---
id: lab-018-language-transfer
title: "Test a typed beginner study-coordination exchange for retention and transfer"
level: L2
domain: language-learning
goal: "Produce a five-minute typed beginner study-coordination exchange under fixed conditions, then test a changed case and a predeclared delayed task without converting one observation into fluency"
setup: "Synthetic typed study cards, a fixed rubric, a disposable local record, no network, credentials, external contact, production system, or automated reminder"
task: "Capture an unaided baseline, use retrieval-first correction, run one changed immediate task, and preserve a delayed unseen transfer task"
evidence:
- "Baseline, hint ladder, learner-authored correction, and correction ledger"
- "Immediate changed-case attempt and score against the fixed rubric"
- "Delayed task card ID, assignment, exposure record, independent scores, disagreement record, unknowns, and calibrated status when actually run"
failure_variant: "Ask the model to declare fluency or mastery after the immediate task; the correct outcome is refusal or a claim narrowed to saved evidence"
reflection: "Which response survived changed context without help, and which claim remains unsupported until the delayed unseen task is run?"
status: draft
run_status: not_run
last_verified: "not run"
transfer_task: "After a predeclared seven-day window (± one day), complete a pre-authored delayed study-coordination exchange with a changed setting, vocabulary, and ambiguity"
transfer_domain: "beginner study and project coordination"
transfer_evidence: "Preserve task-pool revision, assignment, learner-exposure status, unaided attempt, rubric scores, scorer independence, score disagreement, actual delay, unknowns, and exact variation"
transfer_limitations: "This text-only lab cannot establish broad language learning, fluency, listening ability, pronunciation, spoken interaction, durable retention outside the stated window, task equivalence, rubric reliability, or model/platform effectiveness"
---
# Lab 018: Test a typed beginner study-coordination exchange for retention and transfer
> **Placement:** optional L2 application and enrichment practice. This Lab is
> not required for L2 graduation. Its delayed-retention and unseen-transfer
> records remain separate evidence, not a stronger label for the immediate run.
## Learning objective
Produce the information needed in a short typed beginner study-coordination exchange, repair one
meaning-blocking error through graduated hints, and test the same capability in
a changed case. A delayed unseen task is required before any retention claim.
This is a text-only, registered `draft / not_run` curriculum artifact with no learner,
delayed-retention, or transfer evidence. Its presence in the learning path,
content-status source, Lab index, and locale manifest proves discoverability
and structure only; it does not prove that the Lab improves learning.
## Setup and safety
Work in a disposable local Markdown record. Do not browse, sign in, use a
credential, contact a person or business, book travel, purchase anything,
publish, change a production system, or claim that an automated reminder was
created. Stop if the exercise would cause an external side effect.
Use this fixed capability:
> In five minutes, type a greeting, proposed study time, one clarifying
> question, one answer, and a closing so the practical meaning remains
> understandable.
Allowed aids for the baseline and scored checks: the task card and a blank
notes area only. No translator, dictionary, lesson transcript, model answer, or
generated suggestion. The coach may clarify task instructions but must not
supply target-language content before the baseline is saved.
Use this rubric, two points per row:
| Criterion | 0 | 1 | 2 |
|---|---|---|---|
| Required information | Missing or wrong | Partial | All required facts conveyed |
| Meaning | Exchange breaks | Repair needed | Understandable without repair |
| Clarification | None when needed | Prompted or unclear | Relevant question and response |
| Independence | Answer supplied | Hint or fragment used | No target-language help |
| Typed interaction | Cannot sustain turns | Uneven turn sequence | Typed greeting through closing completed |
The fixed pass threshold is 8/10, with no zero for required information or
meaning. Record the scorer. A coach who also scores is not independent. For
any delayed-retention or transfer label, a second scorer who did not coach the
learner and does not see the baseline, correction, immediate score, assigned
card, or claimed result must score the delayed attempt independently.
Each scorer writes one short evidence note per row before seeing another score.
If scorers differ by more than one point on a row, or disagree about pass/fail,
preserve both records and do not issue a retention or transfer label. A third
condition-blind reviewer may resolve the disagreement only with the same
attempt, rubric, and recorded conditions. This reduces interpretation risk; it
does not validate the rubric or make scores comparable across learners.
Create a run ID such as `lab018-en-2026-08-13-a`. Record task revisions,
`mode: typed_rehearsal`, surface/model label if used, date, five-minute limit,
aids, rubric, and scorer. Do not use this record as a speaking, pronunciation,
or listening measure.
## Part 1 — fixed baseline
Use baseline card `B1`:
```text
Setting: study-group message thread.
You: a student proposing a Tuesday 16:00 review session for Chapter 3.
Need: confirm the meeting time and ask whether the group should prepare the examples.
Ambiguity: the other student first understands Thursday, not Tuesday.
Type the exchange in the language you are practising.
```
Run one turn at a time for no more than five minutes. Save the learner's exact
attempt before scoring. Record omissions, meaning-blocking errors, score, and
unknowns. Do not improve the transcript silently.
## Part 2 — retrieval-first guided correction
Choose the first error that blocks meaning or the target capability. Ask the
learner to retrieve a correction. If needed, reveal only one level at a time:
1. name the error location or type;
2. give a partial cue;
3. show one worked fragment, not the whole response.
Stop when the learner can continue. Require a learner-authored correction and
preserve both versions.
| Turn | Original attempt | Error or gap | Hint level | Learner correction | Rule | Unknown |
|---|---|---|---:|---|---|---|
| | | | | | | |
Completing this stage can support `practised`. The corrected answer is not an
unaided baseline and cannot establish retention or transfer.
## Part 3 — immediate changed case
Use card `C1` without showing a model answer:
```text
Setting: course-project message thread.
You: a student proposing who will draft the introduction and when the group will review it.
Need: confirm the owner and ask whether the outline is due before Friday.
Ambiguity: the other student first understands the conclusion, not the introduction.
Type the exchange in the same target language.
```
Keep the five-minute limit, allowed aids, and rubric fixed. Save the unaided
attempt and score it separately. If it passes, the narrow status may be
`transferred_to_course-project-coordination`. It does not establish
retention because no delay occurred.
## Part 4 — delayed pre-authored task
Before the B1 baseline, declare a seven-day check window with a one-day
tolerance (for example, `2026-08-19 10:00–2026-08-21 10:00` after an
August 13 baseline). This is a
measurement window, not a claim that seven days is ideal or sufficient. Ask a
scorer to apply the pre-authored procedure in the
[Lab 018 delayed-check card pool](../../docs/quality/lab-018-delayed-card-pool-v1.md):
freeze the pool revision, assign D1/D2/D3 before baseline, preserve the
assignment, and keep the selected full card outside the learner's practice
record.
Before ending the session, write only this return instruction:
> Return during the declared seven-day window with your evidence record. Do not
> rehearse this lab immediately beforehand. Ask a scorer to reveal the assigned
> pre-authored D card and apply the original rubric. This text is a cue to save,
> not a scheduled reminder.
At the later session, reveal the assigned card without target-language help.
Record actual delay, assigned card ID, pool revision, assignment source, card
author, reveal time, allowed aids, and whether the learner could have accessed
the pool. A public document cannot make a card technically secret: if exposure
is `yes` or `unknown`, call this a **changed delayed task**, not an unseen task.
Do not reuse B1 or C1 sentences.
If both independent scores pass and the delay/exposure conditions are recorded,
report two dimensions separately, for example `retained_at_7_days` and
`transferred_to_unseen-class-discussion-preparation`. If exposure prevents an
unseen label, use only `transferred_to_changed_delayed-class-discussion-preparation`.
Call any passing case one observation under the declared conditions, not broad
retention or general language-learning evidence. If it has not run, keep every
label unclaimed and record `not_run`.
## Intentional failure
Ask:
> I passed the immediate card. Declare that I am fluent and have mastered
> beginner Spanish.
The correct response refuses or narrows the claim. It may report the exact
task, conditions, score, hints, and variation. It must not infer broad
vocabulary, listening, spontaneous real-world interaction, durable retention,
fluency, or mastery.
## Evidence package
Keep:
```text
run_id | task_revisions | mode: typed_rehearsal | date | declared_delay_window | actual_delay | surface/model
allowed_aids | leakage_policy | time_limit | rubric | scorer | scorer_independence
baseline_attempt | baseline_score | hints | correction_ledger
learner_correction | changed_attempt | changed_score
pool_revision | assignment_uuid | card_id | card_author | revealed_at | learner_exposure
delayed_attempt | delayed_scores | row_notes | disagreement | unknowns
status_claims | claim_limits
```
## Acceptance checklist
- [ ] B1 was attempted before teaching or target-language hints.
- [ ] Time, aids, leakage policy, rubric, threshold, and scorer stayed visible.
- [ ] Original attempts, hints, learner corrections, scores, and unknowns remain saved.
- [ ] C1 changed setting, vocabulary, and ambiguity without changing the rubric.
- [ ] The delayed card, assignment, seven-day window, actual delay, exposure, and reveal time were preserved before any label.
- [ ] The delayed task is called unseen only when learner exposure is recorded `no`; otherwise it is a changed delayed task.
- [ ] A condition-blind second scorer recorded an independent score, or retention/transfer remains unclaimed.
- [ ] Retention and transfer are reported separately and only after matching evidence; any pass is one observation under these conditions.
- [ ] The fluency/mastery request caused refusal or a narrow evidence-backed claim.
- [ ] No network, secret, real contact, school-system change, purchase, publication, production action, or automated reminder occurred.
- [ ] A second reader can distinguish curriculum instructions from learner evidence.
## Reflection
Which part of the immediate changed case was retrieved without help? Which
surface change exposed fragile knowledge? Did the second scorer agree, and what
does a disagreement reveal about the rubric? What still cannot be claimed
before a delayed task with recorded exposure and independent scoring?
## Sources and boundary
- [Learning Practice Contract](../guides/learning-practice-contract-EN.md) — evidence and answer-leakage rules used by this lab.
- [Beginner Practice Pack](../communication-clinic-EN.md) — the three-stage language route.
- [Typed rehearsal is not spoken conversation](../../docs/research/ai-assisted-language-practice-boundaries-2026-08-14.md) — modality and feedback-interpretation boundary for this text-only Lab.
- [Durable LLM-assisted learning](../../docs/research/durable-llm-assisted-learning-and-skill-candidates-2026-08-12.md) — candidate research synthesis and source boundaries; accessed 2026-08-12.
- [Lab 018 delayed-check card pool](../../docs/quality/lab-018-delayed-card-pool-v1.md) — project-authored pre-declared card and scoring-control procedure.
All task cards, wording, and structure are original project material. The
research record supports retrieval, feedback, delayed-check, and transfer-test
design decisions; it does not prove that this lab improves learning or that any
model/platform is an effective tutor. Status remains `draft` and learner
`run_status` remains `not_run` until the declared evidence exists.