Precision measurement from item to score to credential
Every LearnVyx assessment runs the same four-stage pipeline: item calibration, adaptive delivery, behavioral signal analysis, and credential issuance. Here's the science behind each stage.
Item Response Theory — why it outperforms static tests
Classical Test Theory (CTT) scores are sample-dependent — a difficult test administered to strong students looks like a weak test. IRT scores are sample-independent. A learner's ability estimate is meaningful regardless of which items they answered.
Four stages from launch to credential
Item Bank Calibration
Items are piloted with a minimum 150-respondent sample. EM-algorithm estimation produces initial a, b, c parameter estimates. Items outside acceptable discrimination or guessing thresholds are flagged for revision before live administration.
Adaptive Delivery
Assessment launches via LTI 1.3 (LMS) or REST API (any platform). The adaptive engine selects the next item in real time. Average session length: 18–22 items for a targeted domain. Completion triggering is configurable — min items, max SEM, or time limit.
Behavioral Signal Analysis
Response latency, confidence ratings, revision patterns, and person-fit statistics (infit/outfit) are combined into the Response Profile Integrity Score (RPIS). Final ability estimate is Bayesian-penalized if RPIS falls below threshold.
Score & Credential Issuance
Scores return to the LMS gradebook (AGS), to your HRIS via webhook, or via the REST API. Passing learners receive an Open Badges 3.0 / CLR-formatted digital credential with a public verification URL. Credentials include the ability estimate and RPIS metadata.
REST API — embed anywhere
LearnVyx exposes a full REST API. Launch assessment sessions, retrieve scores, issue credentials, and query item bank metadata — from any platform, any stack.
- Bearer token authentication
- Webhook support for score events
- JSON response format throughout
- Sandbox environment for development
$ curl -X POST https://api.learnvyx.com/v1/assessments/sessions \
-H "Authorization: Bearer lvx_prod_sk_xxxxxxxxxxxx" \
-H "Content-Type: application/json" \
-d '{"assessment_id":"asmnt_8xR2vN","learner_id":"lrn_49FxZq"}'
{
"session_id": "sess_3KpWmB7j",
"launch_url": "https://assess.learnvyx.com/s/3KpWmB7j",
"expires_at": "2026-06-11T18:30:00Z"
}
$ curl https://api.learnvyx.com/v1/sessions/sess_3KpWmB7j/score \
-H "Authorization: Bearer lvx_prod_sk_xxxxxxxxxxxx"
{
"theta": 1.34,
"se": 0.27,
"rpis": 0.91,
"items_administered": 19,
"credential_url": "https://verify.learnvyx.com/c/BqM7R2"
}Technical & implementation FAQ
The minimum practical item bank size for a 20-item adaptive assessment is 80–100 calibrated items per domain. The engine needs a sufficient pool above and below the target ability range to select optimally without re-exposure. We recommend a minimum of 120 items for a production bank to allow for content balancing and exposure control.
Initial calibration requires a minimum of 150 respondents per item for stable parameter estimates. If you already have response data from a previous platform, we can perform retrospective calibration on your existing data to bootstrap parameters before going live. Cold-start (no existing data) requires a pilot phase of 2–4 weeks depending on your learner population size.
RPIS is a session-level integrity score ranging from 0 to 1. A score above 0.85 indicates a response profile consistent with genuine ability-based performance. A score below 0.65 indicates a statistically improbable pattern that may reflect external assistance, unusual response behavior, or technical test conditions (e.g., very slow internet introducing latency artifacts). Scores between 0.65 and 0.85 are flagged for human review. RPIS does not replace human judgment — it surfaces cases that warrant review, with a confidence interval attached.
Yes. LearnVyx accepts QTI 2.2 format for item import, as well as our JSON-based item schema for API-based authoring workflows. If you currently use a third-party item authoring tool (e.g., TAO Platform, Learnosity Author), we can work with your vendor on a data export. We also offer our own authoring interface if you prefer to author natively.
LearnVyx issues Open Badges 3.0 (OB3) credentials by default, which are verifiable via any OB3-compliant verifier. We also support Comprehensive Learner Record (CLR) 2.0 format for HCM platforms that ingest structured learning records. All credentials include a public verification URL that resolves to a verification page without requiring the verifier to have an account — and include the ability estimate (θ) and RPIS as metadata claims.
See the pipeline run on your content
Book a technical walkthrough. We'll run an assessment in your subject domain, share the item-level IRT output, and show you the RPIS in action on known-integrity test cases.