Recursive Integrity Toolkit v0.1.0

Audit synthetic training data for missing categories, provenance gaps and concentrated ancestry. Recursive Integrity Toolkit is a local Python CLI that reads CSV, JSONL or optional Parquet and produces inspectable JSON and Markdown reports. Runtime needs no API key, model download, telemetry or data upload.

Start here

Use Python 3.11 or 3.12 on Linux or Windows. In a virtual environment, install the wheel attached below:

python -m pip install ./recursive_integrity_toolkit-0.1.0-py3-none-any.whl
rit example --lineage --out ./rit-demo

Open rit-demo/reports/report.md. The included example shows topic coverage falling from 8 categories to 5, diversity from 0.875 to 0.750, and identifies the missing topics and shared ancestry.

For tagged-source installation and more workflows, see the README and full walkthrough. Package-registry publication is outside this release.

Included

• Input-only validation, declarative mapping and dataset audits.
• Declared categorical support/diversity, exact duplicates and tail diagnostics.
• Provenance coverage and direct closure intervals that preserve unknown evidence.
• Supplied lineage graphs, unresolved ancestry, external roots and ancestry concentration.
• Explicit ordered dataset comparisons, including adjacent and optional baseline pairs.
• Seeded closed and constant-source reopened categorical simulations, with assumptions and experimental labels.
• Standard/redacted JSON and Markdown, safe no-overwrite output and packaged examples.

These are research diagnostics for supplied records and declarations. There is no universal integrity or collapse score, model-performance certification or inferred source truth. Empirical causal-intervention ingestion remains deferred. Simulations remain experimental; large complete reports can require several GB of RAM. The existing package development classifier is unchanged.

Official identity and validation

• Release label: official-v0.1; toolkit 0.1.0; report schema 1.3.
• Tag target: 24f46ac375dafe4ee5c407d9b4b08dc4c77914ee (merge of PR #5); tree d7bb2e3984b2838b8fe201b4e798a1b3ef1fe846.
• Package build source: fec89a7ed4da864e97ca1a967c85580bbbdf085a; SOURCE_DATE_EPOCH=1791320488.
• Full runtime evidence source: c5837845da603167e6f0c8216766d0a3a874835e. Candidate run 36976926784 passed all 13 required jobs, including supported core environments, real Parquet, Hero/math, security, all 21 performance cases and installed delivery.
• Publication changes preserve runtime, dependencies, schemas and resources. Two clean builds produce identical wheel bytes and matching sdist payloads; strict distribution and installed wheel/sdist checks pass. Final documentation-only successor and merge preserve every packaged source payload. Full runtime evidence is reused with its original identity.
• Xiangyu Guo accepted the handoff and authorized this merge and release on 2026-10-06 as Theory Owner and acting Technical Maintainer. Maintainers, accepted handoff, validation record.
• Theory-source versions, licensing notes, code license, and third-party notices.

Downloads and checksums

The full source ZIP contains all 358 tracked files from the tag target, including schemas, examples, tests, licenses and validation records. Wheel and sdist have their normal installation scopes. These refreshed packages supersede earlier preparation builds.

File	Bytes	SHA-256
recursive-integrity-toolkit-v0.1.0-source.zip	1803648	803259cdf5d501a93aecdc3b0a8127bc1a0c1d608bb607af514c18916838ce10
recursive_integrity_toolkit-0.1.0-py3-none-any.whl	389517	9d114cd296054d91fb59627d3378ba6271a6b20fd9efc1f6c19fc2044b1552fa
recursive_integrity_toolkit-0.1.0.tar.gz	354650	4a8c3590a3d979836276f51b97c441c70413250d9af1b1d1087937bd20de3296

Verify the three attached archives with sha256sum -c SHA256SUMS.txt.