★★★★★
4.8/5.0 · #1 EHR for PT/OT Rehab

The AI-Native Platform Built for Outpatient Rehab

SPRY connects intake, scheduling, documentation, prior auth, billing, collections, and reporting on one platform. Episteme adds the one layer it does not have: a causal gate that admits an outcome claim only when the graph licenses it, and refuses when it does not.

Ask the gateSee the outcome claims graded
Trusted by 1000+ rehab clinics · Same clinic-day structure, one added gate

The platform — and where the gate sits

Same seven functions SPRY lists. The chip on each says who it belongs to.

Intake

Forms, insurance, consents, history before arrival.

ordinary code

Scheduling

Appointments, providers, locations, waitlist.

ordinary code

Documentation

AI turns the visit into structured notes.

Episteme gate added

Prior auth

Authorization and remaining-visit tracking.

ordinary code

Billing / RCM

Pre-submission claim scrubbing and worklists.

Episteme gate added

Collections

Balances, payment links, denial follow-up.

ordinary code

Reporting

Visits, denials, AR, productivity.

Episteme gate added

Outcome claims, graded

Each claim is exactly as a clinic vendor posts it. The verdict is computed by the real engine against the rehab causal graph — not authored here.

AS POSTED BY THE VENDOR
“SPRY's AI documentation assist reduced clinician documentation burden”
ai_documentation_assist → documentation_burden
Supported
directed causal path ai_documentation_assist -> clinician_documentation_time -> documentation_burden with no unmeasured common cause; the effect is estimable from available telemetry
path: ai_documentation_assist → clinician_documentation_time → documentation_burden
AS POSTED BY THE VENDOR
“SPRY AI documentation cut the time clinicians spend per note”
ai_documentation_assist → clinician_documentation_time
Supported
directed causal path ai_documentation_assist -> clinician_documentation_time with no unmeasured common cause; the effect is estimable from available telemetry
path: ai_documentation_assist → clinician_documentation_time
AS POSTED BY THE VENDOR
“SPRY claim scrubbing reduced claim denials”
claim_scrubbing → denial_rate
Confounded
a causal path exists (claim_scrubbing -> denial_rate), but 1 unmeasured common cause(s) open a backdoor that adjustment cannot close: clinic_growth_trajectory: unobserved. The association is real; attributing 'denial_rate' to 'claim_scrubbing' is not licensed by this data.
path: claim_scrubbing → denial_rate
blocked by: clinic_growth_trajectory
AS POSTED BY THE VENDOR
“SPRY platform adoption drove revenue growth”
ehr_platform_adoption → revenue
Confounded
a causal path exists (ehr_platform_adoption -> claim_scrubbing -> denial_rate -> collections_rate -> revenue), but 1 unmeasured common cause(s) open a backdoor that adjustment cannot close: clinic_growth_trajectory: unobserved. The association is real; attributing 'revenue' to 'ehr_platform_adoption' is not licensed by this data.
path: ehr_platform_adoption → claim_scrubbing → denial_rate → collections_rate → revenue
blocked by: clinic_growth_trajectory
AS POSTED BY THE VENDOR
“SPRY platform improved patient outcomes”
ehr_platform_adoption → patient_outcome_score
Unsupported
no causal path from 'ehr_platform_adoption' to 'patient_outcome_score' and no shared upstream cause. The model contains no mechanism linking them.

Notice the three different dispositions: the documentation-burden claim is supported; the revenue-growth claim is confounded (a real association you may not attribute); the patient-outcome claim has no mechanism at all and is refused. A gate that only said "yes" would be worthless.

Ask the gate

Pick any cause and effect in the rehab graph, or type a claim in words. Verdicts come from a lookup table computed by episteme.verify_text at build time — the page cannot invent one.

A day with SPRY — and the gate

Optional context. The role tabs mirror SPRY's clinic day; the gate shows up where a claim is made.

STEP 1

Book the visit

Right location, provider, appointment type, availability.

STEP 2

Complete forms early

Intake, insurance, consents, outcome measures.

STEP 3

Check in faster

Eligibility, balances, kiosk, reminders.

STEP 4

Continue care

Home programs, follow-ups, payment links.

STEP 1

Prepare for the visit

Schedule, auth, remaining visits, history.

STEP 2

Understand the patient

Goals, prior notes, outcomes, precautions.

STEP 3

Document the session

AI drafts the note — the gate checks each generated field is warranted by the visit record, else it abstains.

STEP 4

Close the loop

Finalize the note, coding, compliance, next-visit readiness.

STEP 1

Fill the schedule

Appointments, availability, cancellations, waitlist.

STEP 2

Get patients ready

Demographics, consents, insurance, forms.

STEP 3

Verify coverage

Eligibility, copays, visit limits, authorizations.

STEP 4

Start smoothly

Kiosk, portal, reminders, balances.

STEP 1

Prepare for the visit

Insurance, auth, visit count, payer rules.

STEP 2

Review the note

Signed documentation, coding, charge readiness.

STEP 3

Submit clean claims

Scrub for missing details and payer rules.

STEP 4

Work the exceptions

Denials, appeals, follow-ups — the gate tells you which denials are attributable to a documentation gap vs payer policy.

STEP 1

See the day clearly

Visits, cancellations, authorizations, AR, denials.

STEP 2

See the money

Claims, collections, productivity across the clinic.

STEP 3

Read the numbers honestly

— the gate reframes a “+37% revenue” headline as a confounded association you cannot attribute to the software.

STEP 4

Grow without overhead

Lean staffing and automation.

Where Episteme adds value — and where it does not

Stated in the open, so the page cannot over-claim.

Episteme's contribution

  • Causal adjudication of outcome & attribution claims (supported / confounded / unsupported)
  • Fail-closed refusal when a variable is unmeasured or out of the graph
  • A groundedness check on AI-generated documentation (cite the source datum or abstain)
  • Identifiability verdict + the named confounding backdoor

Not Episteme — ordinary code

  • The EHR itself: scheduling, intake, kiosk, payment links
  • Claim scrubbing for missing fields and payer rule edits (deterministic rules)
  • Dashes, tabs, and this page's layout
  • Fixing a wrong laterality in a note (that is editing, not adjudication)

Every edge in the rehab graph is author-selected and unsourced. "Supported" means consistent with this graph — it is not evidence the claim is true. That caveat travels with every verdict the engine returns.

Live this run

182ordered pairs adjudicated (14 observed vars)
23supported
17confounded
142unsupported
18graph nodes

Reproduce from the repo:
python3 scripts/verify_claim.py "SPRY platform adoption drove revenue growth" --vertical rehab