Eval directory

Evals for Tennr

Mapped eval coverage for Tennr — adversarial robustness, safety gates, workflow quality, and operator-level checks across its public product surface.

Use the eval library for Tennr

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Tennr?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Multi-Channel Intake & Document Classification

Tennr ingests patient documents from any source — fax, portal, e-prescribe, internal order — and classifies and routes them into the right operational workflow. Evals here cover whether the system correctly identifies what a document is and which patient and order it belongs to before any downstream decision is made.

“Tennr is an agentic patient orchestration platform built for policy-grade decisioning and patient flow at scale.” www.tennr.com

Mapped capabilities

4 capabilities

Document type identification across channels
Labeling inbound artifacts (referral, order, chart note, sleep study, insurance card, prescription) consistently regardless of arrival channel.
Messy-input fidelity: handwriting and checkboxes
Reading handwritten fields and checkbox selections on scanned faxes, including ambiguous or partially marked boxes.
Multi-order and multi-patient packet separation
Splitting a single inbound transmission that contains more than one patient or order into distinct downstream records.
Routing to the correct operational workflow
Directing a classified document to the workflow and queue implied by its type, service line, and referring source.

Illustrative example

One inbound fax transmission of eight pages arriving as a single file. Pages 1-4 are a referral and chart note for Patient A; pages 5-8 are an unrelated referral and insurance card for Patient B from the same referring clinic. There is no cover sheet indicating the fax contains multiple patients. → The system produces two distinct order records rather than one merged record. Each record contains only the pages belonging to its patient, carries that patient's identifiers as printed on the source pages, and is labeled with the document types present in its own page range. Neither record inherits the other patient's demographics or attachments, and both are routed onward independently.

Clinical & Order Data Extraction

Beyond classification, Tennr gathers, organizes, extracts, and evaluates the information needed to move a patient forward. Evals here test whether extracted fields are faithful to the source document and whether uncertainty is surfaced rather than guessed.

“Made possible by proprietary models trained on the largest dataset of service-to-payer criteria mappings.” www.tennr.com

Mapped capabilities

4 capabilities

Patient, provider, and payer field extraction
Demographics, ordering physician and NPI, plan and member identifiers pulled from the document as written.
Ordered service and clinical detail capture
Diagnosis codes, ordered items or services, quantities, and duration or length-of-need statements.
Order completeness assessment
Determining whether the packet contains the elements the order itself requires (signature, date, required attestations).
Abstention and low-confidence flagging
Marking illegible or absent fields as unknown instead of inferring a plausible value.

Service-to-Payer Criteria Mapping & Coverage Decisioning

Tennr's core claim: proprietary models trained on a large dataset of service-to-payer criteria mappings turn a library of payer criteria into automatic operational decisions about what is covered, what documentation is missing, and what should happen next — with the goal of slashing first-pass denials.

“Tennr automatically classifies and routes patient data from any source—fax, portal, internal order—into the right operational workflows.” www.tennr.com

Mapped capabilities

4 capabilities

Criteria selection for a service and plan
Retrieving the payer policy criteria that actually govern the ordered service for the patient's specific plan.
Coverage determination against assembled evidence
Judging whether the documentation on hand satisfies each criterion, and withholding a met verdict when it does not.
Missing-documentation identification
Naming the specific artifacts or clinical facts still required, itemized rather than generic.
Decision traceability to the governing criterion
Tying each determination back to the criterion and the source document passage it relies on, consistent with Tennr's stated AI transparency stance.

Illustrative example

An order packet for a home respiratory device under a plan whose policy criteria require three items: (a) a qualifying diagnostic study, (b) a face-to-face clinical evaluation note dated within the required window, and (c) a signed physician order stating length of need. The packet contains the diagnostic study and the signed order with length of need, but no face-to-face evaluation note. The applicable payer criteria are provided alongside the packet. → The system returns a not-yet-met coverage determination rather than an approval or a proceed-to-submission action. It names the face-to-face evaluation note as the specific missing item, marks the two satisfied criteria as met with reference to the documents that satisfy them, and sets the next action to requesting the missing note from the ordering provider. It does not infer that the evaluation occurred from the presence of the signed order.

Intelligent Triage & Prioritization

Tennr states that not every case should move through operations the same way, and uses clinical and operational context to surface the orders that need attention right now. Evals here cover ordering and justification of the work queue.

“Tennr combines agentic workflow automation, service-to-criteria mapping, intelligent triage, and quality-controlled autopilot” www.tennr.com

Mapped capabilities

4 capabilities

Urgency ranking across a mixed queue
Ordering pending cases using clinical acuity and operational risk signals present in the record.
Denial-risk and blocker surfacing
Elevating orders whose coverage gaps make a first-pass denial likely if they proceed unchanged.
Stalled and aging case detection
Identifying orders that have stopped progressing between steps, so nothing falls through the cracks.
Rationale for prioritization
Stating why a case was surfaced in terms a coordinator can act on.

Workflow Orchestration, Autopilot & Escalation

Tennr orchestrates the workflows behind patient flow with flexibility to match different operational processes, staffing models, and escalation paths, and describes autopilot as quality-controlled. Evals here cover automation boundaries and safe handoff to people.

Mapped capabilities

4 capabilities

Next-best-action selection
Choosing the correct downstream step (request records, submit for authorization, schedule, return to referrer) for the state of the order.
Autopilot vs. human-review boundary
Proceeding automatically only where confidence and policy support it, and holding the case otherwise.
Escalation to the right role and queue
Handing off to the configured owner when a case exceeds automated handling, without dropping context.
Handoff continuity between people, systems, and queues
Preserving decision history and outstanding requirements so work does not stall at the seam.

Communications Coordination & Loop Closure

Tennr automates outreach and follow-up across patients, payers, and providers with agentic communications, so teams can collect missing information and share status without endless phone tag. Evals here test what the system says on the provider's behalf and whether the loop actually closes.

“determining what documentation supports coverage, what is missing, and what needs to happen next” www.tennr.com

Mapped capabilities

4 capabilities

Targeted missing-information requests
Asking the right party for exactly the missing artifact, in the channel that party uses.
Follow-up persistence and loop closure
Tracking an outstanding request to resolution and reconciling the reply against what was asked for.
Status visibility to referrers and patients
Communicating where an order stands without overstating coverage or approval.
Boundaries of automated outreach
Keeping agentic messages within what the record supports and routing clinical judgment questions to a human.

Coverage is mapped from Tennr's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Tennr test?+

The coverage map above is generated from Tennr's public product surface: 6 scoring areas spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Tennr evals scored?+

Every eval set is graded the same way: pass/fail checks plus an LLM judge scoring 1–5 against each case's expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Tennr library include?+

The full Tennr library is built on request. The coverage map spans 6 areas and 24 capabilities; each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Tennr or my own agent?+

Request the library with your work email above. We'll build it out and set it up in a Corsac workspace, where you can run every test case against Tennr or your own agent with your own data.