Demonstration learning programme

This implementation adds 12 paths and 112 learning activities to the combined Reader. The largest path has 20 steps. It covers the 40 supplied question occurrences (39 unique questions), using the eight projected personas already present in the corpus. These are independent teaching proposals, not validated DWP training, benefits advice or a competency qualification.

The authoritative curriculum is programme.yamlld. The combined builder generates real learning/pNN/sNN records, bounded search and locator entries, semantic identities and the Reader's v2 teaching overlay. Teaching references link existing evidence without creating legal assertions. The pilot, full-DMG release and authored logical units remain unchanged. The dependent logical-context Reader is rebound and its evaluation replayed when its combined-descriptor input changes.

This branch builds on DWP 311f7ff9, including the parallel logical-evidence-unit work. Source-page anchors remain valid. The curriculum explicitly distinguishes page locations from complete logical units, governing headings, cross-page continuations, attached examples, qualifications and unresolved references. No old whole-page evidence requirement is silently satisfied by a fragment. The logical-unit Reader remains a separate corpus, linked from the repository's logical-unit guide.

Path design, lesson specifications and question coverage retain the original design baseline. They are historical design artefacts; the authored YAML-LD and generated Reader are the current implementation. The earlier design's statements about absent lesson records and absent assessment software describe that earlier baseline.

Demonstrate in ten minutes

1. Open the generated combined descriptor in the matching Explorer build.
2. Expand Read evidence before trusting an answer. Open Baseline and role, then inspect its supporting persona record. Use Back to return to the lesson.
3. Open Distinguish the layers. Explain why a source page, a complete logical unit and an interpretation have different roles. The logical-unit guide supplies the separate exact-span demonstration; do not claim all combined records have been replaced by logical units.
4. Use the household path and its changed-scenario exercise to show how a plausible answer can omit a controlling qualification.
5. Open Evidence and assessment. Show that an artefact can be submitted for the foundation path but dependent assessment requires an assessed pass.
6. Export a fictional foundation submission. A facilitator reviews it, applies the rubric and returns a signed decision using Explorer's learning_assessor.mjs. Import that decision; demonstrate that the dependent assessment unlocks.
7. Reload: the verified decision persists. Explain that a new programme version or bundle snapshot needs a new review. Export the journal as a backup.
8. Finish with the fair-comparison path. Show measured results only where a retained experiment exists. The form and calculator paths teach design and assurance; they are not a delivered CASA claim service or an award calculator.

For a short meeting, use this excerpt. Do not imply that attending it completes 112 activities or awards a pass for omitted assessments.

Assessment and operator setup

The authored roster contains the local demonstration facilitator's public key. Its private key stays outside the repository; it is not included in a shared bundle. Recipients can browse and prepare submissions without it. To run their own assessments, they generate their own key, replace the authored public roster and rebuild, producing a new snapshot. A signature establishes configured key ownership, not specialist competence. Legal interpretation requires a qualified reviewer; no automatic grading or official credential is claimed.

The rubric has five scores (0–2): traceability, scope, reasoning, counterexample and communication. Pass requires 8/10, full marks for traceability, scope and counterexample, and no critical failure. Browser journals retain attempts and verified decisions. They are local teaching aids, not a managed training record system. Use fictional examples and export a backup before clearing browser data.

Reproduce

uv sync --locked
uv run --locked python scripts/build_combined_reader.py
uv run --locked python scripts/build_combined_reader.py --check
uv run --locked python -m unittest discover -s scripts -p test_learning_paths.py

The generator rejects unknown source/persona routes, incomplete question coverage, changed question-registry bytes, duplicate lesson identities, prerequisite cycles and text exceeding Reader limits. Assessor decisions bind the learner, exact artefact, programme version and generated bundle snapshot.