HOW TO READ THE FICTIONAL DECISION-FIRST CARD
Ahmadreza Samadi | https://a-samadi.com/about.html
Prepared 17 October 2026. Educational planning example; no experiment was run.

The decision is whether to try a supervised pilot for private summaries of public
housing reports. It is not a decision to automate publication, make a property
recommendation or claim that AI has already saved time.

1. Define an action, not a model preference.
The proposed output helps a researcher decide whether a trial is worth running.
The card does not name a model or imply that Ahmadreza uses a particular tool.

2. Make acceptance checkable.
In this fictional plan, each factual claim must be supported by its source and no
material caveat may be omitted. The researcher accepts or rejects the result.
A fluent summary or the model's own confidence is not evidence of acceptance.

3. Resolve source eligibility before starting.
Public access alone is not permission to reuse a report or submit it to a tool.
No actual reports, reuse permissions, provider terms or source coverage have been
checked for this fictional pilot. Verify these for the actual intended use. Do
not include confidential or personal material. The example sends nothing.

4. Compare the same task and record the whole process.
The proposed six reports are an illustrative planning choice, not a recommended
sample size or evidence of reliability. Define the eligible cases, instructions,
review rules and observation cutoff before testing. Apply the same acceptance
criterion to manual and AI-assisted work. Record failed drafts and retries as
well as accepted results, model/tool charges, review effort and other costs.
No costs or working-time measurements are provided by this example.

5. Use the stop rules.
Stop or return to manual work when source-use rights are unclear or claims cannot
be verified. Correcting a draft does not establish that the next draft is safe.
Keep a human responsible for the final decision and any eventual publication.

6. Leave the outcome unknown until observed.
The completed field says "Unknown: no pilot run." It does not claim zero cost,
zero errors, accepted results, savings, performance or a company deployment.
A small pilot would still not establish broad reliability or ROI.

HOW TO USE THE FILE
The example is a separate editable copy of the existing ten-field blank card.
It has explicit fictional labels. The original blank PDF is unchanged. Start
real planning from a clean blank copy; do not carry fictional answers into your
records. PDF viewers differ: native-reader edit/save behaviour is not verified.
There are no scripts, automatic submission actions or external connections.
Visible links open the existing tool or biography only when followed.

Full interactive brief: https://a-samadi.com/tools/before-you-build-ai.html
Related public cost-recording tool: https://github.com/ahmadreza-smdi/ai-workflow-cost
Author-proposed thinking aid, not a validated assessment. AI-assisted preparation.
No company/customer experience, measurements or results are represented here.
