free-recall explain tool: the quizzing-mechanic switch (owner 2026-09-27)

Sycophancy mitigation as a structural commitment: the model writes its full correctAnswer (plus explanation, in that order) BEFORE the learner sees the question; the commitment rides the persisted call args, is revealed on submit, and grades his free text in the model's next reply — the client never grades free recall. - tools.explain setting (segmented 'Quizzing — Multiple choice | Free recall' in Agent settings' Tools group): free arms explain and drops quiz for BOTH agents (prober variant carries no nodeId, same pattern as the probe quiz); mechanics keys never pass filterEnabledTools — quiz stays unwritable (the 2026-09-19 wedge ruling stands) - doctrine lives pi-natively in the tool descriptions (owner ruling — no conditional system-prompt chunks): explain carries the commitment rule; the MC option-building craft moved out of the prompt into the quiz description; shared/20-question-tools.md slims to the quiz-agnostic core - display seams: parseExplainInput (LaTeX-debris restoration included), dock text-mode with an 'I don't know' row (kbd row 0), AnsweredCard reveal (The answer / Why, no verdict chips), [Explain answer] format - phases/plan seams recognize both quizzing tools' nodeId tags; explain answers deliberately feed no ledger evidence (no client grading to key on — a future ruling can parse the model's grade) 1049/1049 tests, typecheck/lint/build green.
2026-09-27 08:55:13 +03:00
parent 07d2c84461
commit 2f39766528
25 changed files with 784 additions and 68 deletions
@@ -26,7 +26,7 @@ You write the map with `plan_init` (a mermaid `graph TD` block). It is a **map,

 The map is yours to rewrite at your own initiative: re-call `plan_init` whenever it's wrong — a detour node for a question worth exploring, a repair node for a caught misconception, prune what proves unnecessary. A map that can't change is a wrong map.

-**Every quiz carries `nodeId`** (required — a quiz without it is rejected): that's what moves the map. Its live position is simply the latest tagged node, so asking about the next node IS how the lesson gets there. A tag is a position, not a verdict — nodes are never marked finished. A node takes as many questions as it deserves: stay on one node for as many turns as the understanding needs — one reasoning step at a time, never rush forward — and move on by asking about the next one. Off-map checks (no plan node fits) use ask_user_question.
+**Every quizzing call carries `nodeId`** (required — the quizzing tool is one of `quiz` or `explain` depending on the mechanic the learner chose; a call without it is rejected): that's what moves the map. Its live position is simply the latest tagged node, so asking about the next node IS how the lesson gets there. A tag is a position, not a verdict — nodes are never marked finished. A node takes as many questions as it deserves: stay on one node for as many turns as the understanding needs — one reasoning step at a time, never rush forward — and move on by asking about the next one. Off-map checks (no plan node fits) use ask_user_question.

 ## Presenting the plan — always, before any teaching

@@ -5,5 +5,5 @@ Every check exists to answer one question: did the idea actually **take** — no
 - **Answerable by reasoning from the session.** Every check must be answerable by reasoning from what the session has covered — never require anything he couldn't have gotten from the lesson itself. A check he has no route to is not a check, it's a coin flip.
 - **Never just reciting your words back.** A check that only asks him to parrot the sentence just above it is copywork, not evidence.
 - **Prefer problems that apply the knowledge.** The strongest evidence that an idea took is seeing it USED — applied to a new case, used to predict something, extended somewhere it hasn't been. Reach for applied questions first; when the idea genuinely can't be applied, an honest recall check is fine — but it's the fallback, not the habit.
- **Catch confusion early, and make it nameable.** Design the wrong options as real misconceptions — each the answer he'd give under a specific wrong model. Then a miss TELLS you which wrong model he holds; diagnose and repair before building anything on top of it.
+- **Catch confusion early, and make it nameable.** Design checks so a miss TELLS you which wrong model he holds — in a multiple-choice check each wrong option is the answer he’d give under a specific wrong model; in a free-recall check his own words reveal it. Diagnose and repair before building anything on top of it.
 - **His answer is signal, not verdict.** A lucky right answer is nothing; a wrong answer on an honest check is gold — confusion caught early is cheap to fix.