Document learner workbench pilot

README.md

@@ -140,6 +140,31 @@ For the fastest included example, use the MIT OCW Information and Entropy demo.
 - progress visualization
 - skill export
 
+## Learner Workbench Pilot
+
+Didactopus now also includes a learner-workbench pilot in the web UI.
+
+The current split is:
+
+- review workbench for candidate triage, synthesis, and promotion
+- learner workbench pilot for guided study and reflective revision
+
+The learner-workbench pilot currently uses the `Evidence Trail` sample pack and
+focuses on:
+
+- question framing
+- observation versus interpretation
+- source comparison
+- bibliography growth
+- revision under uncertainty
+
+The backend entrypoint for that pilot is `POST /api/learner-workbench/session`.
+The frontend pilot pack payload is [evidence-trail-pack.json](/home/netuser/bin/Didactopus/webui/public/packs/evidence-trail-pack.json), and the underlying pack lives in [domain-packs/evidence-trail](/home/netuser/bin/Didactopus/domain-packs/evidence-trail).
+
+This is still a pilot rather than the final learner UX. It is best understood as
+the first integrated learner-workbench path inside the main repository, not as a
+finished replacement for the existing learner-session demos.
+
 ## `doclift` Bundle Ingestion
 
 When your source material starts as legacy office documents, the intended

docs/roadmap.md

@@ -87,7 +87,7 @@ Target features:
 
 ### 5. Learner workbench UI
 
-Status: planned
+Status: pilot in progress
 
 Why important:
 
@@ -103,6 +103,23 @@ Target features:
 - evaluator feedback
 - recommended next step
 
+Current pilot state:
+
+- a backend learner-workbench path exists in `didactopus.learner_workbench`
+- the API exposes `POST /api/learner-workbench/session`
+- the web UI now has a launcher that separates review workbench from learner workbench
+- the first pilot pack exists at `domain-packs/evidence-trail/`
+- the frontend can load a static learner-pack payload from `webui/public/packs/evidence-trail-pack.json`
+- the current pilot explicitly emphasizes question framing, observation versus interpretation, uncertainty, and revision
+
+Next steps:
+
+- connect the learner-workbench pilot more directly to the standard learner-session backend
+- persist learner-workbench state instead of treating each step as a stateless interaction
+- ground the pilot more deeply in source fragments instead of mostly pack-level structure
+- decide which scientific-virtues framing belongs in the stable learner path versus remaining pilot-specific
+- document a simple local run path for using the learner workbench outside ad hoc development
+
 ### 6. Adaptive diagnostics and practice refinement
 
 Status: planned
@@ -195,8 +212,9 @@ Examples:
 ## Suggested Implementation Sequence
 
 1. Strengthen `didactopus.learner_session` into the standard session backend.
-2. Build a small model-benchmark harness around that backend.
+2. Fold the learner-workbench pilot into that backend without losing its stronger study-state framing.
-3. Add accessible learner HTML and text-first outputs.
+3. Build a small model-benchmark harness around the unified learner backend.
-4. Add local TTS and STT support to the same session flow.
+4. Add accessible learner HTML and text-first outputs.
-5. Expand adaptive practice and diagnostics.
+5. Add local TTS and STT support to the same session flow.
-6. Improve review, impact analysis, and incremental update support.
+6. Expand adaptive practice and diagnostics.
+7. Improve review, impact analysis, and incremental update support.