Card cleanup: trim discovery-detail prose, drop project lists + misleading source line, share biospecimen tree + survival section across both cards, reorder so survival_derived sits after the GDC-source story, add non-affiliation note to disclaimer.

README.md

@@ -145,16 +145,12 @@ configs:
 
 # TCGA Patients (Open Access)
 
-Open-access patient data from The Cancer Genome Atlas (TCGA), pulled from the
-National Cancer Institute (NCI) Genomic Data Commons (GDC). One HuggingFace
-(HF) subset per TCGA project; one row per patient.
-
-- **Projects included:** `TCGA-ACC`, `TCGA-BLCA`, `TCGA-BRCA`, `TCGA-CESC`, `TCGA-CHOL`, `TCGA-COAD`, `TCGA-DLBC`, `TCGA-ESCA`, `TCGA-GBM`, `TCGA-HNSC`, `TCGA-KICH`, `TCGA-KIRC`, `TCGA-KIRP`, `TCGA-LAML`, `TCGA-LGG`, `TCGA-LIHC`, `TCGA-LUAD`, `TCGA-LUSC`, `TCGA-MESO`, `TCGA-OV`, `TCGA-PAAD`, `TCGA-PCPG`, `TCGA-PRAD`, `TCGA-READ`, `TCGA-SARC`, `TCGA-SKCM`, `TCGA-STAD`, `TCGA-TGCT`, `TCGA-THCA`, `TCGA-THYM`, `TCGA-UCEC`, `TCGA-UCS`, `TCGA-UVM`
-- **Generated:** 2026-05-09 01:37:04 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
-  each project's `gdc_status.json` for provenance.
+Open-access TCGA data from the NCI Genomic Data Commons (GDC), reshaped as
+one HuggingFace (HF) subset per TCGA project; one row per patient. Covers
+all 33 TCGA projects.
+
+- **Generated:** 2026-05-09 20:13:48 UTC
+- **Schema:** derived from the [GDC Data Dictionary][gdc-dict].
 - **GDC data release:** Data Release 45.0 - December 04, 2025
 
 ## Data model
@@ -172,10 +168,6 @@ case          one patient (TCGA-XX-1234)
             └── aliquot    a vial of that analyte handed off for sequencing
 ```
 
-Every `days_to_*` field anchors to the case's `index_date` (TCGA: almost
-always `"Diagnosis"`), per the dictionary, so clinical and biospecimen events
-share a single timeline.
-
 
 ### Where this dataset deviates from the GDC
 
@@ -194,92 +186,40 @@ tree:
   `Matched_Norm_Sample_UUID`; we additionally resolve those to
   `tumor_sample_id` / `matched_normal_sample_id` so consumers can join
   straight to `samples[]`.
-- **Lifted expression QC counts.** Each Gene Expression Quantification
-  record has the STAR per-feature quality-control counts `N_unmapped`,
-  `N_multimapping`, `N_noFeature`, `N_ambiguous` lifted from the source
-  Tab-Separated Values (TSV) file onto the row as scalar fields. The
-  `stranded_first` / `stranded_second` columns are dropped — the GDC
-  pipeline harmonizes by [treating all RNA-Seq reads as
+- **Stranded RNA-Seq columns dropped.** Each Gene Expression record drops
+  `stranded_first` / `stranded_second`; the GDC pipeline [harmonizes
+  RNA-Seq reads as
   unstranded](https://docs.gdc.cancer.gov/Data/Bioinformatics_Pipelines/Expression_mRNA_Pipeline/),
   so `unstranded` is the canonical column.
 
-## Survival endpoints (`survival_derived`)
-
-Four standard survival endpoints — Overall Survival (OS), Disease-Specific
-Survival (DSS), Progression-Free Interval (PFI), Disease-Free Interval (DFI)
-— **re-derived** from the current GDC data using the algorithm published by
-**Liu et al. 2018** ([DOI 10.1016/j.cell.2018.02.052](https://doi.org/10.1016/j.cell.2018.02.052)).
-
-Each patient row carries a top-level **`survival_derived` struct** with eight sub-fields: `os_event` / `os_time`, `dss_event` / `dss_time`, `pfi_event` / `pfi_time`, `dfi_event` / `dfi_time`. `*_event` is 0/1 (event observed vs censored); `*_time` is days
-from `index_date` (TCGA: diagnosis date). DFI is null for SKCM / THYM / UVM
-/ LAML — Liu specifies no DFI for those tumor types.
-
-**Implementation:** [`tcga2hf_pipeline.survival`][repo-survival]. The
-algorithm is the OR-of-three disease-free check Liu's STAR Methods
-specifies, augmented with `treatment_outcome_first_course` from the BCR
-biotab Clinical Supplements (Liu's original BCR-form data; the modern
-GDC harmonized API drops or under-populates this field).
-
-**Validation:** see [`dev_research/liu_2018/report.html`][repo-liu-report]
-for the full reproduction. Headline agreement against Liu's curated 2018
-CDR (both NA OR populated within 30 days):
-
-| Endpoint | Agreement |
-|---|---|
-| OS  | 98.2% |
-| DSS | 93.2% |
-| PFI | 96.3% |
-| DFI | 90.1% |
-
-**Why we don't ship Liu's curated 2018 values directly:** the CDR is a
-frozen 2018 snapshot derived from a since-modified GDC release. Including
-those values in this dataset would lock in irreproducible source-data
-drift. We re-derive on every build instead, so the values you see here
-always reflect the current GDC and are reproducible from this dataset's
-other tables alone.
-
-[repo-survival]: https://github.com/galtay/tcga2hf/blob/main/packages/tcga2hf-pipeline/src/tcga2hf_pipeline/survival.py
-[repo-liu-report]: https://github.com/galtay/tcga2hf/blob/main/dev_research/liu_2018/report.html
-
-## How this dataset is built
-
-Three GDC REST endpoints feed every row, with filters and field lists
-constructed in [`src/tcga2hf/clinical.py`][src-clinical] and
-[`src/tcga2hf/genomic.py`][src-genomic] of the [`tcga2hf` repo][repo].
-
-**Projects fetched in this build:** `TCGA-ACC`, `TCGA-BLCA`, `TCGA-BRCA`, `TCGA-CESC`, `TCGA-CHOL`, `TCGA-COAD`, `TCGA-DLBC`, `TCGA-ESCA`, `TCGA-GBM`, `TCGA-HNSC`, `TCGA-KICH`, `TCGA-KIRC`, `TCGA-KIRP`, `TCGA-LAML`, `TCGA-LGG`, `TCGA-LIHC`, `TCGA-LUAD`, `TCGA-LUSC`, `TCGA-MESO`, `TCGA-OV`, `TCGA-PAAD`, `TCGA-PCPG`, `TCGA-PRAD`, `TCGA-READ`, `TCGA-SARC`, `TCGA-SKCM`, `TCGA-STAD`, `TCGA-TGCT`, `TCGA-THCA`, `TCGA-THYM`, `TCGA-UCEC`, `TCGA-UCS`, `TCGA-UVM`.
-
-### `POST /cases`
-
-Filter: `project.project_id IN [<projects above>]`. The request `expand`s
-the full nested `case` structure (demographic + diagnoses → treatments +
-follow_ups + exposures + family_histories + samples → portions →
-analytes → aliquots). The response is captured verbatim into
-`<data-dir>/raw/<project>/cases.json` per project and feeds the case-level scalars + the nested `demographic`, `diagnoses[]`, `follow_ups[]`, `exposures[]`, `family_histories[]`, and `samples[]` columns of each patient row. See
-`tcga2hf.clinical.TOP_LEVEL_FIELDS` and `tcga2hf.clinical.EXPANSIONS`
-for the exact field/expand lists.
-
-### `POST /files`
-
-One POST per (project, modality). All requests share
-`cases.project.project_id = <project>` AND `access = open`. The
-remaining clauses lock the format / experimental strategy / workflow
-type so future GDC additions can't silently ship a different pipeline
-under the same `data_type`. Each row's molecular content lands in two top-level array columns on the same patient row:
+## How this dataset was built
+
+Every row originates in the NCI Genomic Data Commons (GDC). Three
+sources feed each project's data:
+
+- **Case-level clinical structure** — fetched from the GDC `/cases`
+  endpoint, returning the full nested case JSON (demographic + diagnoses
+  → treatments + follow_ups + exposures + family_histories + samples →
+  portions → analytes → aliquots).
+- **Per-modality molecular files** — discovered via `/files` (filtered
+  by the clauses in the table below) and downloaded via `/data`. Each
+  combination locks one `data_type` to a specific GDC pipeline so future
+  GDC additions can't silently ship different content under the same
+  `data_type`.
+- **BCR Clinical Supplement files** — original Biospecimen Core Resource
+  (BCR) clinical biotab forms (per-project TSVs: patient, follow_up,
+  nte, drug, radiation, etc.). The harmonized `/cases` endpoint drops
+  or under-populates a number of clinical fields that the BCR-original
+  biotabs preserve — most importantly Liu et al. 2018's
+  `treatment_outcome_first_course`, used by `survival_derived`.
+  Discovered via `/files` filtered to `data_type="Clinical Supplement"`
+  + `data_format="bcr biotab"`, downloaded via `/data`.
 
 | Lands at | data_type | data_format | data_category | experimental_strategy | analysis.workflow_type |
 |---|---|---|---|---|---|
 | `samples_masked_somatic_mutation` array | `Masked Somatic Mutation` | `MAF` | `Simple Nucleotide Variation` | `WXS` | `Aliquot Ensemble Somatic Variant Merging and Masking` |
 | `samples_gene_expression_quantification` array | `Gene Expression Quantification` | `TSV` | `Transcriptome Profiling` | `RNA-Seq` | `STAR - Counts` |
-
-See `tcga2hf.genomic.MODALITY_FILTERS` and `tcga2hf.genomic.FILE_FIELDS`
-for the full request payload.
-
-### `POST /data` → file bytes
-
-UUIDs returned by `/files` are batched (≤50 per request) into `POST
-/data`; the response is a tar.gz of those files. Mutations files are parsed row-by-row into each patient's `samples_masked_somatic_mutation` array; expression files are parsed and projected into `samples_gene_expression_quantification`. See
-`tcga2hf.gdc.bulk_download` for the batching and retry logic.
+| consumed in-memory for `survival_derived` | `Clinical Supplement` | `bcr biotab` | `Clinical` | | |
 
 ### Provenance pinned per build
 
@@ -289,8 +229,34 @@ UUIDs returned by `/files` are batched (≤50 per request) into `POST
   saved alongside the raw data; its SHA-256 is recorded in
   `gdc_status.json`.
 
-[src-clinical]: https://github.com/galtay/tcga2hf/blob/main/src/tcga2hf/clinical.py
-[src-genomic]: https://github.com/galtay/tcga2hf/blob/main/src/tcga2hf/genomic.py
+See the [repository][repo] for full request payloads, filter clauses,
+and the build pipeline source.
+
+## Survival endpoints (`survival_derived`)
+
+The only value-added enrichment in this dataset. Four standard survival
+endpoints — Overall Survival (OS), Disease-Specific Survival (DSS),
+Progression-Free Interval (PFI), Disease-Free Interval (DFI) — re-derived
+from the current GDC data using the algorithm from **Liu et al. 2018**
+([DOI 10.1016/j.cell.2018.02.052](https://doi.org/10.1016/j.cell.2018.02.052)).
+
+Each patient row carries a top-level **`survival_derived` struct** with eight sub-fields: `os_event` / `os_time`, `dss_event` / `dss_time`, `pfi_event` / `pfi_time`, `dfi_event` / `dfi_time`. `*_event` is 0/1 (event observed vs censored); `*_time` is days
+from `index_date` (TCGA: diagnosis date). DFI is null for SKCM / THYM /
+UVM / LAML — Liu specifies no DFI for those tumor types.
+
+We've reimplemented Liu's method against the current TCGA data and find
+broad agreement with the original curated CDR. Differences exist and are
+expected: this is a newer release of the underlying GDC data, so
+re-curated clinical values, post-2018 patient additions, and schema
+migrations all contribute to the gap. This work is evolving; see the
+[repository][repo] for the full reproduction report and per-endpoint
+methodology.
+
+**Why we don't ship Liu's curated 2018 values directly:** the CDR is a
+frozen 2018 snapshot derived from a since-modified GDC release. Including
+those values would lock in irreproducible source-data drift. We re-derive
+on every build, so the values reflect the current GDC and are reproducible
+from this dataset's other tables alone.
 
 ## Loading
 
@@ -364,9 +330,12 @@ Policy references:
 
 ## Disclaimer
 
-Re-derive from the GDC for any analysis where freshness matters; the
-dataset reflects the GDC release pinned in each project's
-`gdc_status.json`. The pipeline that builds this dataset is on GitHub at
+**This project is not affiliated with the NCI, GDC, or the TCGA Research
+Network.** It is an experimental open-source pipeline that may change
+significantly between versions; pin the GDC release and dataset commit
+if you depend on a specific snapshot. Re-derive from the GDC for any
+analysis where freshness matters — the dataset reflects the GDC release
+pinned in each project's `gdc_status.json`. Pipeline source:
 [`galtay/tcga2hf`][repo].
 
 [gdc-dict]: https://docs.gdc.cancer.gov/Data_Dictionary/