Home
The engine

A rules engine that proposes and retires its own rules.

Hive continuously scans your operational graph, finds the patterns that matter, and ships the findings with a dollar value, an owner, and the evidence. The detection library grows itself.

How the engine works

A rules engine that proposes
and retires its own rules.

Five steps, on repeat, without anyone asking. Every step is auditable. Every finding traces back to the rows that triggered it.

  1. Step 1
    Ingest

    Residents, orders, MDS, eMAR, labs, staffing, claims.

  2. Step 2
    Detect

    The catalog runs against every row, every night.

  3. Step 3
    Rank

    Dollar value, urgency, and round-robin balance.

  4. Step 4
    Route

    Into one person's queue, with a draft action attached.

  5. Step 5
    Learn

    The engine grades itself and tunes what fires next.

Capped at 40 per facility
A real operator works 20 to 40 open items at a time. Show them 400 and the feature doesn't work. Show them 40 balanced items and it does.
Round-robin admission
Every pattern gets a turn. A single big-dollar compliance finding never saturates the queue and drowns out the small wins.
The miner

The engine proposes new detection patterns on its own.

Four statistical passes run against your operational graph every night. When one spots something, a proposer drafts a new detection pattern, with its own threshold, dollar math, and owner routing, and ships it as data. No code deploy.

Mode A
Anomaly

A facility’s number on a metric is far enough from the chain median to be suspicious. Robust outlier math, not a trigger-happy z-score.

“Hilltop North’s antibiotic start rate is 3.1× the network median.”
Mode B
Correlation

Residents with feature pattern X have adverse event Y happen at materially higher rates than the base population.

“Residents on 3+ sedating meds fall at 2× the facility baseline.”
Mode C
Temporal

Event A tends to be followed by event B inside a bounded window. Lead indicator, not coincidence.

“Antibiotic start to fall within 14 days, 1.8× chance.”
Mode D
Peer drift

A facility’s own metric diverges from its recent history AND the chain trend. Something changed here, and only here.

“Adams Court agency hours are up 240% while the network is flat.”
The flywheel

At deployment, you get the full catalog. Six months later, the engine has retired the patterns that didn't perform on your data and accreted twenty or thirty new ones specific to your chain.

shadowcanaryactivewarningquarantineretired
The reference layer

We see things a chart alone can't see.

Every EHR reads your data. Hive does too, and cross-joins it against ten public federal datasets inside the same query. The most interesting signals live in that join.

Datasets joined, in-query
  • OIG LEIE
    Federal exclusion list, ~83,000 people banned from Medicare
  • CMS Care Compare
    Star ratings, deficiencies, penalties, QMs, per-facility
  • Payroll-Based Journal
    Every SNF’s day-by-day nurse staffing
  • NPPES NPI registry
    Every provider’s National Provider Identifier
  • CMS PECOS
    Who is enrolled to bill Medicare right now
  • HCRIS Cost Reports
    Every facility’s filed Medicare Cost Report
  • Physician Fee Schedule
    Official per-code reimbursement, refreshed monthly
  • HCPCS code catalog
    Official catalog of procedure and drug billing codes
  • NCCI edits
    Code pairs that can’t be billed together
  • Quality Measures
    Public QM scores per facility
Signal
Excluded Party Detected

Hive matches your active payroll against the full OIG exclusion list, monthly. Hiring or retaining an excluded person is a False Claims Act violation with exposure up to $22,927 per claim. Most chains check at hire and never again. The highest-risk scenario is invisible under a check-at-hire process.

Signal
NH Compare External Risk

Your facility's public CMS profile is what hospital referral coordinators, MA plan network teams, private-pay families, and plaintiff firms look at. We surface it next to your internal findings. A 1-star rating from 2023 stops being invisible while you optimize 2026 coding.

See it on your own data

Thirty minutes. Your facilities. Your queue.

We walk a real resident chart, show what the engine surfaced, and let you drill into the evidence. You'll know by the end whether this fits your chain. No slide deck.