# ClinicalMem

> Persistent, auditable, contradiction-safe clinical memory for healthcare AI agents.
> A 6-layer drug-safety pipeline anchored by a 50,949-parameter BitNet b1.58 ternary
> classifier with bit-identical Q16.16 fixed-point replay across phones, browsers,
> $15 Raspberry Pi Zero 2 W boards, NVIDIA A100s, and NVIDIA GB300 devices.

ClinicalMem is published by **STARGA, Inc.** under the **Apache License 2.0**.
This `llms.txt` follows the proposed [llmstxt.org](https://llmstxt.org) standard
so language models can index canonical resources without scraping the marketing
site.

## What ClinicalMem actually claims (verified)

- **PCCP recall cohort: 139 / 139** (44 contraindicated, 4 major, 69 serious, 22 moderate). 100% recall on every severity class · 0 false negatives.
- **Negative-control cohort: 0 / 10 false positives** on adversarial near-miss pairs (clopidogrel + pantoprazole, atorvastatin + amlodipine, simvastatin + diltiazem, spironolactone + trimethoprim).
- **Cross-architecture determinism: 1200 / 1200 bit-identical replays** (12 pairs × 100 iterations) under Q16.16 fixed-point.
- **Python ↔ JavaScript parity**: in-browser BitNet forward pass produces byte-identical `repro_hash` to the server pass.
- **FHIR cohort: 30 patients · 47 NPIs · 239 resources** — all CMS Luhn-valid synthetic identifiers, zero collision with the known-real validation NPI.
- **CI tests: 1425 / 1425** in the engine + scripts suite (pinned floor).
- **Federation: 21 typed runtime invariants** in `JointMemoryFederation.flow.mind`, all 21/21 PASS in the mock end-to-end demo.

## What ClinicalMem deliberately does NOT claim

- Adverse-event reduction outcomes on real patient data — requires IRB-approved cohort + 4–12 week review.
- Vendor-blinded head-to-head accuracy vs Epic BPA / Cerner First Databank — requires NPI-attributed prospective study.
- Alert-fatigue measurement — requires a real prescribing flow.

## Canonical resources

### Code & releases
- **GitHub repository**: https://github.com/star-ga/clinicalmem
- **PyPI package** (memory layer): https://pypi.org/project/mind-mem/ (>= 4.0.0)
- **Hugging Face model card**: https://huggingface.co/stargainc/clinicalmem-bitnet-b158
- **Live demo**: https://clinicalmem-demo.pages.dev/
- **Devpost submission**: https://devpost.com/software/clinimalmem

### Architecture documents
- **6-layer pipeline overview**: https://github.com/star-ga/clinicalmem/blob/main/docs/architecture.md
- **Federated memory & PHI gate**: https://github.com/star-ga/clinicalmem/blob/main/docs/federated_memory.md
- **BitNet training & calibration**: https://github.com/star-ga/clinicalmem/blob/main/docs/bitnet_training.md
- **Clinical validation methodology**: https://github.com/star-ga/clinicalmem/blob/main/docs/clinical_validation.md
- **Edge / Pi Zero offline notes**: https://github.com/star-ga/clinicalmem/blob/main/docs/edge_pi_offline.md

### Audit artifacts (load-bearing)
- **Reproducibility manifest**: https://github.com/star-ga/clinicalmem/blob/main/docs/reproducibility_manifest.json
- **BitNet weights bundle A** (`bundle_id 1f0f8859…`): https://github.com/star-ga/clinicalmem/blob/main/engine/bitnet_weights.json
- **BitNet weights bundle B** (`bundle_id 5f7ed5f6…`): https://github.com/star-ga/clinicalmem/blob/main/engine/bitnet_weights_b_specialist.json
- **Confusion matrix on 139-pair cohort**: https://github.com/star-ga/clinicalmem/blob/main/docs/bitnet_confusion_matrix.json
- **Synthea FHIR cohort**: https://github.com/star-ga/clinicalmem/blob/main/docs/synthea_demo_cohort.json
- **Drug-pair severity cache**: https://github.com/star-ga/clinicalmem/blob/main/docs/openevidence_cache.json
- **Pharmacology flag table**: https://github.com/star-ga/clinicalmem/blob/main/docs/pharmacology_flags.json

### Flows (MIND-language `.flow.mind`)
- AllergyCrossReactivity · ClinicalTrialMatch · JointMemoryFederation · LabContraindication · MedicationSafetyReview · ProviderDisagreement · WhatIfSimulation
- Source dir: https://github.com/star-ga/clinicalmem/tree/main/flows

### Test pins (the bar every PR must clear)
- Federation invariant count: https://github.com/star-ga/clinicalmem/blob/main/tests/test_scripts/test_federation_invariant_count_pin.py
- BitNet cross-architecture determinism: https://github.com/star-ga/clinicalmem/blob/main/tests/test_scripts/test_bitnet_determinism.py
- Synthea cohort integrity: https://github.com/star-ga/clinicalmem/blob/main/tests/test_engine/test_synthea_cohort_integrity_pin.py
- Reproducibility manifest parity: https://github.com/star-ga/clinicalmem/blob/main/tests/test_scripts/test_reproducibility_manifest.py

## How to cite

```
@software{clinicalmem_2026,
  title  = {ClinicalMem: Auditable BitNet b1.58 Drug-Safety Memory for Healthcare AI},
  author = {{STARGA, Inc.}},
  year   = {2026},
  version = {4.1.0},
  url    = {https://github.com/star-ga/clinicalmem},
  license = {Apache-2.0}
}
```

## Contact

- Email: **info@star.ga**
- Security disclosures: see `SECURITY.md` in the repo (synthetic data only — no PHI is ever stored).