Use this reading version with or without JavaScript. All records are authored exercises; no model or tools run. Make your decision before expanding an explanation.
Trusted task
Compare invented condition snapshots. A useful follow-up changes one intended variable, records what stayed fixed, and keeps missing information visible. These are authored records, not model evaluations.
Exercise criterion
Interpret differences only with the changed and unknown conditions made explicit.
01 / One intended change
The input changes while the recorded conditions match.
Question
Does requesting a one-line answer change the observed formatting?
Observation
No model results are supplied. Assess the comparison design only.
Recorded conditions
Input
Baseline: Summarise the agenda.
Variant: Summarise the agenda in one line.
Target
Baseline: Invented assistant
Variant: Invented assistant
Version
Baseline: Fixture v1
Variant: Fixture v1
Context
Baseline: Synthetic agenda: review the BLUEBIRD meeting note at 14:00.
Variant: Synthetic agenda: review the BLUEBIRD meeting note at 14:00.
Settings
Baseline: Authored fixture; no inference
Variant: Authored fixture; no inference
Input placement
Baseline: Trusted user message
Variant: Trusted user message
Choose a decision
- Useful controlled follow-up; repeatability remains unknown
- This proves the input caused a reliable change
- Different inputs cannot be compared
Read the explanation for this complete record
Useful controlled follow-up; repeatability remains unknown
The recorded target, version, context, settings and placement match; the intended input change is isolated in this design. That supports a controlled follow-up. It does not supply an outcome, establish causality or show repeatability; real systems may vary across repeated attempts.
A useful comparison design, with results and repeatability still untested.
02 / Several conditions changed
A prompt difference is mixed with a version and context change.
Observation
A different answer is reported, but the supplied pair changed more than the input.
Recorded conditions
Input
Baseline: Summarise the agenda.
Variant: Summarise the agenda in one line.
Target
Baseline: Invented assistant
Variant: Invented assistant
Version
Baseline: Fixture v1
Variant: Fixture v2
Context
Baseline: Synthetic agenda: review the BLUEBIRD meeting note at 14:00.
Variant: Synthetic agenda: review the BLUEBIRD meeting note at 14:00.
Additional note: room changed to B.
Settings
Baseline: Authored fixture; no inference
Variant: Authored fixture; no inference
Input placement
Baseline: Trusted user message
Variant: Trusted user message
Choose a decision
- Attribute the answer difference to the prompt
- Separate the changed version and context before attributing an effect
- A matching target name makes the pair equivalent
Read the explanation for this complete record
Separate the changed version and context before attributing an effect
The input, version and context all differ. The target's display name alone does not make the conditions equivalent. Keep these differences in the record and prepare a follow-up that isolates the intended change before attributing an observed difference to the prompt.
Several conditions changed; the prompt's contribution is unresolved.
03 / Unknown is not a match
Two missing versions do not establish the same version.
Record quality
The version is absent from both records. One record also omits its settings.
Recorded conditions
Input
Baseline: Summarise the agenda.
Variant: Summarise the agenda in one line.
Target
Baseline: Invented assistant
Variant: Invented assistant
Version
Baseline: Unknown
Variant: Unknown
Context
Baseline: Synthetic agenda: review the BLUEBIRD meeting note at 14:00.
Variant: Synthetic agenda: review the BLUEBIRD meeting note at 14:00.
Settings
Baseline: Authored fixture; no inference
Variant: Unknown
Input placement
Baseline: Trusted user message
Variant: Trusted user message
Choose a decision
- Both missing versions match, so the pair is controlled
- Keep version and settings unresolved; do not call them matched
- Missing versions prove that different versions ran
Read the explanation for this complete record
Keep version and settings unresolved; do not call them matched
Absent information establishes neither equality nor a difference. Two unknown versions remain unknown, and missing settings remain unresolved. Preserve those gaps rather than guessing or silently treating empty values as equal.
The unknown conditions limit what this comparison can establish.