AI Atlas · ELIZA · Learner 1 of 2 · 20 September 2026

How did it answer that?

Supported practice: predict → reveal → explain.

This is an original two-rule teaching script inspired by ELIZA, not the historical program or a live chatbot. Use neutral authored sentences. This station is separate from the three-room trust quest.

Optional rule support

Read when needed, and record that help. Find the first whole word NEED. Discard the prefix, keep the suffix, remove trailing sentence punctuation, replace MY with YOUR, and insert it after “Why do you need” with a question mark. If NEED is absent, reply “Please say more about that.”

Keep the first response

1. “I need my library card.” Before revealing: predict the rule and reply.

Observe the trace. Which exact words were discarded, kept and changed?

2. “My library card is missing.” Predict before revealing. Record the reply and the step that explains it.

3. “I do not need my library card.” Predict before revealing. Then compare both sentences and their replies.

My first explanation and any revision, with trace evidence:

Rule support read before prediction: ☐ yes ☐ no ☐ not recorded
Access assistance (reading, scribing, other): __________ Content hints: __________
Written or scribed words are records, not automatically verified understanding.

AI Atlas · ELIZA · Learner 2 of 2 · Keep answers separate

A fresh structure

Finish practice first. Start without reopening the rules if you can.

Challenge A: “The sign says I need my library card.”

The same two-rule script runs here. Predict the selected rule and exact reply before revealing. If you need the rule or a content cue, use it and record when. Help is welcome; it changes what this attempt can show.

First prediction — rule and exact reply:

Before reveal — why I expect that:

After reveal — actual reply and exact trace evidence:

Any revision; a supported conclusion and an overclaim I rejected:

Access assistance and when: __________ Content hints and when: __________
Reasoning: ☐ written ☐ scribed ☐ not recorded. Do not replace the first response.

A reason to return

Challenge B: “My library card needs a cover.” Save this for a later visit if possible. It tests a different structure. Predict rule and exact reply before reopening the trace; then explain which step determines the result.

Return date / delay: __________ Prior exposure to this case: __________
Assistance: __________ Exact explanation or scribed words: __________
Completion alone does not demonstrate understanding or delayed retention.

Facilitator only · separate from learner predictions

Evidence, limits and next observation

Worked practice

  1. Match: NEED selects the rule. The suffix “my library card” becomes “your library card”; reply: “Why do you need your library card?”
  2. Fallback: no whole-word NEED occurs. Capture and substitution are skipped. Reply: “Please say more about that.”
  3. Negation: “I do not” is discarded before NEED. Both positive and negated inputs produce the same question. This shows that this rule ignores the prefix, not that every language system ignores negation.

Fresh challenge A: reported speech

Input: “The sign says I need my library card.” The prefix “The sign says I” is discarded. The captured suffix is unchanged from practice, so the reply is “Why do you need your library card?” The rule does not represent who said the words. The trace supports prefix removal; it does not show recognition of the sign as a speaker or identification of the learner's actual need.

Return challenge B: word boundary

Input: “My library card needs a cover.” NEEDS is not the whole word NEED. The fallback is “Please say more about that.” Capture and substitution are skipped. This supports exact keyword matching here; it does not establish a general inability to handle grammar or the behavior of historical ELIZA.

Observe without supplying the answer

Record first prediction, help timing, actual reply and revisions separately. Ask which exact words support the conclusion. Distinguish access support from content hints. A correct revision after seeing the trace differs from a first prediction; a later return attempt is not automatically a measured retention result.

Preserve old practice records as practice. New transfer records use assessment version 2. Selected evidence is checked; optional prose is not graded. No logged content help means only that no in-app help was recorded, not that outside help was absent.

Historical scope

Joseph Weizenbaum's 1966 ELIZA paper describes keyword-triggered decomposition and reassembly. This authored model omits historical ranking, memory and alternative reassembly rules.

Learner understanding, unfamiliar facilitator use, physical devices, screen readers, commercial demand and delayed return still need separate observations.