Skip to content
Meadowlark
Meadowlark · Data

Meadowlark Data

Surfaced, explained, traced.

Analytics and oversight for the people who run the organization: unusual patterns surfaced, explained in plain language, traced to the record. A working demo on synthetic data.

Documentation

Status

State
Demo

What it is

Meadowlark Data is analytics and oversight for the people who run the organization. Unusual patterns in operational data are surfaced automatically, explained in plain language, ranked by likely cause, and traced back to the record that produced them.

A working demo exists. It runs on synthetic data for a fictional organization, on one machine, with no door in front of it. It is not deployed, and it is not yet a product. This page says what it does today and what it would take to change that.

What's inside

Ten detectors
Timely documentation, no-shows, overlapping visits, coding mix, authorization overruns, overdue screenings, referral collapse, crisis geography, missing-data drift, outcome deterioration. Each is classical statistics against a peer group, not a model.
The explainer
Attribution, co-signals and ranked hypotheses for each finding, in sentences a director can read.
Context
Public benchmarks loaded as retrieved, with source, date and license recorded beside every tile.
A risk layer
An illustration of what a clinical risk-grouping system would add, simulated from the same inputs. It is a sketch of the shape, not a licensed method.

What it refuses to do

  • It uses no language model. Every finding is deterministic and can be recomputed.
  • It fakes nothing. A tile whose data did not load is omitted, never filled in.
  • No client name leaves the server. People appear as codes and staff as initials.
  • It never writes to the source. Every statement against the original database is a read, and the clone it works on is checked by identity before anything runs.

What has been measured

Ten anomalies are planted on a clone of the data. The detectors and the explainer do not read the plant, and the ledger reports which were matched, which were missed, and what else was found. It is a recall check, not a formal evaluation, and the page says so.

What it would take to become a product

A deploy, so it runs somewhere other than a laptop.
The family's shared door in front of it.
An evaluation against more than the planted ten, stamped with the run that produced it.
A name that is not a working title.

Also in the lab

Discover

A way to reconstruct how an environment works from what is available, keeping findings, assumptions, contradictions and open questions apart. A name and a design, not yet code.

Capture

Sixty-second visit-outcome capture on a phone between sessions. Pick the client, say what happened, answer the instruments that are due, sign. A working prototype inside Care, not deployed; nothing is saved yet.

View

A patient-record workspace that adapts to the clinician's role. The prototype Care's Patient 360 was ported from.