Codex plugin
Journal Sampling
Turn variable accounting journals into reproducible audit samples.
Codex handles format ambiguity, column mapping and review questions. Deterministic Python scripts extract, normalize, sample and preserve the audit trail.
What it does
From mixed journals to a sample with diagnostics
Use it when a journal is readable but not standardized: good Excel, CSV, print-style Excel exported from PDF, or text PDF. Codex guides only the uncertain parts.
input/
journal.xlsx
print_export_from_pdf.xlsx
general_journal.pdf
out/
inspection.json
suggested_recipe.json
normalized_journal.csv
normalization_diagnostics.json
sample/
journal_sample.csv
journal_sample.xlsx
sampling_audit.json
Inputs
What it reads
The plugin works with local files already available to the firm. OCR-only scanned PDFs are not a v1 target.
Outputs
What it produces
The result is a reviewable work folder: mapped recipe, normalized population, sample files, diagnostics and audit trail.
Codex is the adaptive layer, not the sampling engine
Codex inspects the customer format, proposes a mapping and asks only the smallest unresolved question: header row, account column, debit/credit convention or sample assumptions.
The scripts remain responsible for extraction, normalization, filtering, sampling and exports. If a parser misses a pattern, the improvement is a new deterministic rule or script, not a silent model correction.
At the end of a run Codex can write a short improvement note and ask before emailing it to the maintainer.
Use
How it runs in Codex
Open the work folder in Codex, call the plugin, and let Codex inspect the journal before running the deterministic scripts.
Prompts
Starter prompts for beta users
Start from one of these prompts and replace the folder path, language, method and sample size. Codex will ask before resolving essential ambiguities.
Download
Download the plugin
Requires Codex desktop. The same ZIP installs Plugins4Accountants.
First use
In a Codex chat, provide the journal folder, language, sampling method and target size.
Use Journal Sampling on /path/input.
Language: en. Document language: auto.
Inspect the files, ask only for essential mapping ambiguities, then generate sample, diagnostics and audit trail.
Method: random. Size: 25.