blind

Output language

Read INTERFACE_LANG from .course-meta (default en). All user-facing prose — chat output, clarification questions, and narrative sections of any MD file you write — must be in that language. Keep in English regardless: file paths, slash command names, pattern IDs (P1, P2…), YAML keys (pattern:, error_type:, problem_id:, source:, date:, summary:), LaTeX, code, and the literal section anchors downstream tools regex on (## One-line verdict, ## Page N, # Vision-OCR transcription).

Load skills/exam-drill/SKILL.md. Read course-index/patterns.md.

Target: $ARGUMENTS

Procedure:

1. Load problem statement ONLY from converted/homework/<n>.md or converted/textbook/<ch>.md (for textbook examples). Do NOT open the solution yet.

   Confirm a reference solution exists first — before spending the user's effort. A blind drill grades the user's strategy against the known solution, so without one there is nothing to check. Verify (do NOT read its contents yet) that the matching solution is ingested: converted/solutions/<n>_sol.md for a homework problem (e.g. 3.3 → converted/solutions/hw3_sol.md), or the worked-example section inside converted/textbook/<ch>.md for a textbook example. If it is missing, stop here without requesting a strategy and tell the user (in $INTERFACE_LANG): no reference solution for this problem is ingested, so a blind strategy-check can't be verified. Offer the concrete next steps — add the solution to materials/solutions/ and run /paideia:ingest, or pick a problem that has a solution (list which converted/solutions/*_sol.md exist), or drill it instead with /paideia:twin <id> or /paideia:quiz (neither needs a pre-existing per-problem solution). Never fabricate a solution to grade against.

2. Present the problem verbatim to the user.

3. Request strategy (3–5 lines in $INTERFACE_LANG prose, no math typing). The 3 axes to ask about:

   Strategy only — no equations needed.
   1) Which pattern(s) will you use? (Pk number from course-index/patterns.md)
   2) Which variables held fixed; which expanded?
   3) What form do you expect the final answer to take?
   

   Render those three axes in $INTERFACE_LANG.

4. Wait for response. Do NOT proceed until the user answers.

5. Load solution from converted/solutions/<n>_sol.md (or the worked-example section of converted/textbook/<ch>.md for an example) — the same file confirmed present in Step 1. Compare 3 axes:

   a. Pattern identification — correct Pk(s)? b. Variable choice — correct hold-fixed set? c. End-form prediction — matches actual answer structure?

6. Feedback protocol:

   • ✅ all three → confirm, then copy the relevant part of the solution into derivations/blind-<id>.md for permanent reference
   • ❌ on any axis → point out specifically which axis failed, WITHOUT revealing correct answer. Ask for revision.
   • After 2 failed attempts on same axis → give a one-line hint referencing the relevant pattern name.

7. Log errors if user needed revision. Use the canonical schema from skills/answer-processing/SKILL.md Step 6 — same keys /grade writes, so statusline and weakmap see /blind errors too. Write through the deterministic writer (replaces any prior entries for this drill's source, so re-running /blind on the same problem never double-counts):

   python3 "${CLAUDE_PLUGIN_ROOT}/scripts/log_tool.py" append --source="blind/<id>" <<'YAML'
   - problem_id: <id>
     pattern: <Pk>
     error_type: pattern-missed | wrong-variable | wrong-end-form
     summary: "<1 line>"
     source: blind/<id>
     date: <ISO>
   YAML
   

   Map strategy axis → error_type: pattern axis → pattern-missed, variable axis → wrong-variable, end-form axis → wrong-end-form. Never hand-edit errors/log.md appends.

8. Close (in $INTERFACE_LANG): "To check retention on the same type, do one variant via /twin <id>."