Our technology

From sound to score.

A six-stage pipeline that turns one hour of session audio into clinical evidence.

6 stages 1 score every finding timestamped audio only — no cameras

Millions

of hours of speech behind the foundation models we build on

clinical domains scored per session

40+

observable criteria from the BACB RBT task list

functional classes for every therapist utterance

0.01 s

timing resolution — every pause measured

100%

of findings timestamped & traceable

Internal evaluation: 96.6% agreement with expert review · formal validation study in progress

The pipeline

Six stages. Every one of them measurable.

Signal

One studio standard, so no session is judged by its microphone.

Transcription

Every word timestamped to the millisecond.

Separation

Therapist and client, told apart.

Reconstruction

Streams fused, and every pause timed.

Classification

Each utterance labeled; small talk excluded.

Scoring

Measured against the BACB RBT Task List.

How scoring works

Judgment with rules.

The final stage reviews the full timestamped transcript the way a senior clinician would, following the same written discipline every time.

RULE 01

Only what’s observable.

If the transcript can’t show it, it isn’t scored. No inference, no vibes.

RULE 02

Timing is testimony.

Every measured pause is evidence: reinforcement speed, response time, trial pacing.

RULE 03

Patterns over one-offs.

One slow moment is noise; the same issue repeating is signal. Consistency is what’s weighed.

RULE 04

Both sides of the ledger.

Strengths are named with the same rigor as issues. A fair record, not a complaint list.

Every session returns: a 0–100 score · 8 domain scores · a competency level · timestamped strengths & issues, ranked by severity · a plain-language clinical summary

The standard

We don’t grade on vibes. We measure against numbers.

Reinforcement delivered after a correct response < 2.0 s C.1

Wait time after an instruction, before prompting 3–5 s C.3

Interval between teaching trials 1–5 s C.3

Prompting follows a least-to-most hierarchy wait → gesture → verbal → model C.7

Praise names the behavior, not just “good job” behavior-specific C.1

Reinforcement is contingent, never after an error 100% contingent D.3

Respectful, age-appropriate tone throughout client dignity F.5

BACB® and RBT® are registered marks of the Behavior Analyst Certification Board. EvidentABA is an independent product and is not affiliated with or endorsed by the BACB.

The output

Findings, not impressions.

Every finding the pipeline produces carries a timestamp, the task it maps to, and the evidence behind it — so a score is never something you have to take on faith.

[ 12:41 ] C.1 · Reinforcement Strength

Behavior-specific praise delivered 0.8 s after a correct response.

“Nice job touching red!”

[ 27:03 ] C.7 · Prompting Needs attention

Full verbal prompt given 1.2 s after the instruction — before the client had time to respond independently.

Standard: wait 3–5 s, then prompt least-to-most.