Map Tax Docs

Look for a file matching pozn*.txt or pozn*.docx.txt. If found:

• Read it fully.
• Extract every named category/heading (e.g. "Allianz penzijko odpocet", "VZP", "Zonky", "Krypto").
• Ignore Google Drive / external URLs — treat them as already downloaded.
• Note any quantitative hints embedded in the notes (e.g. "úrok 55,20 EUR, daň 11,03 EUR") — these can confirm file-to-category matches.

4. Read each document

For every non-notes file read enough to identify:

• Document type (income confirmation, ELDP, insurance statement, brokerage report, etc.)
• Issuer / institution name
• Tax year / period covered
• Taxpayer name (to confirm it belongs to this case)
• Any key amounts

For large or garbled OCR files, read the first ~100 lines — enough to identify, no more.

5. Map files to categories

Match each file to a pozn category. Rules:

• A file may belong to more than one category (note it in both, mark as dual).
• If the pozn lists N links for a category but only M < N files are found, flag the gap.
• .failed files: report their error message content (usually "Incorrect password" or a tool error), their likely category based on filename, and any password hint visible in the notes.
• Duplicate files (same content, different attachment IDs from datová schránka): mark clearly.
• Files not referenced in pozn but present in the directory: add them to an "Extra files" section, classifying as tax-relevant or administrative/personal.
• Subdirectories: if a vysledek/ or similar output folder exists, list its contents separately as "Output / filed documents" — these are not input source documents.

6. Identify issues

Flag explicitly:

• Password-protected files that failed extraction (include password hint if found in notes).
• Missing documents (pozn references something with no matching local file).
• Ambiguous mappings (file could belong to multiple categories, reason unclear).
• Duplicate files.
• Prior-year documents that are supplementary rather than 2024 income documents.

7. Write source-map.md

Write the analysis to <target-directory>/source-map.md.

Structure of the output file:

# Source Document Mapping — Tax Year <YEAR>

Inferred from `<notes-file>` and directory contents.

---

## 1. <Category name from pozn>
*Brief description of what this category covers*

| File | Notes |
|------|-------|
| `filename.ext` | What it is, key amounts, period, status. |

---

## 2. <Next category>
...

---

## vysledek/ directory (OUTPUT — not source documents)
...

---

## Extra files not referenced in notes
...

---

## Issues / Missing documents

| Issue | Detail |
|-------|--------|
| ⚠️ ... | ... |
| ❓ ... | ... |
| ℹ️ ... | ... |

Formatting rules:

• Use ✓ when a file clearly matches a pozn category.
• Use ⚠️ for failures, missing files, or password issues.
• Use ❓ for ambiguous/unresolved items.
• Use ℹ️ for informational notes (duplicates, context-only documents).
• Include key figures (income amounts, interest, tax withheld) directly in the Notes column when readable — they allow quick cross-checking without reopening files.
• If OCR quality is poor, say so and recommend checking the raw file.

8. Confirm completion

Report back to the user:

• Total files processed.
• Number mapped to pozn categories.
• Number of extra/unreferenced files.
• Number of issues found (failures, missing docs, ambiguous).
• Path to the written source-map.md.

Reference example

A completed example for tax year 2024 exists at: examples-cases-data/txtified/2024-2025/source-map.md

Use it as a reference for output format and depth of analysis expected.

Notes on Czech tax document types

Common document types you will encounter: