Writing Evidence Card: From Tool Polish to Accountable Delivery

Copy this card into a private note for one real writing task at a time. Do not include passwords, client data, unauthorised third-party information, exam answers, or unnecessary personal details. A final version may be public without making the raw draft, feedback, and tool record public.

1. Define the Task, Reader, and Responsibility

# Writing Evidence - YYYY-MM-DD

Real situation:
Task: explain / request / status / decide / narrate / reflect / hand over
Audience and relationship:
What should they understand, decide, or do afterwards?
Genre, length, and deadline:
Completion standard:
Dictionary / translation / search / AI / human help allowed:
Confidentiality, copyright, privacy, and authorship boundary:
Final accountable owner:

When an institution, school, client, or hiring process prohibits assistance, follow the current rule.

2. Preserve an Unaided Draft

Raw draft location:
First-version time and length:
Outline or sources allowed:
One sentence the reader should remember:
Fact I trust least:
Gap most likely hidden by attractive prose:
Action I expect from the first version:

Do not check whole-sentence translations or ask AI for prose first. Keep pauses, repetition, and failed paragraphs as the baseline.

3. Build a Fact and Responsibility Ledger

Claim/content	Type	Source, version, and location	Check date	Confidence	Publishable?	What changes it?
	fact / number / quotation / experience / inference / advice / other-person data

What I knew then and learned later:
Third-party content requiring permission:
Part still unverifiable:
Content to remove or anonymise before delivery:

4. Preserve the Document Skeleton

Problem facing the reader:
Direct answer:
Support 1:
Support 2:
Concrete example/minimum fact:
Limit, counterexample, or unknown:
Reader's next step:

Paragraph	Function	Must keep	Delete/move	Reason
	context / problem / claim / reason / evidence / example / turn / limit / request / decision / handover

5. Record Translation Meaning Changes

What I mean	English first draft	Translation/tool candidate	Final decision	Reason

Did agent or responsibility change?
Was certainty strengthened or weakened?
Did negation or condition scope change?
Does tone fit the relationship?
Non-equivalent institutional/cultural term:
Glossary and reviewer:

Natural, fluent, and faithful are different judgments. The final author explains the trade-off.

6. Complete Five Revision Passes

Pass	Main problem	Before	After	Why it changes the task result
Task and fact
Structure and evidence
Sentence and meaning
Register and voice
Delivery and boundary

Necessary correction:
Ambiguity repair:
Register choice:
Personal voice retained:
Change I rejected because I could not explain it:

7. Record Feedback Uptake

Ask for a retelling before comments:

Reader's retold point:
Evidence and limits the reader heard:
Next step the reader understood:
Place requiring guessing:

Feedback	Source/relationship	Category	Decision	Reason	Used in parallel text?
	real reader / peer / teacher / editor / AI	correction / ambiguity / structure / evidence / register / style	accept / partly / reject / defer

An improved old draft does not prove new-task use. Feedback begins to show transfer when it enters a parallel text.

8. Record AI and Automated Tool Use

Tool / model / version and date:
Draft and sources that existed before use:
Request: diagnosis / reader simulation / fact candidate / correction
         / version comparison / counterexample / other
Scope generated or rewritten by the model:
Facts, quotations, or sources it may have invented:
My item-by-item verification:
What I accepted/rejected, and why:
Disclosure required for school, client, employer, or reader:
Parallel task after closing the tool:

Model output is a candidate. It does not own authorship or carry final accountability.

9. Check Asynchronous Delivery

Check	Content	Accepted?
One-sentence conclusion/subject
Completed work and evidence location
Risk, block, and unknown
Who decides what by when
Next step, owner, and checkpoint
What happens without a response
Failure, stop, and rollback
Links, attachments, and permissions

Can the reader continue without another meeting?
Real response or execution result:
Gap that still required oral explanation:

10. Test Delayed Retention and Transfer

Time	What is removed	Changed condition	Output	Result
Day 1	No AI-written prose	Original task	Draft and delivery version
Days 3-7	Old sentences and comments removed	New topic, same genre	Parallel text
Day 14	Recent feedback and translation removed	New audience, same information	Register-transfer text
Day 30	All practice prompts closed	Real email, report, application, or handover	Complete delivery

Structure and judgment available after seven days:
Problem that returned with a new topic:
Did whole-sentence translation/full rewrite dependence decrease?
Did the reader guess less?
One variable for the next cycle:

11. Score and Decide Whether to Sign

Score 0-2: 0=not completed, 1=prompt-dependent or unstable, 2=completed under the current condition.

Item	Day 1	Day 7	Day 14	Evidence reason
Leave an independent draft
Make task, reader, and next step clear
Separate fact, source, and inference
Support conclusion with structure and evidence
Explain translation/AI/feedback changes
Let async reader continue or roll back
Deliver after topic, audience, or genre changes

Current version: willing to sign / needs verification / narrow scope / do not deliver
Most important evidence:
Conclusion I still cannot draw:
Smallest next task:
Next review date:

Related chapters: Writing: From Draft to Verifiable Revision | Grammar | Reading | Learning English with AI | Job-search English | Artifacts