You handle the setup.
They make the discoveries.

Props and materials

• ✊ RPS cards
• ⑤ Five-round score card
• ⇢ Classifier-vs-rule map

Tool handoff and actor

Prerequisites

• Open Teachable Machine in a separate tab and choose an image project.
• Keep the camera aimed at cards and props, not faces.
• Print or draw three cards with both an icon and a distinct color.

Expected save

• RPS Classifier v1
• Mission 1 Dev Log

Recovery

If every card looks the same to the model, add brighter light and a plain background. If the camera is blocked, use image files for practice and return to live testing with an adult.

Props and materials

• ◆ Three icon sets
• ▥ Data Tally
• H Held-out markers
• Aa Label map

Tool handoff and actor

Prerequisites

• Choose three non-person labels and write them exactly as they should appear.
• Make a tally row for each class and reserve one example per class for testing.
• Remind the learner that equal counts are a starting point, not proof of fairness or accuracy.

Expected save

• Image Model v1
• Mission 2 Dev Log

Recovery

If a class dominates, compare the counts and backgrounds before retraining. If the model learns the table instead of the card, move every class through the same locations.

Props and materials

• 9 Nine Job Cards
• △ Three system station cards
• ↦ System Map

Tool handoff and actor

Prerequisites

• Place the CLASSIFIER, GEMINI DRAFT, and EXACT RULE station cards on the table.
• If Gemini Apps access is unavailable, use the supplied sample draft cards.
• Keep Gemini use on an adult-controlled screen and enter no private information.

Expected save

• System Map v1
• Mission 3 Dev Log

Recovery

If a job seems to need two stations, split it into generation and exact decision steps. If Gemini access is unavailable, use the printed draft examples and keep the same learning proof.

Props and materials

• ♫ Command cards
• … BACKGROUND card
• ▤ Audio tally
• V1 Audio Model v1 passport

Tool handoff and actor

Prerequisites

• Choose two short commands and a BACKGROUND class.
• Close apps that may record audio and show the learner the browser microphone indicator.
• Plan to stop the microphone and export the model when the mission ends.

Expected save

• Audio Model v1
• Mission 4 Dev Log

Recovery

If BACKGROUND wins every time, add clearer command examples at normal volume. If commands swap, shorten them or choose words with more different sound shapes.

Props and materials

• 1st First-Test Log
• ☀ NEW LIGHT card
• ∠ NEW ANGLE card
• ? Shortcut detective card

Tool handoff and actor

Prerequisites

• Choose two conditions that were not used for training.
• Write the conditions down before any new examples are added.
• Keep the original model unchanged during the first test.

Expected save

• First-Test Log v1
• Mission 5 Dev Log

Recovery

If everything works, make the next condition meaningfully different rather than simply darker. If nothing works, confirm the correct Image Model v1 and label map are loaded.

Props and materials

• % Score tracker
• ≥ Threshold guide
• ▶ ACT card
• ? ASK AGAIN card

Tool handoff and actor

Prerequisites

• Lay out the score cases without revealing the answer key.
• Explain that the slider is a game rule placed after the model score.
• Use small, harmless game costs only.

Expected save

• Threshold Policy v1
• Mission 6 Dev Log

Recovery

If every case acts, raise the threshold or confirm ASK AGAIN is connected. If no case acts, lower the threshold only after checking the scores and labels.

Props and materials

• ✦ Structured Prompt Builder
• G Goal cards
• C Context cards
• ! Constraint cards
• Ex Example cards
• {} Format cards

Tool handoff and actor

Prerequisites

• Confirm Gemini Apps access is allowed for the account and region, or use the offline sample draft.
• Use an adult-controlled screen and a sanitized fictional prompt.
• Keep the requested output to level data, not executable code.

Expected save

• LevelSpec v1
• Mission 7 Dev Log

Recovery

If the draft is missing fields, make the output format and constraints more explicit. If Gemini is unavailable, revise the supplied imperfect draft using the same checklist.

Props and materials

• S Fictional Source Pack
• C Claim cards
• ↔ Evidence board
• × Unsupported markers

Tool handoff and actor

Prerequisites

• Use only the facts printed in the course source pack.
• If Gemini is used, submit the sanitized mission facts from an adult-controlled screen.
• Mix supported, uncertain, and unsupported clues for the learner to inspect.

Expected save

• Grounded Route v1
• Mission 8 Dev Log

Recovery

If two sources conflict, use NEEDS CHECK instead of guessing. If a clue has no source, reject it even when it sounds helpful.

Props and materials

• ▣ Image input cards
• ♫ Audio input cards
• ⊕ FusionTrace
• AND Exact AND-rule board

Tool handoff and actor

Prerequisites

• Confirm the Audio Model v1 identifier matches the one saved in Mission 4.
• Draw two input boxes leading into an AND gate and then ACT or ASK AGAIN.
• Keep live model testing in Teachable Machine tabs supervised by an adult.

Expected save

• FusionTrace v1
• Mission 9 Dev Log

Recovery

If the combo acts when one input fails, replace OR logic with the required AND rule. If the audio identifier changed, return to the frozen Mission 4 artifact instead of silently substituting it.

Props and materials

• V2 Version Lab log
• ⚓ Regression anchors
• ⇄ Comparison grid
• ↶ Rollback card

Tool handoff and actor

Prerequisites

• Copy the model project so Image Model v1 stays recoverable.
• Choose two logged failures and four previously successful anchor cases.
• AI Studio, if demonstrated, stays adult-operated and produces no learner-facing API client.

Expected save

• Version Lab v2 Comparison
• Version Workflow Demo Checked
• Mission 10 Dev Log

Recovery

If the comparison uses different cases, reset and run the same locked set. If v2 loses an anchor, keep v1 while planning a smaller targeted change.

Props and materials

• ▦ Condition Matrix
• ♥ Privacy and accessibility checklist
• MC Model Card
• ✋ HUMAN DECIDES card

Tool handoff and actor

Prerequisites

• Create bright, dim, plain, busy, quiet, and noisy test stations using safe props.
• Provide touch, visual, and silent alternatives for the play loop.
• Keep every classifier label about props or signals, never people or traits.

Expected save

• Release Model Card v1
• Mission 11 Dev Log

Recovery

If an alternative skips the lesson, route it through the same decision and explanation goal. If the game cannot handle a condition safely, document the limit and abstain instead of hiding it.

Props and materials

• 12 Finale setup checklist
• LIVE Live FusionTrace
• ▶? ACT and ASK AGAIN proof card
• ✓ Final dev log

Tool handoff and actor

Prerequisites

• Open the image and audio model projects in supervised external Teachable Machine tabs.
• Confirm the audio identifier matches Mission 4 and aim the camera at a tabletop prop zone.
• The parent observes the live demonstration and helps the learner read both signal panels.
• Stop camera and microphone permissions when the demonstration ends.

Expected save

• Final FusionTrace and Dev Log
• Mission 12 Dev Log

Recovery

If a sensor is unavailable, use Practice mode to learn the interface, but do not mark the supervised capstone complete. If the audio identifier differs, return to the frozen Audio Model v1 before continuing. If every pair acts, reconnect the threshold and AND rule so uncertain cases reach ASK AGAIN.