evidence / Engineering depth
The Data Layer -- a leak-safe, as-of-stamped, keyless multi-sport data platform that audits its own completeness
Data engineering at scale, built to be reproducible and honest about what it does not yet have. The single truth-source for any figure below is docs/JOB_EVIDENCE_PACKET.md (sections F and G); the full pipeline/cache reference is docs/DATA.md. data/ is local-only and gitignored, so every number here is read from a committed artifact under scripts/platformkit/analytics_showcase/out/ or a committed builder, never from the private corpora themselves.
strongest single receipt
the claim
The platform ingests four sports from keyless-first public sources, folds them into point-in-time-correct feature corpora that are leak-safe by construction, and then audits its own completeness and publishes the audit. No number is presented without its provenance and an honesty label; a feed miss is logged and skipped, never filled with a guess. This is a calibration/measurement layer -- no dollar/edge/ROI claim is made anywhere, and the artifacts carry that fact in their own metadata.
receipts
| claim | receipt | committed artifact |
|---|---|---|
| MLB Statcast coverage | 693,037 pitches; framing cols 99.63-100% non-null | analytics_showcase/out/statcast_showcase.json |
| Fact-claims corpus | 103,048 generated / 103 families; 101,864 of 101,865 sampled verified (98.85% covered) | analytics_showcase/out/claims_corpus_meta.json |
| Matchup matrix | 291,625 pairs from 2,214 tracking files / 3 seasons | data/cache/coverage_faced_allseasons.parquet; scripts/intel/build_coverage_allseasons.py |
| As-of feature builders | shift(1) expanding-window; debut => NaN assertion | scripts/platformkit/asof_common.py; scripts/intel/outcome/build_player_availability.py |
| Keyless acquisition pipelines | 6 pipelines; `bbref_advanced_extended.parquet` = 1,470 rows | scripts/platformkit/data_frontier/*.py |
| Completeness auditor | NBA: 61 attributes / 6 windows / 505 players / 0 all-null | scripts/platformkit/data_frontier/profile_completeness.py -> data/frontend/ops/profile_completeness.json |
why this matters
The hire signal here is not raw volume -- it is that the volume is reproducible, provenance- stamped, and self-audited. The hard part of data engineering is not fetching rows; it is building corpora that cannot leak, that rebuild identically from scratch, and that tell you where the gaps are before a model quietly trains on a hole. This layer does all three: leak-safety is a construction invariant with an assertion behind it, keyless-first sourcing removes a paid-API dependency for the default slate, and the completeness auditor plus the generated-vs-validated split mean no number reaches a reviewer without its own disclaimer.
This site reports calibration and sharpness only, never a dollar edge, ROI, or bankroll result. An honest null is a success.