# PDR 004: Review imported recipes before publication

- HTML version: https://robbiepalmer.me/projects/recipe-site/pdrs/004-review-imported-recipes-before-publication
- Project: Recipe Site
- Status: Accepted
- Opened: 2026-07-08
- Decided: 2026-07-08

## Context

Recipe imports can begin with a structured file, a web page, photographs, or
model output. Even a plausible result may omit an instruction, misread a
quantity, join two source pages incorrectly, or infer the wrong ingredient.
Publishing that result immediately would put uncertain content into recipe
boxes and public pages while hiding what the importer actually produced.

Import processing may continue after the browser closes. Durable background
execution solves that operational problem, but completion of a job is not the
same as a user's decision to keep the recipe.

## Decision

Every import produces an editable draft before it can create a canonical
recipe. The user reviews that draft and explicitly accepts it. Successful
parsing, a confidence score, a known archive format, or a shared batch default
cannot replace item-level acceptance.

Keep the generated result and its source provenance intact while the user
edits a separate draft. Record the accepted recipe and the correction from the
generated result so the ingestion pipeline can be evaluated without treating
model output as ground truth.

An unfinished draft stays outside the normal recipe box and outside public
discovery. Retry, skip, reject, and cancellation do not create a canonical
recipe. Import agents may prepare a draft, but publication remains an explicit
human action.

## Consequences

Imports take an extra review step, including sources that usually parse cleanly.
That cost is deliberate. It prevents uncertain content from looking finished
and gives users one place to correct title, ingredients, quantities,
instructions, attribution, and audience before publication.

The system must preserve immutable generated evidence separately from mutable
editor state. It also needs durable drafts so review survives navigation,
restarts, and long-running imports.

## Evaluation

Tests must prove that no import path creates a canonical recipe before an
authenticated acceptance action. A reviewer must be able to compare the source,
generated draft, edited draft, and accepted recipe. The accepted record must
retain enough provenance to explain its source and later undo a safe migration.

## Migration note

This record extracts the review boundary first published with ADR 012. ADR 024
later applied the same policy to collection and photo batches. The migration
preserves the original date, Accepted status, and explicit-review requirement.
It does not make any proposed batch-import architecture Accepted.

## Related records

- Project: [Recipe Site](https://robbiepalmer.me/projects/recipe-site.md)
- Implemented by ADR: [ADR 012: Cloudflare Workflows for Recipe Ingestion](https://robbiepalmer.me/projects/recipe-site/adrs/012-cloudflare-workflows-recipe-ingestion.md)
- Implemented by ADR: [ADR 024: Batch Recipe Import Staging](https://robbiepalmer.me/projects/recipe-site/adrs/024-batch-recipe-import-staging.md)

---

Markdown index of this site: https://robbiepalmer.me/llms.txt