Skip to content
Get started

Evidence-based assessment rubric

Compare work artifacts with explicit criteria and performance levels, inspect exact evidence quotations and calculate scores from supplied human ratings.

Input

Rubric criteria CSV: id,name,weight,definition C1,Traceability,2,Claims are linked to identifiable source records. C2,Clarity,1,The method and its limitations are stated explicitly. Performance levels CSV: id,criterion,points,descriptor …

Result

The workshop report is reviewed against traceability and clarity. The supplied traceability rating contributes to the human score; the missing clarity rating keeps the aggregate incomplete under the default policy. A plausible AI clarity…

A useful rubric separates the criterion, its performance descriptors, the submitted evidence and the actual rating. This tool keeps those layers visible. The AI proposes an existing performance level for each artifact and criterion, backed by an unchanged source quotation, or leaves the suggestion unresolved. Deterministic code calculates weighted aggregate scores only from explicitly supplied human ratings. The output is a review draft with evidence gaps and disagreements, not an automatic grading decision.

Your texts, pages and images are processed on our own servers in the EU. Trust Center

Try it now Pricing

Who it is for

Training designers, workshop facilitators and quality teams reviewing work artifacts against a maintained rubric. Suitable for preparing a structured discussion or checking the consistency of supplied ratings. It does not replace qualified reviewers or authorize educational, employment, credit or other consequential decisions about people.

Typical uses

  • Review each artifact against every supplied criterion and its existing performance levels.
  • Compare AI suggestions with independently supplied human ratings without silently replacing either.
  • Calculate a weighted normalized human score with an explicit scale and missing-rating policy.
  • Keep original quotations, rubric version, level descriptors and unresolved evidence gaps together.

Example

Input

Rubric criteria CSV: id,name,weight,definition C1,Traceability,2,Claims are linked to identifiable source records. C2,Clarity,1,The method and its limitations are stated explicitly. Performance levels CSV: id,criterion,points,descriptor L1,C1,0,No identifiable supporting source is provided. L2,C1,2,The main claim cites an identifiable source record. L3,C2,0,The method or its limitations are absent. L4,C2,2,The method and at least one limitation are explicit. Artifact evidence CSV: id,name,text E1,Workshop report,The conclusion uses record DATA-7. The method compares two samples and cannot establish causality. Supplied human ratings CSV: id,evidence,criterion,level H1,E1,C1,L2 Confirm supplied definitions: Confirmed for this review Confirmed definition version: RUBRIC-1 Quotation detail: Detailed Result order: Source order Criterion weighting: Supplied criterion weights Missing human-rating policy: Show incomplete score Aggregate score scale: 100 Score decimal places: 2 Displayed performance-level order: Ascending points

Result

The workshop report is reviewed against traceability and clarity. The supplied traceability rating contributes to the human score; the missing clarity rating keeps the aggregate incomplete under the default policy. A plausible AI clarity suggestion remains separate and cannot fill that missing human rating.

A fixed example, shortened. Real results are AI-generated: please check them.

Price and limits

Price
usually 1–3 credits
Available with
Free, With package
Input limit
One page with up to 80,000 characters of HTML

Privacy

Processed on our own servers in the EU (Poland). Inputs and results are deleted after 7 days (Free) or 30 days (packages), or right after you see the result if you choose not to store them. Images are deleted right after processing.

In the API

Tool name: assessment_rubrics. Create a job with POST /api/v1/jobs and fetch the result with GET /api/v1/jobs/{id}. You find your API key in your account.

curl -X POST https://do4u.eu/api/v1/jobs \
  -H "Authorization: Bearer d4u_…" -H "Content-Type: application/json" \
  -d '{"tool": "assessment_rubrics", "url": "https://example.com/"}'

Full API documentation →

Questions and answers

Can the model invent a performance level or assign a final grade?

No. A proposed level must be an existing ID belonging to the exact criterion. Its quotation must occur unchanged in the artifact text. The model may leave a level unresolved. A structurally valid quotation still requires semantic review; it does not prove that the proposed level is correct.

How are weights and different point scales combined?

Each supplied human rating is divided by that criterion's maximum supplied points. The normalized values are combined using the selected supplied or equal weights and multiplied by the configured aggregate scale. The calculation uses exact rational arithmetic and rounds half up only for display. This arithmetic does not establish a valid assessment method.

What happens to missing human ratings?

Incomplete withholds the aggregate when any criterion lacks a human rating. Zero explicitly counts missing ratings as zero while preserving warnings. AI suggestions never enter either calculation. Supplied ratings are not identity-verified approvals.

Is my text stored?

Only as long as needed: 7 days on Free, 30 days with a package – or not at all if you choose not to store it. Processing happens on our own servers in the EU (Poland).

Can I use the result right away?

Results are AI-generated. Please read them before you publish or send them.