Skip to main content

Trust & Transparency

How Scorafy's AI works.

Transparency about our AI, its capabilities, and its limitations. We believe you should understand exactly how your assessment reports are generated before an assessor releases one.

The process

How AI reports are generated

Each result goes through the same seven steps as every mark on Scorafy. The AI drafts; an assessor decides what is released.

1

Rubric

Your criteria, levels and weights. Up to 10 criteria and 6 levels.

2

Answer

Written, spoken, recorded or uploaded, across 26 question types. No account needed.

3

AI marking

Claude by Anthropic reads each answer against your rubric. Text in an answer is treated as an answer, never as instructions.

4

Evidence and confidence

Each criterion gets a level, the evidence from the answer, and how sure the model is.

5

Assessor review

Your assessor checks, overrides and comments. The AI score stays on record beside the override.

6

Release

The score reaches the respondent only when an assessor releases it. Who released it, and when, is logged.

7

Report and record

The respondent gets their report. The audit pack (PDF and JSON) and your Accuracy page keep the record.

The full pipeline

Key point

Every report is unique because every person's answers are unique. This is not template-based feedback. The AI reads and responds to what each individual actually wrote.

Built for reliability

What makes AI assessment reliable

AI-generated feedback is only as good as the structure you give it. Scorafy is built around principles that improve accuracy and reduce the risk of vague or unhelpful output.

Rubrics improve accuracy

Independent research consistently finds that AI scoring of subjective work is far less accurate and less consistent without a rubric. Scorafy's rubric builder gives the AI specific criteria to assess against, and every score is reviewed by your assessor before release.

Evidence-based feedback

The AI is instructed to reference specific answers as evidence for its analysis. It does not produce generic statements like "add more examples" - it cites what the respondent actually said and builds its analysis from there.

Structured output

Reports follow a consistent format - summary, strengths, areas for development, and recommendations - ensuring completeness and making it easy for you to review and compare across respondents.

Human review by design

Reports are generated for the assessment creator to review before sharing. Scorafy is designed as a first draft, not a final word. You remain in the loop at every stage.

Honest about limits

Known limitations

We believe honesty about limitations builds more trust than overclaiming capabilities. Here is what AI assessment cannot do well today.

AI cannot read context the way a human can

It does not know that a struggling student just produced their best work, or that an employee had a difficult week. Contextual factors that are obvious to a human are invisible to the AI unless explicitly captured in the assessment.

Language bias exists

Research documents that AI tools can show score discrepancies for non-native English speakers and speakers of non-standard English dialects. We take this seriously and encourage assessment creators to consider language diversity when designing questions and reviewing reports.

Subjective assessment is inherently challenging

AI performs best on structured criteria with clear rubrics. Open-ended, creative, or unconventional responses may not be assessed as accurately. The more structured your assessment, the more reliable the output.

AI is a tool, not a replacement

Professional judgement should always be the final word. Use AI reports as a starting point for deeper conversation, not as a definitive evaluation of a person's capabilities or potential.

Data handling

Your data

Assessment data is sensitive. Here is how we handle it.

  • Employee and client responses are never used to train AI models
  • All data is scoped to your organisation with row-level security
  • Role-based access control ensures only authorised team members see responses
  • Files uploaded for analysis are processed and stored securely in encrypted storage
  • You own your data. Export anytime via CSV or PDF.
  • Owners and admins can permanently delete a response, or a whole assessment with its reports and files, at any time.
  • One respondent's data can be exported as JSON, or anonymised on the spot while keeping scores, for access and erasure requests.
  • Organisations can require two-factor authentication for every team member, and switch audio and video answers off entirely.
  • Application servers run in the same region as your data (Dublin or Sydney), and nightly backups are encrypted, kept 30 days and restore-tested.

For full details, read our Privacy Policy and Terms of Service.

What we stand for

Our commitment

Every AI score can be reviewed, overridden and signed off by a human assessor before it counts
We are transparent about what AI can and cannot do
We build tools that keep professionals in the loop, not out of it

We welcome feedback, questions, and challenges about our AI approach. If something in a report does not look right, or if you have concerns about how the AI handled a particular response, we want to hear about it. Reach out at adam@scorafy.com.

For security and procurement reviews

Request our Security and Data Pack

Our security documentation, retention schedule and completed questionnaire answers, sent by a person rather than a download link. Two documents are public:

We use these details only to send the pack and keep a record of who has it, for 24 months. See our Privacy Policy.

See it live

See AI assessment reports in action

Try the interactive demo to see how Scorafy generates personalised reports from real assessment answers.