@inflexa-ai/harness 0.15.0 → 0.16.0

package/dist/reference-data/catalog.js

@@ -562,6 +562,118 @@ export const REFERENCE_DATA_CATALOG = deepFreeze(ReferenceDataCatalogSchema.pars
                 },
             ],
         },
+        {
+            // The three canonical-pathway subcollections MSigDB still redistributes
+            // that nothing else in this catalog covers. Reactome and WikiPathways are
+            // deliberately NOT here — they have their own datasets, from upstream
+            // directly and on upstream's release cadence rather than MSigDB's snapshot.
+            id: "msigdb-canonical-pathways-human",
+            version: "2026.1",
+            title: "MSigDB human canonical pathways (KEGG MEDICUS, BioCarta, PID)",
+            description: "Three curated pathway subcollections from MSigDB C2:CP for Homo sapiens (symbols GMT) — 1,146 sets over roughly 300 KB. The pathway resources that Reactome and WikiPathways do not duplicate: KEGG's redistributable MEDICUS subset, plus the BioCarta and NCI-PID snapshots. Reach for these when a question names a pathway that the other two resources do not carry.",
+            organism: "human",
+            sourceUrl: "https://data.broadinstitute.org/gsea-msigdb/msigdb/",
+            license: { identifier: "MSigDB-License", url: "https://www.gsea-msigdb.org/gsea/msigdb/license.jsp" },
+            recommendation: { group: "pathways", recommended: true },
+            artifacts: [
+                {
+                    path: "c2.cp.kegg_medicus.v2026.1.Hs.symbols.gmt",
+                    url: "https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2026.1.Hs/c2.cp.kegg_medicus.v2026.1.Hs.symbols.gmt",
+                    format: "gmt",
+                    contents: "658 sets, one per line, tab-separated: set name (KEGG_MEDICUS_*), source URL, then member gene symbols (HGNC). This is the MEDICUS subset, the only KEGG content MSigDB may redistribute — it is NOT the legacy KEGG_* collection, so set names and membership do not match older KEGG results and the two are not comparable across releases.",
+                },
+                {
+                    path: "c2.cp.biocarta.v2026.1.Hs.symbols.gmt",
+                    url: "https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2026.1.Hs/c2.cp.biocarta.v2026.1.Hs.symbols.gmt",
+                    format: "gmt",
+                    contents: "292 sets, one per line, tab-separated: set name (BIOCARTA_*), source URL, then member gene symbols (HGNC). Small, signalling-focused sets. BioCarta is no longer maintained upstream, so this is a historical snapshot — good for classical signalling cascades, unreliable for anything characterised in the last decade.",
+                },
+                {
+                    path: "c2.cp.pid.v2026.1.Hs.symbols.gmt",
+                    url: "https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2026.1.Hs/c2.cp.pid.v2026.1.Hs.symbols.gmt",
+                    format: "gmt",
+                    contents: "196 sets, one per line, tab-separated: set name (PID_*), source URL, then member gene symbols (HGNC). The NCI Pathway Interaction Database, retired upstream and preserved here. Curated signalling pathways with tight membership — the sets are small, so enrichment on them is sensitive to a handful of genes.",
+                },
+            ],
+        },
+        {
+            id: "msigdb-perturbations-human",
+            version: "2026.1",
+            title: "MSigDB human chemical and genetic perturbation gene sets",
+            description: "The MSigDB C2:CGP collection for Homo sapiens (symbols GMT): 3,555 expression signatures from published chemical and genetic perturbation experiments, roughly 2.9 MB. The collection to test a signature against 'what happens when this drug or gene is perturbed' — the natural gene-set complement to connectivity-style drug repurposing.",
+            organism: "human",
+            sourceUrl: "https://data.broadinstitute.org/gsea-msigdb/msigdb/",
+            license: { identifier: "MSigDB-License", url: "https://www.gsea-msigdb.org/gsea/msigdb/license.jsp" },
+            recommendation: { group: "pathways", recommended: false },
+            artifacts: [
+                {
+                    path: "c2.cgp.v2026.1.Hs.symbols.gmt",
+                    url: "https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2026.1.Hs/c2.cgp.v2026.1.Hs.symbols.gmt",
+                    format: "gmt",
+                    contents: "3,555 sets, one per line, tab-separated: set name, source URL, then member gene symbols (HGNC). Names are keyed to the publication's first author (WANG_*, LEE_*, MIKKELSEN_*) and carry a _UP/_DN direction — the two halves of a signature are a pair and must never be merged. These are observed responses in one study's model system, so a hit means the profile resembles that experiment, not that the same mechanism is engaged; report the perturbation and the system it was measured in.",
+                },
+            ],
+        },
+        {
+            id: "msigdb-regulatory-targets-human",
+            version: "2026.1",
+            title: "MSigDB human regulatory target gene sets (TF and miRNA)",
+            description: "MSigDB C3 regulatory target sets for Homo sapiens (symbols GMT): 506 transcription-factor target sets from GTRD and 2,377 miRNA target sets from miRDB, roughly 4.1 MB combined. Both are computationally predicted, which is what distinguishes them from the curated signed regulons catalogued separately — reach for these for breadth of coverage, not for confident single-edge claims.",
+            organism: "human",
+            sourceUrl: "https://data.broadinstitute.org/gsea-msigdb/msigdb/",
+            license: { identifier: "MSigDB-License", url: "https://www.gsea-msigdb.org/gsea/msigdb/license.jsp" },
+            recommendation: { group: "pathways", recommended: false },
+            artifacts: [
+                {
+                    path: "c3.tft.gtrd.v2026.1.Hs.symbols.gmt",
+                    url: "https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2026.1.Hs/c3.tft.gtrd.v2026.1.Hs.symbols.gmt",
+                    format: "gmt",
+                    contents: "506 sets, one per line, tab-separated: the set name is the regulating TF's own gene symbol, then the source URL, then its predicted target symbols (HGNC). Targets are inferred from GTRD ChIP-seq binding near the TSS, so membership means 'bound', not 'regulated', and the sets carry NO direction — a set cannot say whether the TF activates or represses. Use a signed regulon resource when direction matters.",
+                },
+                {
+                    path: "c3.mir.mirdb.v2026.1.Hs.symbols.gmt",
+                    url: "https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2026.1.Hs/c3.mir.mirdb.v2026.1.Hs.symbols.gmt",
+                    format: "gmt",
+                    contents: "2,377 sets, one per line, tab-separated: set name is the miRNA identifier, then the source URL, then predicted target gene symbols (HGNC). Sequence-based predictions from miRDB, unvalidated — false-positive rates are high per edge. Because miRNAs repress their targets, enrichment of a miRNA's target set among UP-regulated genes is evidence that the miRNA is LESS active, which is the opposite of the naive reading.",
+                },
+            ],
+        },
+        {
+            id: "msigdb-cancer-modules-human",
+            version: "2026.1",
+            title: "MSigDB human computational cancer modules",
+            description: "The MSigDB C4 collection for Homo sapiens (symbols GMT): 1,006 gene sets derived computationally from cancer expression compendia rather than curated, roughly 670 KB. Mixed provenance — co-expression modules, cancer gene neighbourhoods, and single-cell meta-programs — useful for recognising a recurrent tumour expression programme that no curated pathway names.",
+            organism: "human",
+            sourceUrl: "https://data.broadinstitute.org/gsea-msigdb/msigdb/",
+            license: { identifier: "MSigDB-License", url: "https://www.gsea-msigdb.org/gsea/msigdb/license.jsp" },
+            recommendation: { group: "pathways", recommended: false },
+            artifacts: [
+                {
+                    path: "c4.all.v2026.1.Hs.symbols.gmt",
+                    url: "https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2026.1.Hs/c4.all.v2026.1.Hs.symbols.gmt",
+                    format: "gmt",
+                    contents: "1,006 sets, one per line, tab-separated: set name, source URL, then member gene symbols (HGNC). The name prefix identifies the derivation and they are not interchangeable: MODULE_* are cancer modules, GNF2_/MORF_/GCM_* are co-expression neighbourhoods of a named seed gene, GAVISH_* are single-cell meta-programs. All are data-derived, so a set carries no mechanism and many are unnamed by biology — report the module identifier and its derivation, never a mechanistic gloss.",
+                },
+            ],
+        },
+        {
+            id: "msigdb-celltype-human",
+            version: "2026.1",
+            title: "MSigDB human cell type signature gene sets",
+            description: "The MSigDB C8 collection for Homo sapiens (symbols GMT): 866 cell-identity marker sets curated from single-cell studies of human tissue, roughly 1.1 MB. The gene-set route to cell-type identity — for scoring clusters or deconvolved bulk profiles against published cell populations where no matched annotated reference is available.",
+            organism: "human",
+            sourceUrl: "https://data.broadinstitute.org/gsea-msigdb/msigdb/",
+            license: { identifier: "MSigDB-License", url: "https://www.gsea-msigdb.org/gsea/msigdb/license.jsp" },
+            recommendation: { group: "pathways", recommended: false },
+            artifacts: [
+                {
+                    path: "c8.all.v2026.1.Hs.symbols.gmt",
+                    url: "https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2026.1.Hs/c8.all.v2026.1.Hs.symbols.gmt",
+                    format: "gmt",
+                    contents: "866 sets, one per line, tab-separated: set name, source URL, then member gene symbols (HGNC). Names key to the source study and its tissue (DESCARTES_*, TRAVAGLINI_*, HE_*), so the same cell type appears several times with different membership and the studies do not agree — treat agreement across studies as the signal, and always report which study a call came from. Signatures are tissue-context specific: a marker set from fetal tissue or a different organ can score high for the wrong reason.",
+                },
+            ],
+        },
         {
             id: "collectri-human",
             version: "2.0",
@@ -829,6 +941,430 @@ export const REFERENCE_DATA_CATALOG = deepFreeze(ReferenceDataCatalogSchema.pars
                 },
             ],
         },
+        {
+            id: "scgate-human-generic",
+            version: "v0.17",
+            title: "scGate human generic cell-type gating models",
+            description: "Broad human cell identities — immune lineages, T/B/NK, myeloid, epithelial, stromal, endothelial and their common subsets. 29 gating models plus the signature table they all share — a few KB in total. scGate assigns cells by walking a marker hierarchy rather than by clustering, labelling one population and marking everything else impure, which makes it the route when no annotated reference exists to transfer from. Its own model downloader cannot run without egress, so these files ARE the model database. From the immutable v0.17 release.",
+            organism: "human",
+            sourceUrl: "https://github.com/carmonalab/scGate_models",
+            license: { identifier: "GPL-3.0", url: "https://github.com/carmonalab/scGate_models" },
+            recommendation: { group: "cell-typing", recommended: true },
+            artifacts: [
+                {
+                    path: "Bcell.GerminalCenter_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Bcell.GerminalCenter_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Bcell / GerminalCenter. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Bcell.NonGerminalCenter_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Bcell.NonGerminalCenter_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Bcell / NonGerminalCenter. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Bcell_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Bcell_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Bcell. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CAF.SMC.pericyte_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/CAF.SMC.pericyte_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CAF / SMC / pericyte. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD4T_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/CD4T_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD4T. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8TIL_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/CD8TIL_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8TIL. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8T_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/CD8T_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8T. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Endothelial_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Endothelial_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Endothelial. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Epithelial_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Epithelial_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Epithelial. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Erythrocyte_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Erythrocyte_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Erythrocyte. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Female_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Female_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Female. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Immune_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Immune_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Immune. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Macrophage_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Macrophage_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Macrophage. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Male_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Male_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Male. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Mast_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Mast_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Mast. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Megakaryocyte_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Megakaryocyte_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Megakaryocyte. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "MoMacDCMast_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/MoMacDCMast_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for MoMacDCMast. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "MoMacDC_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/MoMacDC_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for MoMacDC. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "MoMac_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/MoMac_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for MoMac. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Monocyte_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Monocyte_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Monocyte. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Myeloid_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Myeloid_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Myeloid. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "NK_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/NK_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for NK. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Neutrophils_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Neutrophils_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Neutrophils. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "PanBcell_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/PanBcell_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for PanBcell. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "PlasmaCell_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/PlasmaCell_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for PlasmaCell. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Stromal_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Stromal_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Stromal. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Tcell.alphabeta_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Tcell.alphabeta_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Tcell / alphabeta. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Tcell_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/Tcell_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Tcell. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "master_table.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/master_table.tsv",
+                    format: "tsv",
+                    contents: "Two columns, tab-separated: `name`, `signature`. Signatures are semicolon-separated gene symbols with a trailing `-` marking a negative marker (`KRT18-`). Every model in this dataset carries an EMPTY signature column and resolves against this file by name, so it must be staged with them and its path passed explicitly when loading — the loader resolves the default bare filename against the working directory, not against the model's folder.",
+                },
+                {
+                    path: "panDC_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/generic/panDC_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for panDC. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+            ],
+        },
+        {
+            id: "scgate-human-tme",
+            version: "v0.17",
+            title: "scGate human tumour microenvironment gating models",
+            description: "The HiTME panel: the cell types making up a human tumour microenvironment, gated as one consistent hierarchy. 15 gating models plus the signature table they all share — a few KB in total. scGate assigns cells by walking a marker hierarchy rather than by clustering, labelling one population and marking everything else impure, which makes it the route when no annotated reference exists to transfer from. Its own model downloader cannot run without egress, so these files ARE the model database. From the immutable v0.17 release.",
+            organism: "human",
+            sourceUrl: "https://github.com/carmonalab/scGate_models",
+            license: { identifier: "GPL-3.0", url: "https://github.com/carmonalab/scGate_models" },
+            recommendation: { group: "cell-typing", recommended: true },
+            artifacts: [
+                {
+                    path: "Bcell_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/Bcell_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Bcell. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD4T_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/CD4T_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD4T. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8T_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/CD8T_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8T. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Endothelial_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/Endothelial_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Endothelial. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Epithelial_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/Epithelial_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Epithelial. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Erythrocyte_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/Erythrocyte_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Erythrocyte. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Fibroblast_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/Fibroblast_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Fibroblast. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Mast_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/Mast_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Mast. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "MoMac_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/MoMac_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for MoMac. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "NK_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/NK_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for NK. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Neutrophils_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/Neutrophils_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Neutrophils. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "PlasmaCell_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/PlasmaCell_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for PlasmaCell. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "Platelet_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/Platelet_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for Platelet. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "gdT_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/gdT_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for gdT. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "master_table.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/master_table.tsv",
+                    format: "tsv",
+                    contents: "Two columns, tab-separated: `name`, `signature`. Signatures are semicolon-separated gene symbols with a trailing `-` marking a negative marker (`KRT18-`). Every model in this dataset carries an EMPTY signature column and resolves against this file by name, so it must be staged with them and its path passed explicitly when loading — the loader resolves the default bare filename against the working directory, not against the model's folder.",
+                },
+                {
+                    path: "panDC_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/HiTME/panDC_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for panDC. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+            ],
+        },
+        {
+            id: "scgate-human-cd8-til",
+            version: "v0.17",
+            title: "scGate human CD8 TIL state gating models",
+            description: "CD8 tumour-infiltrating T-cell states — naive, effector-memory, tissue-resident, exhausted, precursor-exhausted, TEMRA, MAIT and innate-like. 8 gating models plus the signature table they all share — a few KB in total. scGate assigns cells by walking a marker hierarchy rather than by clustering, labelling one population and marking everything else impure, which makes it the route when no annotated reference exists to transfer from. Its own model downloader cannot run without egress, so these files ARE the model database. From the immutable v0.17 release.",
+            organism: "human",
+            sourceUrl: "https://github.com/carmonalab/scGate_models",
+            license: { identifier: "GPL-3.0", url: "https://github.com/carmonalab/scGate_models" },
+            recommendation: { group: "cell-typing", recommended: true },
+            artifacts: [
+                {
+                    path: "CD8_EM_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/CD8_EM_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8 EM. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8_MAIT_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/CD8_MAIT_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8 MAIT. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8_N_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/CD8_N_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8 N. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8_TEMRA_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/CD8_TEMRA_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8 TEMRA. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8_TEX_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/CD8_TEX_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8 TEX. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8_TPEX_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/CD8_TPEX_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8 TPEX. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8_TRM_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/CD8_TRM_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8 TRM. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "CD8_Tinn_scGate_Model.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/CD8_Tinn_scGate_Model.tsv",
+                    format: "tsv",
+                    contents: "Hierarchical gating model for CD8 Tinn. Four columns, tab-separated: `levels`, `use_as`, `name`, `signature`. Rows are gating steps applied in level order, each marking a population positive (retain) or negative (exclude). The `signature` column is empty by design and is filled from master_table.tsv.",
+                },
+                {
+                    path: "master_table.tsv",
+                    url: "https://raw.githubusercontent.com/carmonalab/scGate_models/v0.17/human/CD8_TIL/master_table.tsv",
+                    format: "tsv",
+                    contents: "Two columns, tab-separated: `name`, `signature`. Signatures are semicolon-separated gene symbols with a trailing `-` marking a negative marker (`KRT18-`). Every model in this dataset carries an EMPTY signature column and resolves against this file by name, so it must be staged with them and its path passed explicitly when loading — the loader resolves the default bare filename against the working directory, not against the model's folder.",
+                },
+            ],
+        },
+        {
+            // ProjecTILs projects a query onto a fixed reference embedding rather than
+            // re-clustering it, so the atlas defines the state vocabulary the results are
+            // reported in. Its own loader downloads from figshare and cannot run here;
+            // these ARE the reference maps, passed to `load.reference.map()` as a path.
+            id: "projectils-cd8-til-human",
+            version: "1.0.0",
+            title: "ProjecTILs human CD8 TIL reference atlas",
+            description: "Reference map of human CD8+ tumour-infiltrating T cells for ProjecTILs: 11,021 single-cell transcriptomes from 20 samples across 7 tumour types. Projecting onto it places query cells in a fixed, curated state space — naive through effector-memory, tissue-resident and exhausted — rather than in de novo clusters, so results are comparable across datasets. The route to exhaustion and precursor-exhausted phenotyping, which unsupervised clustering names inconsistently. ~258 MB, opt-in.",
+            organism: "human",
+            sourceUrl: "https://figshare.com/articles/dataset/ProjecTILs_human_reference_atlas_of_CD8_tumor-infiltrating_T_cells_CD8_TIL_version_1/23608308",
+            license: { identifier: "CC-BY-4.0", url: "https://creativecommons.org/licenses/by/4.0/" },
+            recommendation: { group: "cell-typing", recommended: false },
+            artifacts: [
+                {
+                    path: "CD8T_human_ref_v1.rds",
+                    url: "https://ndownloader.figshare.com/files/41414556",
+                    format: "rds",
+                    contents: "Seurat reference object carrying the atlas embedding and its cell-state labels. Pass the file path to ProjecTILs' reference loader — it takes a local path directly and does not download. Human gene symbols, so mouse queries need ortholog conversion first. The projection reports every query cell in this atlas's state vocabulary, so states absent from it cannot be discovered by projection alone.",
+                },
+            ],
+        },
+        {
+            id: "projectils-cd4-til-human",
+            version: "1.0.0",
+            title: "ProjecTILs human CD4 TIL reference atlas",
+            description: "Reference map of human CD4+ tumour-infiltrating T cells for ProjecTILs: 12,631 single-cell transcriptomes from 20 samples across 9 tumour types. The CD4 counterpart to the CD8 atlas — regulatory, helper and follicular-helper states in a fixed coordinate system. ~262 MB, opt-in.",
+            organism: "human",
+            sourceUrl: "https://figshare.com/articles/dataset/ProjecTILs_human_reference_atlas_of_CD4_tumor-infiltrating_T_cells_version_1/21981536",
+            license: { identifier: "CC-BY-4.0", url: "https://creativecommons.org/licenses/by/4.0/" },
+            recommendation: { group: "cell-typing", recommended: false },
+            artifacts: [
+                {
+                    path: "CD4T_human_ref_v1.rds",
+                    url: "https://ndownloader.figshare.com/files/39012395",
+                    format: "rds",
+                    contents: "Seurat reference object carrying the atlas embedding and its cell-state labels. Pass the file path to ProjecTILs' reference loader. Human gene symbols. CD4 and CD8 atlases are separate maps — subset the query to the right lineage before projecting, or the projection forces cells into a state space that does not describe them.",
+                },
+            ],
+        },
+        {
+            id: "projectils-til-mouse",
+            version: "3.0.0",
+            title: "ProjecTILs mouse TIL reference atlas",
+            description: "Reference map of murine tumour-infiltrating T cells for ProjecTILs: 16,803 single-cell transcriptomes from 25 samples (B16 melanoma, MC38 colon adenocarcinoma) across six studies. This is the atlas ProjecTILs downloads by default, so it is the one every published example assumes — but it is MOUSE, and applying it to human data is the mistake the default invites. ~496 MB, opt-in.",
+            organism: "mouse",
+            sourceUrl: "https://figshare.com/articles/dataset/ProjecTILs_murine_reference_atlas_of_tumor-infiltrating_T_cells_version_1/12478571",
+            license: { identifier: "CC-BY-4.0", url: "https://creativecommons.org/licenses/by/4.0/" },
+            recommendation: { group: "cell-typing", recommended: false },
+            artifacts: [
+                {
+                    path: "ref_TILAtlas_mouse_v1.rds",
+                    url: "https://ndownloader.figshare.com/files/41398167",
+                    format: "rds",
+                    contents: "Seurat reference object carrying the atlas embedding and its cell-state labels, in MGI mouse gene symbols. Pass the file path to ProjecTILs' reference loader. For human queries use the human CD8 or CD4 atlas instead; converting a human query to mouse orthologs to reuse this one loses the genes that differ between species, which are disproportionately immune genes.",
+                },
+            ],
+        },
         {
             id: "azimuth-pbmc",
             version: "1.0.0",

package/dist/runtime/boot.d.ts

@@ -20,11 +20,17 @@
  *      misconfigured pool fails loudly at boot, not under load.
  *   5. `assembleCoreRuntime(core)` — registers the durable-workflow cohort and
  *      builds the conversation agent (register-before-launch invariant).
- *   6. `beforeLaunch()` — embedder hook for host-specific pre-launch work that
+ *   6. `backfillConversationDisplayEnvelopes(...)` — validates every stored
+ *      display envelope and freezes one for each legacy turn that has none, so
+ *      the runtime read path never meets a row without a display projection.
+ *      Runs AFTER assembly because migrating a legacy call's one-line detail
+ *      needs the assembled conversation roster, embedder-contributed tools
+ *      included; assembly starts no traffic, so this is still before any read.
+ *   7. `beforeLaunch()` — embedder hook for host-specific pre-launch work that
  *      must attach before DBOS launch re-emits events (scheduled sweeps, an
  *      legacy-workflow reap, an agent-switch install). Runs after registration
  *      so it may close over the registered callables.
- *   7. `launchDbos(...)` — the last registration-dependent step.
+ *   8. `launchDbos(...)` — the last registration-dependent step.
  *
  * Boot-step errors PROPAGATE (the embedder's composition root catches them and
  * releases whatever it acquired). Only `shutdown` swallows per-step failures —