Card: add How this dataset is built section with verbatim GDC request payloads

README.md

@@ -30,7 +30,7 @@ National Cancer Institute (NCI) Genomic Data Commons (GDC). One HuggingFace
 (HF) subset per TCGA project; one row per patient.
 
 - **Projects included:** `TCGA-CHOL`, `TCGA-DLBC`, `TCGA-LUAD`
-- **Generated:** 2026-05-05 03:50:12 UTC
+- **Generated:** 2026-05-05 05:47:17 UTC
 - **Source:** GDC `/cases` endpoint, open-access tier only.
 - **Schema:** derived from the [GDC Data Dictionary][gdc-dict]; the live
   dictionary the GDC was serving when the data was fetched is hashed into
@@ -82,6 +82,114 @@ tree:
   unstranded](https://docs.gdc.cancer.gov/Data/Bioinformatics_Pipelines/Expression_mRNA_Pipeline/),
   so `unstranded` is the canonical column.
 
+## How this dataset is built
+
+Three GDC REST endpoints feed every row in this dataset. Both endpoints
+and payloads below are reproduced verbatim from the build code; the
+on-disk `gdc_status.json` for each project additionally pins the GDC
+data release and dictionary SHA-256 the data was fetched against.
+
+### `POST /cases` — clinical entities
+
+One paginated POST per build, returning the nested case JSON for every
+patient in the requested projects.
+
+```json
+{
+  "filters": {
+    "op": "in",
+    "content": {
+      "field": "project.project_id",
+      "value": [
+        "TCGA-CHOL",
+        "TCGA-DLBC",
+        "TCGA-LUAD"
+      ]
+    }
+  },
+  "fields": "case_id,submitter_id,primary_site,disease_type,index_date,consent_type,days_to_consent,days_to_lost_to_followup,lost_to_followup",
+  "expand": "demographic,diagnoses,diagnoses.treatments,follow_ups,exposures,family_histories,project,samples,samples.portions,samples.portions.analytes,samples.portions.analytes.aliquots",
+  "format": "JSON",
+  "size": 200,
+  "from": 0
+}
+```
+
+The full nested response (top-level scalars + `demographic` + `diagnoses[]`
+with `treatments[]` inside + `follow_ups[]` + `exposures[]` +
+`family_histories[]` + `samples[]` with `portions[].analytes[].aliquots[]`
+inside) is written to `<data-dir>/raw/<project>/cases.json` and feeds the
+`cases` config (and the consolidated patient row).
+
+### `POST /files` — molecular file discovery
+
+One POST per (project, data_type) pair, listing every open-access file
+of that type. Below is the request for `Masked Somatic Mutation` against
+`TCGA-LUAD`; the build also runs the same shape with
+`data_type` = `"Gene Expression Quantification"`. The responses are
+written to `<data-dir>/raw/<project>/<modality>/manifest.json` and feed
+the `files` config.
+
+```json
+{
+  "filters": {
+    "op": "and",
+    "content": [
+      {
+        "op": "=",
+        "content": {
+          "field": "cases.project.project_id",
+          "value": "TCGA-LUAD"
+        }
+      },
+      {
+        "op": "=",
+        "content": {
+          "field": "access",
+          "value": "open"
+        }
+      },
+      {
+        "op": "=",
+        "content": {
+          "field": "data_type",
+          "value": "Masked Somatic Mutation"
+        }
+      }
+    ]
+  },
+  "fields": "file_id,file_name,file_size,md5sum,data_category,data_type,data_format,experimental_strategy,workflow_type,access,cases.case_id,cases.submitter_id,cases.project.project_id,cases.samples.sample_id,cases.samples.submitter_id,cases.samples.sample_type,cases.samples.tissue_type,cases.samples.portions.analytes.aliquots.aliquot_id,cases.samples.portions.analytes.aliquots.submitter_id",
+  "format": "JSON",
+  "size": 500,
+  "from": 0
+}
+```
+
+### `POST /data` — file bytes
+
+For every file UUID returned by `/files`, the build batches up to 50
+UUIDs per request:
+
+```json
+{"ids": ["<file_uuid_1>", "<file_uuid_2>", "..."]}
+```
+
+The response is a tar.gz of the files. Mutations files are parsed
+row-by-row into the `masked_somatic_mutation` config; expression files
+are parsed gene-row-by-gene-row into the
+`gene_expression_quantification` config. (See `tcga2hf.gdc.bulk_download`
+for the batching + retry logic.)
+
+### Other GDC calls captured for provenance
+
+- `GET /status` — once at the top of every build; `data_release` /
+  `tag` / `commit` are saved in each project's `gdc_status.json`.
+- `GET /v0/submission/_dictionary/_all` — once at the top of every build;
+  written to `<data-dir>/raw/gdc_dictionary.<major>.<minor>.json` and its
+  SHA-256 is recorded in `gdc_status.json`. The static schema in
+  `tcga2hf.schema` is regenerated from this dictionary; see the package
+  `scripts/regenerate_clinical_fields.py`.
+
 ## Loading
 
 The [`tcga2hf` package][repo] ships a typed `TcgaHfPatient` pydantic model