Toolkit · Template 2 of 5 · v1.0

Second-Reader Checklist

A record that someone other than the maker looked at a change, what they looked at, what they tried to break, and what nobody checked.

For anyone shipping AI-assisted or hand-made work (code, a page, a document, a message) who wants a second reader before calling it done. “The checks passed” and “ready to release” are different statements.

  • Second reader: a different person or a different AI model from the maker, who checks the work against the requirement before it is called done. A second session of the same model does not count.
All examples are invented and illustrative. Using this template does not guarantee any result.

Copy the templateWorked exampleWhy these fieldsAll five templates

How to use it

  1. The second reader fills this in from the live result, not from the maker’s summary of it.
  2. If “Made by” and “Read by” are the same, the checklist does not count. Two sessions of the same AI model tend to share blind spots, so use a person or a different model.
  3. Try to break the change on purpose and write down what you tried, including the attempts that failed to break it.
  4. State plainly what was not checked. An honest gap is more useful than a confident blank.
  5. Read the conclusion as a statement about these checks only. A second reader lowers the chance of a miss; it does not remove it.
  6. Working alone? You cannot be a true second reader of your own work. The nearest substitute: wait 24 hours and re-read it cold against this checklist, or ask any peer. A cold re-read by you is not a second read, so record it as “Read by: me, cold re-read after 24 hours (not a second read)”, and the checklist then does not count as a second read. A peer who is not the maker is a second reader. Neither removes the chance of a miss.

Copy the blank template

Plain text. Paste it into a document or notes app. Nothing on this page collects what you type. Free to copy, adapt and share.

Second-Reader Checklist  (template v1.0)

What changed:
  [Plain description of the change. A change can be code, a page, a document or a message.]

Made by:
  [Who or what produced it.]

Read by:
  [Who or what is checking it. Must differ from “Made by”.]

Automated checks run:
  [Which checks, and pass or fail for each.]

Checks done by hand:
  [What the second reader actually did, not only re-ran.]

Cases tried to break it:
  [What was deliberately attempted, and what happened.]

Not checked:
  [What was left out, stated plainly.]

Second reader’s conclusion:
  [In the reader’s own words, about these checks only.]

Evidence:
  [A screenshot, log excerpt or output the conclusion rests on.]

Date and time:
  [When the reading was finished.]

Worked example

Example: reading an account-recovery change. All names and numbers are invented and illustrative.

FieldExample entry
What changedAccount-recovery links now stop working after one hour. Before, they never expired.
Made byAgent A
Read byAgent B, a different AI model from Agent A, given only the requirement and not Agent A’s notes.
Automated checks runLink-expiry check: pass. Recovery-email send check: pass.
Checks done by handRequested a recovery link, waited past the hour, confirmed the old link was refused.
Cases tried to break itOpened the link at 59 minutes (worked, as intended). Opened it at 61 minutes (refused, as intended).
Not checkedBehaviour when the reader’s device changes time zone during the hour.
Second reader’s conclusionMeets the requirement as written, for the cases above.
EvidenceTwo screenshots: one before expiry, one after.
Date and timeExample only: 2000-01-01, 10:00.

Keep the checklists as a dated log. Do not edit a past entry. If you learn something new, add a new entry that refers to the old one.

Why these fields are here

The plain fields explain themselves. These are the ones people skip, and the reason each one is there.

Read by
The checker needs a different point of view. A different person, or a different AI model given only the requirement, has one. A second session of the same model has less of one.
Checks done by hand
Re-running the maker’s checks finds what the maker already found.
Cases tried to break it
Only an attempt to break it shows where its edges are, and a failed attempt is as useful to record as a successful one.
Not checked
A blank reads as “all clear”. A stated gap shows the next reader what nobody has looked at.
Evidence
A conclusion with nothing behind it cannot be re-read later.

All five templates