Week 08 · lesson

Lab: Trace Recurrent State

Mission

Run the local Recurrent State Trace Lab. Compare sequences that reuse the same update rule, preserve every calculation, and export a Recurrent State Trace Record.

Guided start

Select robots → build and run the state trace.
Confirm the final state is 0.800.
Preserve both candidate-score calculations.
Select build → robots and run again.
Explain why the final state changes although the tokens do not.

Procedure

Predict the final state for each supplied sequence.
Run all three traces and preserve each arithmetic step.
Compare the two sequences containing the same two tokens.
Trace how adding quickly changes the carried state.
Preserve your first understanding claim and write a bounded correction.
Complete Observation, Measurement, Decision, and Limitation.
Export the record and confirm it names ai-for-beginners-recurrent-state-lab.

Success criteria and boundary

every update can be reproduced from the displayed operands;
the same recurrence rule is visible at every step;
order dependence is supported by two matched-token traces;
candidate scores remain separate from probabilities and facts; and
the limitation rejects claims about training, understanding, safety, or real-model performance.

Use only supplied tokens and rules. The Lab contacts no model or service.

Robotnix Lab · contract v1

Recurrent State Trace Lab

Trace deterministic recurrent updates, compare token order, and bound claims about language-model state and scores.

Help and boundaries

A deterministic browser simulation uses invented token values and equations and contacts no model or service.

No personal data, private prompts, or API keys. This Lab makes no hidden remote writes and does not persist your work after the page session.

No result generates external text, trains a model, or authorizes action.

Lab draft is active for this page session only.

Objectives

Trace prior state and current input through a recurrent update.
Demonstrate order dependence with matched tokens.
Separate teaching scores from probabilities, meaning, and real-model performance.

Procedure

Inspect sequenceRecord ordered tokens, token values, initial state, and update rule.
Trace statePreserve every prior-state, input-value, and new-state calculation.
Compare orderRun the two matched-token sequences and compare final states and candidate scores.
Bound output claimSeparate supplied arithmetic from probability, understanding, truth, safety, and real-model behavior.

Workbench

Trace every state update before interpreting the output.

Token sequence

Select a supplied sequence and run the deterministic trace.

Boundary: This invented recurrence is not trained. Its states and scores are not probabilities, meanings, facts, safety judgments, or evidence about a real language model.

Text and static fallback

Every token, input, prior state, update, final state, score, and boundary appears in accessible text and tables.

Robots-build ends 0.800; build-robots 1.000; robots-build-quickly 0.600.

Evidence record

Record only the evidence requested. Do not enter names, personal information, private prompts, or credentials. This draft stays in memory only unless you explicitly export it.

Sequence declaration *Record ordered tokens, supplied values, initial state, and recurrence.State traces *Record every update and both candidate scores for all three sequences.State claim repair *Preserve and repair one state-means-understanding claim.Limitation *State what the teaching recurrence cannot establish about trained language models.

Success criteria

The sequence and recurrence boundary are explicit.
Every state and score is reproducible from displayed operands.
The correction follows deterministic arithmetic only.
The limitation rejects unsupported probability, understanding, safety, and performance claims.

Complete every procedure step and required evidence field before export.

Import evidence JSON

Vocabulary lab

Flip the idea, not just the card

Explain the term before you reveal the back. Then compare your explanation with the definition, example, and warning.

Read all terms without animation

sequence: An ordered collection of input elements. Example: Robots followed by build. Do not confuse it with: A sequence is not an unordered bag of tokens.
state update: The calculation combining prior state and current input into a new state. Example: 0.5 times 0.8 plus 0.4 equals 0.8. Do not confuse it with: It does not preserve every prior token exactly.
order dependence: A result changes when the same elements appear in a different order. Example: The two-token traces end at 0.8 and 1.0. Do not confuse it with: The token set remains the same.
candidate: One possible output considered by a scoring rule. Example: Safely and quickly are supplied candidates. Do not confuse it with: A candidate is not a verified answer or action.
Recurrent State Trace Record: Evidence preserving inputs, state calculations, scores, claim repair, and limitations. Example: The Lab export. Do not confuse it with: It is not evidence from a trained language model.