Blame for data/schema.json

{
  "version": "1.2.1",
  "study_date": "2026-10-30",
  "revision_date": "2026-11-02",
  "encoding": "UTF-8",
  "csv_nested_values": "Lists and dictionaries are JSON strings; JSONL preserves nested structure.",
  "units": {
    "account": "Frozen publishing identity; not necessarily a unique person or original builder.",
    "model": "Frozen repository entry; not necessarily a unique training run.",
    "person": "Source-linked deduplicated identity in the35-person detailed evidence set."
  },
  "files": {
    "account_inventory": "All495accounts; reviewed type, geography, model summaries and all applicable linked-person reasons.",
    "model_inventory": "All917models; source-coded provenance with evidence level, separate model-builder decisions when available.",
    "people_classification": "35deduplicated evidence people;19primary UK connections and16context records.",
    "model_decisions": "55detailed model records for UK-related or historical/context assessments.",
    "builder_classification_inputs": "Compact public curated evidence input to classify_builders.py.",
    "account_inventory_annotations": "Public account evidence input to classify_inventory.py.",
    "model_inventory_annotations": "Public model evidence input to classify_inventory.py.",
    "builder_descriptive_summary": "Multiple-response counts by institutional/independent route; explicit people denominators.",
    "context/dfe_education_context": "All222official sex/FSM/ethnicity observations for the Englandage15in2020/21 cohort followed toHEby19in2024/25.",
    "context/dfe_education_comparisons": "24major/detailed ethnicity rows; foursex/FSMrates, all-sex totals and originaldenominators/entrantcounts."
  },
  "provenance_values": [
    "from_scratch",
    "derived",
    "untrained",
    "mirror",
    "unresolved"
  ],
  "evidence_levels": "Publisher statements, reviewed training evidence, detailed assessments, exact-file identity and labelled development-evidence inferences are recorded per release. Inferences retain confidence and basis.",
  "uk_build_status": {
    "yes": "UK-based individual or joint model development, with direct or explicitly labelled dated-location inference.",
    "no": "Saved model-specific evidence places relevant building outside UK.",
    "unknown": "Builder location at build time not established; current overseas account location is not automatically historical build location.",
    "not_applicable": "Mirror or non-building contribution where builder location is not attributed to publisher."
  },
  "strict_independence": "At least one UK-built from-scratch natural-language model, no university attendance anywhere, no institutional affiliation ever, affirmative independent project and funding evidence. A tech day job is an affiliation even when project sponsorship is unknown.",
  "location_status": {
    "uk_public_location": "Public UK personal or labelled entity location.",
    "uk_affiliation_only": "UK institutional link without separate public location.",
    "non_uk_public_location": "Sourced non-UK geographic connection; may be institutional or historical rather than residence.",
    "mixed_locations": "Mixed/distributed/conflicting sources; row evidence distinguishes these.",
    "unknown": "No attributable geographic connection established."
  },
  "privacy": "No private source-access logs, transcripts, instructions, contact data or inferred demographics are included.",
  "geographic_assessment": {
    "uk_source_connection": "Sourced UK personal, professional or entity connection.",
    "uk_estimate": "Best country/base estimate is UK; source confidence and period recorded.",
    "uk_affiliation_only": "UK institutional affiliation without a separately located base.",
    "mixed_connections": "Distributed, mixed or conflicting geographic connections.",
    "non_uk_connection_or_estimate": "Sourced non-UK connection or best country/base estimate.",
    "region_or_timezone_clue": "Region or timezone evidence when country remains unestablished.",
    "unresolved": "No supported country or regional assessment after follow-through."
  },
  "recursive_location_estimate": "Best country/base or regional estimate, confidence, evidence type, period and source basis. Recorded separately from builder location at training.",
  "upstream_attribution": "Original training, architecture/code lineage, inherited checkpoint, subsequent operation and creator contribution are separate fields."
}