Reclaim OS
The engine

Built for the practice. Engineered for the audit.

Audit-grade detail on every layer of the Reclaim recovery pipeline. Read it. Rip it apart. The math holds.

Differential analysis kernel

The kernel is a streaming Postgres pipeline. For every inbound 835 / paid claim row, the kernel hash-keys on (npi, cpt, modifier_set, payer_id, state, dos_quarter) and joins against the live ReimburseOS rate cube. Result: contracted rate at the row’s effective period, in O(1) lookup.

Edge cases handled in-kernel: same-day bundling (modifier 59), multiple-procedure reduction (MPR), timed-code unit accumulation (97110 / 15-min), and POS-modifier corrections. Each correction is logged so the contracted rate cited in the appeal letter matches CMS NCCI guidance for the exact code combination.

$0.005 per claim · sub-200ms p95

Cost target: under half a penny per scanned claim, end-to-end. P95 latency from remit-ingest to flag-or-clear: 187ms in current production load (10K claims / minute test).

Achieved by: (1) materialised rate-cube reads from a hot Supabase replica, (2) in-process LRU on the top-2K NPI×payer×CPT combos (covers ~78% of SMB-practice volume), (3) async queue write to Denial OS via direct Postgres FDW rather than HTTP. No round trips on the critical path.

Cross-product data flow

Reclaim is a citizen of the Twin Flame Revenue Cycle OS. It reads from payers + payer_playbooks (shared schema). It writes to reclaim_underpayments + reclaim_recovery_jobs (Reclaim-owned). It queues into denials with a source = ‘reclaim’ tag, so Denial OS picks the appeal up automatically.

Every flag carries a reclaim_underpayment_id back-reference on the resulting denial row, so recovery → outcome → ledger reconciles end-to-end. No dropped data, no orphan rows, no double-billing.

HIPAA append-only audit

Every scan run, every flag, every queue, every recovery is logged to the shared audit_log table — append-only, RLS-locked, signed with the practice’s row-level ed25519 key. PHI fields encrypted at rest with the family-shared PHI_MASTER_KEY and never logged in plaintext.

HHS audit pack export: one-click CSV + cryptographic manifest of every Reclaim event in the requested window. Delivered in under 10s for any practice up to 100K claims.

The numbers we track for you

  • Total at stake — sum of all open shortfalls
  • Recovered YTD — confirmed back-payment ledger
  • Recovery rate — recovered / queued
  • Days-to-recovery (p50 / p95) — appeal latency
  • Payer rank — who underpays you the most, by absolute and percentage
  • CPT rank — which procedures get clipped most often

Phase 2: live PMS connectors

Phase 1 (today): CSV upload + SFTP drop covers AdvancedMD, Tebra, WebPT, TherapyNotes, ChiroFusion, Athena, NextGen, eClinicalWorks, Kareo, OptiMantra.

Phase 2 (Q3): live API connectors for the same PMS list + Availity/Waystar/Change Healthcare 835 streams. Real-time scan: remit hits → flag is on dashboard within 60 seconds.

The hard question — what about false positives?

Every flag carries a confidence score. Below 0.92 confidence, Reclaim parks the flag in review rather than queued and surfaces it for a billing manager. False-positive sources we explicitly handle: (1) MRF effective-date drift across the claim’s service period, (2) NPI-vs-TIN attribution mismatches in BlueCard scenarios, (3) bundled-care POs where the contracted rate is a global, (4) plan-tier mismatches on self-funded ASO contracts. Reclaim’s own measured first-letter false-positive rate, on production volume: 3.1%.