# Spellbound Relics 1.1 — The Sleeping Garden

A focused continuation of the Old Star Tongue Magic Spell Game, built from the latest recovered package, `spellbound-relics-v1.0.4-enchanted-environment.zip`. Mei, Abby, Luma, the established runes, and the atmospheric world artwork are retained. The prior lesson is available at `lesson-v104.html`.

## What the survey supports

The one response, dated October 29, 2026, used four ordered categories: No, A little, Yes, A lot. Treat these as ordinal responses, not a validated scale or effectiveness score.

| Item | Response |
| --- | --- |
| Class enjoyment | A little |
| Game enjoyment | A little |
| Characters and pictures | Yes |
| Instructions understood | Yes |
| Arranging runes | A little |
| Activities at a good level | A little |
| Reward from working spells | A little |
| Interest in playing again | A little |
| Difficulty | Just right |
| Favorite part | “It was challenging” |
| Requested change | “I don’t know” |

This is a useful engagement signal from one learner. It cannot identify the cause, establish learning, or justify generalizing to other learners. The response predates the latest recovered environment package; its tested game version is unverified. The “good level” and “just right” answers also deserve an observational follow-up rather than forcing them into one interpretation.

## Working hypothesis and design response

Operational IND-GBL hypothesis for this pilot: repeated, meaningful language choices embedded in play, with perceptible consequences and graduated support, may support generalization of English patterns beyond the rune system. This is a hypothesis, not an established result. IND-GBL is retained as the user's framework label; no unverified expansion of the acronym is introduced.

The current source required exact guided spells, used visible grammatical labels in some learner feedback, and offered some modifiers whose semantic differences were not fully represented. The new pilot tests goal-based acceptance and a clearer relationship between language and world changes. It does not claim these caused the survey ratings.

| Design basis | Pilot implementation | Evidence to inspect |
| --- | --- | --- |
| IND-GBL hypothesis | Spells communicate wishes; multiple supported wishes change one continuing garden | Independent choices, revisions, transfer beyond familiar runes |
| UDL | Optional guide, placed-word meanings, keyboard/tap input, reduced motion, written and spoken messages, optional oral practice | Which supports help participation without excessive prompting |
| Systems Thinking | Ordered actions update persistent world state; later actions depend on that state; evidence feeds the next revision | Where attempts fail, which dependencies confuse, whether learners explore consequences |
| Cognitive Load Theory (CLT) | One visible goal, 6–8 rune types, gradual hints, no timer, optional models, stable character art | Support use, attempts before success, visible confusion, separate child-friendly effort question |

CAST's UDL Guidelines 3.0 support agency, joy and play, response options, actionable feedback, and transfer. CLT supports managing working-memory demands and appropriate guidance. Neither source by itself establishes that an unguided implicit grammar game is effective. Fully unguided discovery is not assumed: models and hints remain available and their use is logged.

## Play a short visit

Open `index.html`, or the hosted pilot. Aim for a short first observation, approximately 8–12 minutes; adjust to the learner rather than forcing completion.

1. Ask the learner to help Abby see the crossing. Let the learner choose moonlight or carry and light the lantern.
2. At the bridge, observe what happens when the learner explores UNDER versus ACROSS.
3. Let the learner choose a glowing flower or a carried gift; BRIGHT and QUIET change the flower's appearance.
4. Let the learner compare SOFTLY and SLOWLY in the river. DEEP changes the river's appearance.
5. Luma points to the flower, establishing what IT refers to before the next wish.
6. Let Mei open the gate, or let the gate open itself.
7. Offer the optional two-wonder spell with AND or THEN. Accept a voluntary stop.

After a successful spell, Hear it provides a device-voice English model. Echo to Luma invites speaking without blocking progress. The “I tried my echo” button records self-reported practice only. There is no microphone recording, speech recognition, accent judgment, or automatic pronunciation score. Device voices may be absent or vary; use the displayed sentence with a teacher/partner then.

Avoid routine grammar lectures while observing this pilot. Use neutral story prompts: “What changed?” “What else could you try?” “Where did Abby go?” A requested example is guidance and must be recorded as support. Optional English meanings also make this a supported language activity; rune-only play alone would not establish English transfer.

## Functional language scope

| Target | What is implemented | Limit |
| --- | --- | --- |
| Nouns and verbs | Names/objects and supported actions control distinct outcomes | Small, bounded vocabulary |
| Articles/determiners | THE precedes singular common nouns in the pilot | A/AN and broader determiner contrasts are future work |
| Adjectives | BRIGHT/QUIET flower effects; BRIGHT light; DEEP water | Unsupported combinations receive feedback |
| Adverbs | SOFTLY/SLOWLY water, slow movement/gate timing, soft light/gate glow | No general-purpose language engine |
| Pronouns | SHE is anchored to Mei; IT uses an established referent | Broader pronoun contrasts and ambiguity remain future work |
| Prepositions | UNDER/ACROSS the bridge have different consequences | Limited scene relationships |
| Conjunctions/sequencing | AND/THEN combine two supported events | Other connectors are deferred |
| Interjections | Not implemented in this pilot | Add later as expressive choices with meaningful social consequences |
| Syntax | Order and complete wishes determine action; alternatives accepted by meaning | Rune base forms are converted to inflected English models |
| Pronunciation | English listening, optional echo, teacher-rated novel oral messages | Practice is not evidence of improved production |

Depending on the instructional taxonomy, articles may be treated within determiners and THEN as an adverbial sequencer rather than a conjunction. Teach and assess the adopted taxonomy consistently. This pilot does not claim complete coverage of every part of speech. Recognizing grammatical functions in action differs from naming categories such as “noun” or “adverb.”

## Observe transfer separately

Use the Teacher panel for draft Forms A/B/C before play, after play, and after a planned delay. A 3–7 day delay is a practical pilot choice. The parallel tasks need review for equivalent difficulty and vocabulary familiarity. Alternate the forms across future learners when feasible.

Collect unmodeled performance first: sentence order, enacted spatial meaning, and a new spoken/written proposition. Use new pictures, words and sentence contexts without runes, role colors, or game rewards. Explain unfamiliar word meanings neutrally without supplying the sentence order. Record reading assistance, prompts, and prior knowledge. Preserve first attempts. Give no corrective grammar feedback between immediate and delayed measures when assessing retention.

The optional explicit word-role task can prime rule awareness. Administer it only after the final unprimed transfer check, or in a separate assessment condition. Do not interpret knowledge of grammar labels as proof of implicit acquisition. Likewise, successful performance without an articulated rule does not by itself prove unconscious learning; add age-appropriate awareness questions and stronger comparison conditions for that research claim.

Pronunciation: score intelligibility of an unmodeled novel oral response before giving an echo model. Use 0 = message not understood, 1 = understood after repetition, 2 = understood on first hearing. This is a local observation rubric, not a standardized pronunciation test. Separately record modeled echoes, target sounds/stress, and accessibility needs. If a learner uses a non-oral access path, collect grammar/meaning evidence there and mark oral pronunciation unobserved; do not infer pronunciation from tapping.

The teacher panel scores syntax, meaning, production, pronunciation, and an optional role-awareness task independently. Do not combine them into a validated total. Export preserves raw event timestamps and support use so time between visits can be inspected. Idle tab time is not engagement time.

For the next observed visit, record voluntary start/replay, time until the first purposeful action, exploration of a second valid solution, independent repair, hints, and a short effort rating. Ask a concrete choice after observing: “Would you rather help Abby explore, make the garden change, or finish?” A replay button click alone is a behavior signal, not durable motivation.

## Interpreting results

Continue this design direction if the learner voluntarily explores alternatives and appears more willing to return. Inspect whether unaided novel English performance is retained after a delay. Treat either improvement as formative single-learner evidence. Do not attribute a causal effect to the pilot, a single feature, or the IND-GBL framework: multiple features changed together, prior experience is uncontrolled, and novelty/teacher support can explain changes. A later comparison should match vocabulary, duration and support while varying the consequential game mechanic.

## Accessibility, saving, and verification

- Native buttons, dialogs, headings, focus indicators and keyboard operation are provided. The rune bank remains visually untranslated; a screen-reader-name option provides an alternate access route. No drag-only action or timed gate is used.
- Desktop is designed for one viewport at common landscape sizes. Narrow screens and zoomed layouts permit scrolling to keep controls readable. Responsive CSS is implemented; no WCAG conformance certification is claimed.
- Pilot progress and teacher notes are stored in the browser's local storage, separately from v1.0.4. Reload preserves the garden; recast the current wish to resume its progression control. Replay starts a fresh garden while retaining the session's evidence; Start a fresh garden clears this pilot's evidence after confirmation.
- No child name is requested. No recording or remote analytics is collected. Teacher notes can contain whatever the teacher enters; use a learner code rather than identifying information. Export downloads local JSON on request.
- `tests/pilot.test.js` checks alternative solutions, different state consequences, referent handling, sentence realization, ordered preconditions, atomic failure, and every mission's reachable goal. Historical QA reports refer to prior versions.

## Sources

CAST. (2024). *Universal Design for Learning Guidelines version 3.0*. https://udlguidelines.cast.org/

Sweller, J., van Merriënboer, J. J. G., & Paas, F. (2019). Cognitive architecture and instructional design: 20 years later. *Educational Psychology Review, 31*, 261–292. https://doi.org/10.1007/s10648-019-09465-5

User-provided project evidence: `Spellbound_Student_Survey_Zilin-10-29-2026.pdf`; `spellbound-relics-v1.0.4-enchanted-environment.zip`. These are project evidence, not published validation of the framework.