files — checksum lists, file manifests.",
+        note: "vs index role: an inventory OF these files is a manifest; a table that exists to be joined against data is index-role.",
+    },
+    {
+        id: "reference-data",
+        definition: "Consumed-not-produced canonical inputs — genome builds, GTFs, pathway sets, spectral libraries, sites-only VCFs.",
+        note: "vs everything: the analysis consumes it and did not produce it.",
+    },
+    {
+        id: "other",
+        definition: "Outside the catalogue — carry a free label saying what it is.",
+        note: "A wrong specific category is worse than an honest `other` with a good label. Electrophysiology, phylogenetic trees, model weights, database dumps, and plate-reader tables belong here today.",
+    },
+];
+// ── Dimension categories ───────────────────────────────────────────────────
+export const DIMENSION_CATEGORY_IDS = [
+    "subject",
+    "sample",
+    "organism-species",
+    "model-system",
+    "cohort-arm",
+    "disease-state",
+    "treatment",
+    "timepoint",
+    "tissue-site",
+    "cell-type",
+    "genetic-perturbation",
+    "sex",
+    "age-group",
+    "developmental-stage",
+    "environment",
+    "assay-modality",
+    "assay-target",
+    "variant-origin",
+    "sample-pairing",
+    "processing-level",
+    "assay-version",
+    "batch",
+    "replicate",
+    "tool-variant",
+    "library-prep",
+    "data-partition",
+    "other",
+];
+export const DIMENSION_CATEGORIES = [
+    {
+        id: "subject",
+        scope: "biological",
+        definition: "The individual the data is about — patient, animal, donor. For a xenograft the subject is the donor; the host is genotype context.",
+        defaultTreatment: "dimension",
+        note: "Never split. Not the specimen — see sample.",
+    },
+    {
+        id: "sample",
+        scope: "biological",
+        definition: "Physical specimen. Nests under subject where one exists; community and environmental samples have none.",
+        defaultTreatment: "dimension",
+        note: "Never split. Not the library or aliquot, which is technical — see library-prep.",
+    },
+    {
+        id: "organism-species",
+        scope: "biological",
+        definition: "Species, when it varies across the dataset.",
+        defaultTreatment: "split",
+        note: "Species picks the reference genome. A constant organism is an identity field, not a dimension.",
+    },
+    {
+        id: "model-system",
+        scope: "biological",
+        definition: "Cell line, organoid, or model identity in subject-less panels.",
+        defaultTreatment: "dimension",
+        note: "A line used as the subject surrogate is still model-system, not subject.",
+    },
+    {
+        id: "cohort-arm",
+        scope: "biological",
+        definition: "Grouping assigned by design — randomisation, dose groups, discovery/validation cohorts. Assigned-by-design is the sole test.",
+        defaultTreatment: "dimension",
+        note: "Comparative models consume all arms together. Split only for independent cohorts, as a stated deviation. Observed conditions are disease-state.",
+    },
+    {
+        id: "disease-state",
+        scope: "biological",
+        definition: "Observed diagnosis or condition varying across subjects, including observational case/control.",
+        defaultTreatment: "dimension",
+        note: "Assigned-by-design grouping is cohort-arm.",
+    },
+    {
+        id: "treatment",
+        scope: "biological",
+        definition: "Compound, dose, or intervention as the varying quantity, including protocol-administered diet.",
+        defaultTreatment: "dimension",
+        note: "Dose-response curves are consumed together. Named designed groups are cohort-arm; ambient exposure is environment.",
+    },
+    {
+        id: "timepoint",
+        scope: "biological",
+        definition: "When the measurement was taken, in subject or visit time.",
+        defaultTreatment: "dimension",
+        note: "Known-ambiguous and documented as such. Acquisition, run, and processing dates are batch; duration of exposure is treatment.",
+    },
+    {
+        id: "tissue-site",
+        scope: "biological",
+        definition: "Organism part, sampling site, biofluid, or subcellular fraction, when it varies.",
+        defaultTreatment: "dimension",
+        note: "Ambiguous like timepoint. A constant tissue is an identity field.",
+    },
+    {
+        id: "cell-type",
+        scope: "biological",
+        definition: "Sorted population or annotated type, when it varies across files.",
+        defaultTreatment: "dimension",
+        note: "Per-cell annotations inside one matrix are file content, not a dimension.",
+    },
+    {
+        id: "genetic-perturbation",
+        scope: "biological",
+        definition: "Strain, genetic modification, KO/WT, CRISPR guide, shRNA, or ORF per member.",
+        defaultTreatment: "dimension",
+        note: "Perturbation as the study's designed comparison is cohort-arm.",
+    },
+    { id: "sex", scope: "biological", definition: "Sex as a varying attribute.", defaultTreatment: "dimension", note: "A constant is an identity field." },
+    {
+        id: "age-group",
+        scope: "biological",
+        definition: "Age or age band as a varying attribute.",
+        defaultTreatment: "dimension",
+        note: "Calendar age; biology is developmental-stage.",
+    },
+    {
+        id: "developmental-stage",
+        scope: "biological",
+        definition: "Developmental stage as a varying attribute.",
+        defaultTreatment: "dimension",
+        note: "Age-group is calendar; stage is biology.",
+    },
+    {
+        id: "environment",
+        scope: "biological",
+        definition: "Diet, stimulus, growth condition, field location.",
+        defaultTreatment: "dimension",
+        note: "Administered per protocol is treatment; plot or block structure is batch.",
+    },
+    {
+        id: "assay-modality",
+        scope: "technical",
+        definition: "Which measurement technology — RNA-seq vs proteomics vs imaging.",
+        defaultTreatment: "split",
+        note: "Different substrates. Different-marker panels are a different modality or target, not a version; a version of one assay is assay-version.",
+    },
+    {
+        id: "assay-target",
+        scope: "technical",
+        definition: "The molecular target an assay is directed at — a ChIP antibody target, a panel's target set.",
+        defaultTreatment: "split",
+        note: "Each target's output is a distinct substrate. Modality is the technology; target is what it points at.",
+    },
+    {
+        id: "variant-origin",
+        scope: "technical",
+        definition: "Somatic vs germline.",
+        defaultTreatment: "split",
+        note: "The canonical substrate case.",
+    },
+    {
+        id: "sample-pairing",
+        scope: "technical",
+        definition: "Tumour/normal or matched-pair membership.",
+        defaultTreatment: "split",
+        note: "Split with the pair id kept on both halves, because paired callers consume both together. Pairing is within-subject; cohort-arm is between-subjects.",
+    },
+    {
+        id: "processing-level",
+        scope: "technical",
+        definition: "Raw, filtered, and normalised tiers of the same product.",
+        defaultTreatment: "split",
+        note: "Steps consume one tier. Alternate tools at the same tier are tool-variant.",
+    },
+    {
+        id: "assay-version",
+        scope: "technical",
+        definition: "Panel, kit, or pipeline version over the same target set.",
+        defaultTreatment: "dimension",
+        note: "Batch-like. Different marker sets are assay-modality or assay-target.",
+    },
+    {
+        id: "batch",
+        scope: "technical",
+        definition: "Plate, lane, centre, run or acquisition date.",
+        defaultTreatment: "dimension",
+        note: "Never split. Subject-visit time is timepoint.",
+    },
+    {
+        id: "replicate",
+        scope: "technical",
+        definition: "Replicate index. The label must state whether the replication is biological or technical.",
+        defaultTreatment: "dimension",
+        note: "Never split.",
+    },
+    {
+        id: "tool-variant",
+        scope: "technical",
+        definition: "Alternate callers or aligners over the same input.",
+        defaultTreatment: "dimension",
+        note: "Consumed together for consensus. Pipeline versions over time are assay-version.",
+    },
+    {
+        id: "library-prep",
+        scope: "technical",
+        definition: "Library, label, capture, or aliquot.",
+        defaultTreatment: "dimension",
+        note: "A biofluid or subcellular fraction is biological — tissue-site.",
+    },
+    {
+        id: "data-partition",
+        scope: "technical",
+        definition: "Chromosome shard, file chunk, read-pair half, paged export.",
+        defaultTreatment: "dimension",
+        note: "Never split, rarely promoted — pure mechanics that stays on the set.",
+    },
+    {
+        id: "other",
+        scope: "technical",
+        definition: "Outside the catalogue — carry a free label saying what varies.",
+        defaultTreatment: "dimension",
+        note: "Monitored: a wrong specific category is worse than an honest `other`.",
+    },
+];
+// ── Probe list ─────────────────────────────────────────────────────────────
+export const DIMENSION_PROBE_IDS = ["subject", "sample", "cohort-arm", "timepoint", "batch"];
+/**
+ * A probe outcome, exactly one per probe.
+ *
+ * `not-found` is a correct and complete answer once the searched set is named — that is
+ * what removes the completeness pressure that otherwise invents dimensions. `attested`
+ * records prose-only evidence and can never justify a split.
+ */
+export const PROBE_OUTCOME_IDS = ["found", "not-found", "found-but-constant", "attested"];
+/**
+ * The only dimensions the agent actively looks for. The searchable set is bounded and
+ * named: files in metadata- or documentation-role groups plus clinical-table and
+ * sample-annotation members, up to about ten files. Everything off this list is recorded
+ * only when encountered during ordinary orientation reading.
+ *
+ * Assay-modality and replicate are deliberately absent: varying modality is already
+ * surfaced by group categories, and probing for replicates invites inferring replicate
+ * structure from sample counts, which is fabrication by another name.
+ */
+export const DIMENSION_PROBES = [
+    {
+        id: "subject",
+        guidance: "Identifier slots in per-subject trees; subject-ID columns in clinical or sample tables. A reference-only tree is a legitimate not-found — a genome build's species is not a subject.",
+    },
+    { id: "sample", guidance: "Sample-ID slots or columns; sample sheets. Nests under subject when both exist." },
+    {
+        id: "cohort-arm",
+        guidance: "Arm or group columns whose values are assigned by design. An observed condition or diagnosis column is disease-state — answer not-found and point to where it landed. Observational data typically has no arms.",
+    },
+    { id: "timepoint", guidance: "Visit, day, or cycle columns; timepoint tokens in names. Acquisition and run dates are batch." },
+    { id: "batch", guidance: "Plate, lane, or run columns; run-date tokens; centre identifiers. Subject-visit time is timepoint." },
+];
+// ── Lookups ────────────────────────────────────────────────────────────────
+const GROUP_CATEGORY_BY_ID = new Map(GROUP_CATEGORIES.map((entry) => [entry.id, entry]));
+const DIMENSION_CATEGORY_BY_ID = new Map(DIMENSION_CATEGORIES.map((entry) => [entry.id, entry]));
+export function groupCategoryEntry(id) {
+    return GROUP_CATEGORY_BY_ID.get(id);
+}
+export function dimensionCategoryEntry(id) {
+    return DIMENSION_CATEGORY_BY_ID.get(id);
+}
+/**
+ * The scope a dimension takes.
+ *
+ * Derived from the category for everything in the catalogue. `other` is by definition
+ * outside it, so there is nothing to derive from and the scope is the agent's to declare:
+ * a biological thing that varies and has no category is still biological, and pinning
+ * every `other` to technical would misfile it. Undeclared, it defaults to technical.
+ */
+export function dimensionScope(id, declared) {
+    if (id === "other")
+        return declared ?? "technical";
+    return DIMENSION_CATEGORY_BY_ID.get(id)?.scope ?? "technical";
+}

package/dist/contracts/target-dossier.d.ts

@@ -9342,6 +9342,27 @@ export declare const DossierBodySchema: z.ZodObject<{
         coverage: z.ZodLiteral<"not_loaded">;
         reason: z.ZodOptional<z.ZodString>;
     }, z.core.$strip>], "coverage">;
+    entity: z.ZodObject<{
+        canonicalId: z.ZodString;
+        symbol: z.ZodString;
+        displayName: z.ZodString;
+        entityType: z.ZodEnum<{
+            gene: "gene";
+            rna: "rna";
+            protein: "protein";
+            complex: "complex";
+        }>;
+        ontology: z.ZodString;
+        identifiers: z.ZodObject<{
+            hgnc: z.ZodOptional<z.ZodString>;
+            ensembl: z.ZodOptional<z.ZodString>;
+            uniprot: z.ZodOptional<z.ZodString>;
+            chembl: z.ZodOptional<z.ZodString>;
+            ncbiGene: z.ZodOptional<z.ZodString>;
+        }, z.core.$strip>;
+        synonyms: z.ZodDefault<z.ZodArray<z.ZodString>>;
+        proteinFamily: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>;
     off_target_panel: z.ZodDiscriminatedUnion<[z.ZodObject<{
         coverage: z.ZodLiteral<"available">;
         data: z.ZodObject<{
@@ -9565,27 +9586,6 @@ export declare const DossierBodySchema: z.ZodObject<{
         inferred_therapeutic_area: z.ZodNullable<z.ZodString>;
         no_liabilities_disclosure: z.ZodOptional<z.ZodString>;
     }, z.core.$strict>;
-    entity: z.ZodObject<{
-        canonicalId: z.ZodString;
-        symbol: z.ZodString;
-        displayName: z.ZodString;
-        entityType: z.ZodEnum<{
-            gene: "gene";
-            rna: "rna";
-            protein: "protein";
-            complex: "complex";
-        }>;
-        ontology: z.ZodString;
-        identifiers: z.ZodObject<{
-            hgnc: z.ZodOptional<z.ZodString>;
-            ensembl: z.ZodOptional<z.ZodString>;
-            uniprot: z.ZodOptional<z.ZodString>;
-            chembl: z.ZodOptional<z.ZodString>;
-            ncbiGene: z.ZodOptional<z.ZodString>;
-        }, z.core.$strip>;
-        synonyms: z.ZodDefault<z.ZodArray<z.ZodString>>;
-        proteinFamily: z.ZodOptional<z.ZodString>;
-    }, z.core.$strip>;
     indications: z.ZodDiscriminatedUnion<[z.ZodObject<{
         coverage: z.ZodLiteral<"available">;
         data: z.ZodObject<{

package/dist/execution/input-signature.d.ts

@@ -2,14 +2,21 @@
  * The input signature — the fixed-width comparand a completed profile persists in
  * place of a per-file identity list.
  *
- * `inputFileIds` had exactly one consumer, the staleness comparison, so a set of
- * several thousand identifiers was persisted, detoasted on every read of the row, and
+ * A per-file identity list would have exactly one consumer, the staleness comparison —
+ * several thousand identifiers persisted, detoasted on every read of the row, and
  * compared element-wise to answer a question 40 bytes answer. The signature covers
  * `size` and `mtimeMs` as well, so it also notices that the bytes behind a path
  * changed — a `fileId` is a path identity and survives an in-place edit unchanged.
  *
  * Order independence is a property of the digest, not of the caller: entries are
  * sorted canonically before hashing, so two enumerations of the same set agree.
+ *
+ * It digests KEPT files only — quarantine runs before the digest, on both sides of the
+ * comparison. A partial-download twin or an OS droppings file was never part of what was
+ * profiled, so its appearance, edit, or removal is not drift, and a profile that
+ * re-triggered on it would re-run on churn nobody caused. A change to the quarantine rules
+ * themselves recomputes every signature once on the next scan, which is the correct
+ * consequence: the definition of "kept" changed.
  */
 import type { DataProfileInputSignature } from "../state/data-profile.js";
 import type { StagedInput } from "./staged-input.js";
@@ -18,9 +25,15 @@ export interface SignableInput {
     readonly fileId: string;
     readonly size: number;
     readonly mtimeMs?: number;
+    /**
+     * Path relative to the data dir — what quarantine reads. An entry carrying none is
+     * always kept: a caller that cannot name a path cannot be asked to prove its file is
+     * not junk, and dropping it would be the one outcome worse than including it.
+     */
+    readonly relativePath?: string;
 }
 /**
- * Digest a staged input set.
+ * Digest a staged input set, quarantine first.
  *
  * A workflow input persisted before `StagedInput.mtimeMs` existed deserializes without
  * it despite the compile-time type; such an entry contributes `-` in the mtime field

package/dist/execution/post-step-pipeline.js

@@ -212,7 +212,7 @@ export async function vectorIndexStepOutputs(deps, postCtx, artifacts) {
     let indexed = 0;
     let failed = 0;
     for (const entry of liveMetadata) {
-        const ok = await indexOne(`/${input.analysisId}/${entry.dbPath}`, entry.description, { text: entry.description, type: "output", ...entry.metadata });
+        const ok = await indexOne(`/${input.analysisId}/${entry.dbPath}`, entry.description, { text: entry.description, type: "output", ...entry.metadata, path: entry.dbPath });
         if (ok)
             indexed++;
         else

package/dist/execution/staged-input.d.ts

@@ -26,10 +26,10 @@ export interface StagedInput {
     readonly hash: string;
     readonly size: number;
     /**
-     * Last-modification time in epoch milliseconds. With `size` and `fileId` it forms
-     * the file's drift signature — the value a consumer compares against a completed
-     * profile's `inputFiles` to decide whether the same bytes were profiled, since a
-     * `fileId` is a path identity and survives an in-place content edit unchanged.
+     * Last-modification time in epoch milliseconds. With `size` and `fileId` it feeds
+     * the kept-files `inputSignature` a completed profile persists — what lets a
+     * consumer notice the bytes behind a path changed, since a `fileId` is a path
+     * identity and survives an in-place content edit unchanged.
      *
      * Embedder-supplied and opaque: the CLI reads it from the same `stat` that yields
      * `size`; a managed service supplies its object store's last-modified epoch. The

package/dist/index.d.ts

@@ -115,8 +115,12 @@ export { DATA_PROFILE_RUN_LITERAL } from "./contracts/data-profile.js";
 export { initCortexState } from "./state/init.js";
 export { upsertAnalysis } from "./state/analyses.js";
 export { clearDataProfile, loadDataProfileStatus, tryRetryDataProfile, reconcileOrphanedDataProfile } from "./state/data-profile.js";
-export type { DataProfileAxis, DataProfileCoverage, DataProfileFile, DataProfileInputFile, DataProfileInputSignature, DataProfileKind, DataProfileLifecycleStatus, DataProfileOrganism, DataProfileQualityAssessment, DataProfileResult, DataProfileSubjectSource, } from "./contracts/data-profile.js";
+export type { DataProfileAxis, DataProfileChecked, DataProfileCompanionCompleteness, DataProfileCoverage, DataProfileDimension, DataProfileFile, DataProfileGroup, DataProfileGroupSlot, DataProfileInputSignature, DataProfileKind, DataProfileLifecycleStatus, DataProfileMemberAnnotation, DataProfileObservation, DataProfileOrganism, DataProfilePartition, DataProfileProbeReport, DataProfileQualityAssessment, DataProfileQuarantine, DataProfileRecipeStep, DataProfileResult, DataProfileSubjectSource, } from "./contracts/data-profile.js";
 export type { DataProfileStatus } from "./state/data-profile.js";
+export { DIMENSION_CATEGORIES, DIMENSION_CATEGORY_IDS, DIMENSION_PROBES, DIMENSION_PROBE_IDS, GROUP_CATEGORIES, GROUP_CATEGORY_IDS, GROUP_ROLES, GROUP_ROLE_IDS, PROBE_OUTCOME_IDS, dimensionCategoryEntry, dimensionScope, groupCategoryEntry, } from "./contracts/profile-vocabulary.js";
+export type { DimensionCategory, DimensionCategoryEntry, DimensionProbe, DimensionProbeEntry, DimensionScope, DimensionTreatment, GroupCategory, GroupCategoryEntry, GroupRole, GroupRoleEntry, ProbeOutcome, } from "./contracts/profile-vocabulary.js";
+export { dimensionCardinalities, profileCaveats, profileDatasetFileCount, profileDimensions, profileFileRecords, profileGroups, } from "./app/data-profile-view.js";
+export type { ProfileDimensionView, ProfileGroupView } from "./app/data-profile-view.js";
 export { createAnalysisPurge } from "./state/purge-analysis.js";
 export type { AnalysisPurge, AnalysisPurgeDeps, AnalysisPurgeOutcome } from "./state/purge-analysis.js";
 export type { StagedInput } from "./execution/staged-input.js";

package/dist/index.js

@@ -204,6 +204,13 @@ export { upsertAnalysis } from "./state/analyses.js";
 // analysis's input set empties (deferring on a running workflow), so the UI
 // stops advertising a profile that describes files the analysis no longer has.
 export { clearDataProfile, loadDataProfileStatus, tryRetryDataProfile, reconcileOrphanedDataProfile } from "./state/data-profile.js";
+// The controlled vocabulary the profiler draws from, as data. A consumer rendering a
+// stored group needs the category's meaning, and the package owns the catalogue — a host
+// does not extend it.
+export { DIMENSION_CATEGORIES, DIMENSION_CATEGORY_IDS, DIMENSION_PROBES, DIMENSION_PROBE_IDS, GROUP_CATEGORIES, GROUP_CATEGORY_IDS, GROUP_ROLES, GROUP_ROLE_IDS, PROBE_OUTCOME_IDS, dimensionCategoryEntry, dimensionScope, groupCategoryEntry, } from "./contracts/profile-vocabulary.js";
+// Reading a persisted profile of either era through one shape — the projection every
+// consumer uses so that a row written before groups existed still renders.
+export { dimensionCardinalities, profileCaveats, profileDatasetFileCount, profileDimensions, profileFileRecords, profileGroups, } from "./app/data-profile-view.js";
 // Analysis reclamation — `purgeAnalysis` reclaims an analysis's whole Postgres
 // footprint from its id alone; workspace files stay the embedder's to dispose of.
 export { createAnalysisPurge } from "./state/purge-analysis.js";