Minarai · Industrial skill transfer

From observable evidence and existing expert knowledge to reviewed interpretation—and only then to learning

New evidence can be compared with prior expert episodes, approved documents, and earlier extractions. Current-video evidence and contextual knowledge always retain separate provenance.

Evidence, expertise elicitation, judgment episode, record, and learning process

Preparing the evidence layer…
Minarai evidence-to-learning process Five-stage diagram. Work video produces verified speech and timestamped observations. The AI asks about specific moments and retrieves separately cited expert corpora, documents, and prior extractions. Expert review populates a connected judgment episode. Verified episodes may later support a separate practice and feedback cycle. 1 Current evidence What can be verified? Work video audio + frames Verified speech timestamped Visible / audible observation Derived cue motion · track · sound 2 Video-cued interview Ask about the moment AI presents a moment with a grounded question Specific expert question moment + cue + uncertainty Response source evidence Open question retained SEPARATELY CITED INPUT Existing knowledge context Reviewed expert episodes accepted corpus + corrections Documents and standards version + applicability retained Prior extracted graphs source + review state retained 3 Expert review and judgment Accept · correct · reject · leave open Interpretation candidate evidence + context stay separate Expert resolves applicability? REJECT / OPEN ACCEPT / CORRECT Rejected or left open question remains visible Reviewed judgment episode situation · cue · interpretation · decision criterion · action · outcome · adjustment 4 Reviewable record / draft SOP A projection—not a reasoning chain DUAL PROVENANCE · SEPARATE SOURCE IDS Source-backed fields Context · Signals · Criteria · Action Reasoning · Exception Open questions missing expertise stays explicit 5 Learning design and practice loop Clockwise · future capability · human-authorized feedback 1 · Observe 2 · Predict 3 · Expertexplanation 4 · Practice 8 · Transferassessment 7 · Variation 6 · Reflection 5 · Feedback reviewed updates only current evidence context suggests

1 · Current evidence

Work video produces timestamped verified speech, visible or audible observations, and derived cues.

2 · Video-cued interview

The AI presents one moment and asks a specific question. A response becomes evidence; otherwise the question stays Open.

Existing knowledge context

Separately cited sources may suggest an interpretation, but do not become current-video evidence.

  • Reviewed expert episodes
  • Documents and standards
  • Prior extracted graphs

3 · Expert review and judgment

The expert accepts, corrects, rejects, or leaves the interpretation Open. Only reviewed relationships form a judgment episode.

4 · Reviewable record / draft SOP

Source-backed fields and Open questions are projected without collapsing current evidence and contextual knowledge IDs.

5 · Learning design and practice loop

A future, human-authorized cycle adds learner action, feedback, reflection, variation, and transfer assessment.

  1. 1 · Observe
  2. 2 · Predict
  3. 3 · Expert explanation
  4. 4 · Practice
  5. 5 · Feedback
  6. 6 · Reflection
  7. 7 · Variation
  8. 8 · Transfer assessment
Current-video evidence Existing knowledge context Expert-reviewed relationship Future learner activity Visual evidence proposes questions; existing knowledge proposes interpretations; expert review decides applicability.