andresnator Bidirectional Translation
Activation
Run delayed native-to-target reconstruction of a supplied bilingual text. This method accepts an earlier-session text directly; no dialogue directory, named agent, or sibling is required.
Inputs
Target/native text and languages, passive date, trustworthy current/due date, and optional learner attempt. Optional: explicit output destination and permission to save this drill record. Without a destination, return it inline. Never infer a path.
Method
Check that delayed practice is eligible: the initial policy is three days after passive exposure, configurable by the caller. If it is too early or the date is unknown, offer comprehension practice without claiming a completed delayed attempt; retain the future drill.
Show only native-language text before the attempt. Let the learner translate from memory, then reveal the original and compare. Natural equivalents count; different wording alone is not a mistake. Ask what changed in meaning before explaining.
Classify material differences as word choice, structure/order, missing chunk, or grammar pattern. Give specific feedback and invite another attempt. Mark completed only when the learner preserves essential meaning in natural target-language production and demonstrates comprehension; unresolved material differences mean needs-another-attempt. Keep feedback conversational, not exam-like.
Return missing phrases as vocabulary candidates and structural gaps as optional retention proposals. Nothing is registered, exported, scheduled, or demoted automatically. A proposed grammar card still needs a post-teaching preview and actual learner selection. Match synthetic categories to supplied gaps without replaying an occurrence as new evidence.
Output
Return actual learner practice evidence, material differences, proposals, and status using assets/bdt-session-template.md when useful. Dates come from the caller/calculator; unscheduled follow-up remains explicit when unavailable. Save only to an explicitly supplied authorized destination; otherwise output inline. Preserve learner-produced evidence only inside this explicitly started drill, never import private correction history.