AsamAce/pharma-serialized-events

ZigoTrace Pharma — Serialized Events (synthetic)

Feature vectors extracted from a synthetic DSCSA-style serialized medicine supply chain, generated by packages/intelligence/src/synthetic.ts in the zigo-pharma engine and exported via hf/generate_fixtures.mjs. Used to train and validate the diversion-detection model (zigotrace/pharma-authenticity-model).

⚠️ Synthetic data notice

This is entirely synthetic — a seeded generator (generateChain), not real distributor or patient data. It exists to prove the modeling pipeline and give a reproducible gold set with labelled ground truth. A -real variant, gated and under a distributor's consent/DPA terms, is planned once a lighthouse dataset is secured (see docs/huggingface-model-plan.md Phase 5). Do not treat metrics on this set as predictive of real-world performance.

Files

File	Rows	Purpose
train.jsonl	420	training split (fed to train.ts's weak-supervision logistic fit)
test.jsonl	180	held-out split, used for the parity test and dry-run metrics

Schema (one JSON object per line)

{"id": "sgtin:0600100000.LOT-1004.7", "features": [0.83, 1, 1, 1, 1, 1, 4, 0.0], "label": 1, "subclass": "diversion_route"}

Field	Type	Meaning
id	string	the synthetic serial (SGTIN)
features	float[8]	see order below — the model's exact input contract
label	0/1	1 = diversion (route or ring). This is the model's trained scope — auth/clone/expiry/recall/cold-chain gaps are caught by ZigoTrace's deterministic layer, not this model; labelling those as positives here would misrepresent what this specific model is for.
subclass	string	the generator's injected pattern: genuine, diversion_route, diversion_ring, cloned_serial, counterfeit_unverifiable, cold_chain, expired, recalled

Feature order (must match model.json's featureNames)

route_rarity, num_receives, num_dispenses, distinct_receive_locations,
aggregation_consistent, attestation_ok, hop_count, case_dispense_anomaly

case_dispense_anomaly is the relational feature — the fraction of a unit's aggregation-case siblings dispensed at a location rare for the lot. It is what lets the model catch diversion rings (each individual unit looks normal; the pattern is only visible across the case).

Generation & reproducibility

node hf/generate_fixtures.mjs   # regenerates train/test/model/expected_scores from source

Seeded (seed=7), so re-running produces byte-identical output. Provenance: git commit of the zigo-pharma repo at export time (see the model card for the paired revision).

Intended use

Training/validating an advisory diversion-anomaly model that fuses into a larger Dempster-Shafer authenticity-confidence engine — never a standalone classifier, never a release/dispense gate. See zigotrace/pharma-authenticity-model's model card for the full intended-use statement.