Input Inspection

Check Input

Stop guessing. Narrow the ask.

What It Is

Check Input is a public protocol for inspecting an AI ask before blaming the model. It helps identify what the input left missing, unclear, implied, conflicting, or unstated.

Use the protocol with any AI system. It should return an Input Summary and a Detected Gaps list using controlled terms.

What It Does

Check Input helps you inspect the original ask and identify what it forced the AI system to guess.

After the AI returns an Input Summary and Detected Gaps, bring that output back to PENGO and map it into the Workbench.

Open Explore Results

What It Does Not Do

It does not diagnose the model.
It does not prove causality.
It does not validate correctness.
It does not classify a failure.
It does not produce receipts.
It does not replace DIVU.

Copyable Protocol

Copy this protocol and paste it into an AI system with the input you want inspected.

AI Input Check Protocol

AI Input Check Protocol

Review the following input for missing or ambiguous criteria that could force an AI system to guess.

Do not rewrite the input.
Do not solve the task.
Do not improve style.
Do not infer hidden intent.
Do not assume business context.
Do not assume authority.
Do not assume risk tolerance.
Do not assume the desired outcome beyond what the input states.

Only identify what is missing, unclear, ambiguous, conflicting, unstated, undefined, incomplete, or implied.

First, identify the following only if they are present in the input:
- actor
- trigger
- action
- outcome
- scope
- constraint
- dependency
- sequence

If an element is not present, write:
- [element]&#58; not specified

Then identify gaps using only this diagnostic vocabulary.

Elements:
- actor
- trigger
- action
- outcome
- scope
- constraint
- dependency
- sequence

Gap types:
- missing
- unclear
- ambiguous
- unstated
- undefined
- conflicting
- incomplete
- implied

Rules for findings:
- Express every finding as: [gap type] [element]
- Do not invent new labels
- Do not use synonyms
- Do not explain with long paragraphs
- Keep findings short and literal
- List only meaningful gaps that could change interpretation, behavior, or output
- If the input would require the AI system to guess, mark the relevant element as not specified or as a detected gap
- Every Detected Gap must use only one allowed gap type and one allowed element
- Do not create new elements such as audience, channel, sender, recipient, tone, service details, format, or booking method
- If a missing detail does not fit the allowed element list, map it to the nearest allowed element or omit it

Return results in exactly this format:

Input Summary
- actor: ...
- trigger: ...
- action: ...
- outcome: ...
- scope: ...
- constraint: ...
- dependency: ...
- sequence: ...

Detected Gaps
- ...
- ...
- ...

Now review this input:

[PASTE INPUT HERE]

After the AI returns an Input Summary and Detected Gaps, bring the output back to PENGO and use Explore Results to map it into related Patterns and useful Lenses.

Controlled Vocabulary

Elements

actor
trigger
action
outcome
scope
constraint
dependency
sequence

Gap Types

missing
unclear
ambiguous
unstated
undefined
conflicting
incomplete
implied

Every finding should use this shape: [gap type] [element].

Examples: missing actor, unclear scope, unstated constraint.

missing actor = who is responsible, affected, addressed, or expected to act is not named.
unclear scope = where the request applies or stops is not clear.
unstated constraint = a rule, limit, requirement, or exclusion is assumed but not said.

See all Gap Examples

Example Output

Input Summary
- actor: not specified
- trigger: not specified
- action: write a refund email
- outcome: not specified
- scope: unclear
- constraint: not specified
- dependency: refund policy not specified
- sequence: not specified

Detected Gaps
- missing actor
- unstated outcome
- unclear scope
- unstated constraint
- missing dependency

Boundary

Check Input surfaces controlled gap terms. These terms can help surface related Patterns and useful Lenses, but they are related structural references, not a final diagnosis.

Check Input turns an AI problem back toward the ask.