# Annotation portfolio evidence pack

Prepared and published 19 October 2026. A portable companion to the LearnWithZavi annotation lab: six synthetic records, preserved fictional mistakes, explained corrections and reusable templates.

## Start here
1. Read learner/rules.md, then learner/sources.csv or sources.json. There are six selected records.
2. Work in a copy of my-first-pass-template.csv. Preserve your first attempt before opening the worked-example folder. If you already know these answers, record that exposure; do not call this an unseen test.
3. Compare your attempt to the corrected pass. Use my-decision-log-template.csv for corrections and preserved decisions.
4. Complete human-review-template.csv and my-README-template.md honestly. A blank reviewer field is not approval.

The worked-example folder contains an intentionally flawed fictional first pass, a corrected pass and a six-row decision log. Its answers were not produced by a real learner. First-pass matches are intent 1/3, whole entity records 1/2, reply ratings 1/1. Corrected counts are 3/3, 2/2 and 1/1 after key review. Reasons are checked separately. The example has three intentional errors: C05, C10, N01.

## File handling
Use a text editor if spreadsheet import changes values. CSV files are UTF-8 with a header row. answer_json contains a JSON string for intent/rating or a JSON array for entities. Keep identifiers as text. A correct N01 answer includes ORDER_ID text "0042". Blank cells mean unanswered, while [] explicitly means no entities. The JSON worked copies are optional readable alternatives; the CSV and JSON answers agree. No macros, scripts or external connections are required.

Keep the example files unchanged. Save your own first pass separately, then create a corrected copy. Do not claim the fictional first pass as your work. Files and any checksums can be edited; neither proves authorship, chronology or an independent review.

## Source and scope
The six source texts and corrected answers are copied from the existing LearnWithZavi practical training bundle dated 12 October 2026, guideline 1.0. The fictional first pass follows the reviewed editorial draft dated 15 October 2026. On 19 October 2026, all six selected source texts and corrected answers were compared with the existing public lab CSVs and answer keys and matched exactly. This six-record selection is not the full 26-record lab.

All customer records are synthetic. The 16 November date belongs to the fictional scenario and is not the practice date. AI-assisted packaging and source review were used. No human reviewer or learner pilot has signed off. The material makes no claim about real clients, paid work, accreditation, learning gains, speed or job outcomes.

## Continue learning
Annotation lab: https://www.learnwithzavi.com/course/ai-annotation-jobs/07-lab-read-the-rules
Quality handoff: https://www.learnwithzavi.com/course/ai-annotation-jobs/11-lab-quality-handoff
Documenting work: https://www.learnwithzavi.com/course/ai-portfolio/04-documenting-your-work
These are existing learning destinations.

## Open the files
- [Rules](learner/rules.md)
- [Six sources](learner/sources.csv)
- [First-pass template](learner/my-first-pass-template.csv)
- [Decision-log template](learner/my-decision-log-template.csv)
- [README template](learner/my-README-template.md)
- [Human-review template](learner/human-review-template.csv)
- [Worked example with answers](worked-example/README.md)

## Questions and review
Contact learn@acelytics.com. A completed human review must name the reviewer and record the actual date, records checked, findings and unresolved corrections. Publication approval is not record-by-record validation.