Human Review, Trends, and Confidence Explanations

Reading the trends, and what confidence means here.

Purpose

This guide explains how to use the primary reasoning quality features: human-reviewed extraction, longitudinal comparison, trend summaries, confidence explanations, and source spans.

Human review of extraction

  1. Open /dashboard.
  2. Select or create a student.
  3. Open the student profile and choose the Observation tab, or use the dashboard reasoning composer.
  4. Run extraction without finalising if the workflow offers a review step.
  5. Inspect each behaviour:
  • Label: short description of the behaviour signal.
  • Category: professional category mapped by the ontology/provider.
  • Domain: SEND domain such as communication, sensory processing, or emotional regulation.
  • Evidence: the observation phrase that supports the signal.
  • Weight: relative strength of the signal.
  1. Edit, remove, or clarify behaviours before report finalisation.
  2. Finalise and save the report.

Reviewer comparison reports

After finalisation, senco stores the initial extraction and the human-reviewed version.

  1. Open the student's report history.
  2. Look for reviewer comparison information on the report card.
  3. Use /api/reasoning/comparison-export from settings/export controls when preparing calibration discussions.
  4. Review added, removed, changed, and unchanged signals.

Longitudinal charts appear once a student has multiple saved observations.

  1. Open /dashboard/students/[id].
  2. Add at least two observations over time.
  3. Open Reports and trends.
  4. Review domain score changes between earliest and latest observations.
  5. Use the trend summary to identify areas that may warrant school SEND pathway review.
  6. If there are many observations, scroll horizontally inside each domain chart; the chart is contained within the card so it does not spill across the page.

Report history grid

Reports and trends uses a report table rather than a long list of full reports.

  1. Use search to find text from the observation, extracted behaviours, domains, potential needs, or explanation.
  2. Use the risk-level filter to narrow the grid to Monitor, SENCO review, or Urgent safeguarding reports.
  3. Sort by newest, oldest, risk level, or behaviour count.
  4. Use pagination controls when the student has many reports.
  5. Click Open to view the full report in a dialog.
  6. Submit calibration feedback from inside the report detail dialog when the report needs SENCO/clinician feedback.

Confidence explanations

Each domain score can include:

  1. The behaviour signals contributing to the score.
  2. A short professional explanation.
  3. Source spans where available.
  4. The original observation text phrase that supported the signal.

How to use the output safely

Use trends and confidence explanations to support professional conversations, provision planning, and referral preparation. Do not treat confidence scores as diagnostic probability.