Process and review the complete five-document pilot corpus #18

New issue

Open

opened 2026-07-24 19:38:18 -04:00 by JackFrostbyte · 0 comments

JackFrostbyte commented

2026-07-24 19:38:18 -04:00

Owner

Goal

    Regenerate, validate, and manually review the complete five-document pilot corpus, then record the evidence that Milestone 1 is complete.

    ## Why

    Automated reproducibility is necessary but does not replace human review of provenance, license evidence, extracted text, transformation lineage, chunk boundaries, and citation locations.

    ## Scope

    - Run the documented clean-checkout processing workflow for all five approved documents.

Review every source record and its provenance, license evidence, acceptance state, and bundle eligibility.
Review original file identity, byte size, and SHA-256.
Review extracted and normalized text against the originals.
Review transformation lineage.
Review generated chunk text, ordering, boundaries, half-open offsets, and available page, section, or heading data.
Record validator and test results.
Correct defects through this focused corpus-review pull request only when they are data or fixture corrections; create a separate implementation issue for substantive pipeline defects.
Add a concise milestone completion record or report in the repository.
```
  ## Out of scope

  - Adding a sixth pilot document.
```
OCR or safety-critical content.
Search index generation.
Embeddings, LLMs, translation, or speech.

A production public release.

  ## Dependencies

  - #10 — Select and approve the five-document pilot corpus

#17 — Add clean-checkout regeneration and reproducibility tests

  ## Acceptance criteria

  - All five original files and records are present under the approved layout.

Every source is legally approved and bundleable.
All source, document, and chunk records validate.
All file sizes, SHA-256 values, lineage links, chunk texts, and offsets are verified.
A clean checkout regenerates the committed output.
Human review finds no silent extraction or normalization loss.

The milestone completion record confirms that every Milestone 1 criterion is satisfied.

  ## Validation

  - Run the complete clean-checkout regeneration and unit-test workflow.

Run schema and cross-record validation over all committed pilot records.
Manually compare representative and boundary passages from every document to the original.
Recalculate all five checksums.

Review the milestone completion record against the milestone description.

  ## Suggested branch

  `feature/issue-18-pilot-corpus-review`

  <!-- arkive-planning-automation:2026-07-24-v3:M1-13 -->

## Goal Regenerate, validate, and manually review the complete five-document pilot corpus, then record the evidence that Milestone 1 is complete. ## Why Automated reproducibility is necessary but does not replace human review of provenance, license evidence, extracted text, transformation lineage, chunk boundaries, and citation locations. ## Scope - Run the documented clean-checkout processing workflow for all five approved documents. - Review every source record and its provenance, license evidence, acceptance state, and bundle eligibility. - Review original file identity, byte size, and SHA-256. - Review extracted and normalized text against the originals. - Review transformation lineage. - Review generated chunk text, ordering, boundaries, half-open offsets, and available page, section, or heading data. - Record validator and test results. - Correct defects through this focused corpus-review pull request only when they are data or fixture corrections; create a separate implementation issue for substantive pipeline defects. - Add a concise milestone completion record or report in the repository. ## Out of scope - Adding a sixth pilot document. - OCR or safety-critical content. - Search index generation. - Embeddings, LLMs, translation, or speech. - A production public release. ## Dependencies - #10 — Select and approve the five-document pilot corpus - #17 — Add clean-checkout regeneration and reproducibility tests ## Acceptance criteria - All five original files and records are present under the approved layout. - Every source is legally approved and bundleable. - All source, document, and chunk records validate. - All file sizes, SHA-256 values, lineage links, chunk texts, and offsets are verified. - A clean checkout regenerates the committed output. - Human review finds no silent extraction or normalization loss. - The milestone completion record confirms that every Milestone 1 criterion is satisfied. ## Validation - Run the complete clean-checkout regeneration and unit-test workflow. - Run schema and cross-record validation over all committed pilot records. - Manually compare representative and boundary passages from every document to the original. - Recalculate all five checksums. - Review the milestone completion record against the milestone description. ## Suggested branch `feature/issue-18-pilot-corpus-review`

JackFrostbyte added this to the Reproducible Acquisition & Processing v0.1 milestone

2026-07-24 19:38:18 -04:00

JackFrostbyte added the

type/data

area/metadata

labels

2026-07-24 19:38:18 -04:00

JackFrostbyte added a new dependency

2026-07-24 20:10:49 -04:00

#10 Select and approve the five-document pilot corpus

JackFrostbyte added a new dependency

2026-07-24 20:10:49 -04:00