Recorded
An attributable manifest contains the claim. The record can be inspected and corrected; its presence does not make the claim legally conclusive.
The AI Bill of Materials.
A public corpus
with receipts.
01 / CORPUS
OpenWALDO turns a resolved corpus selection into an immutable BOM: index revision, manifests, sources, license assertions, canonical shards, exact counts, and hashes—inspectable before training and carried into the models built from it.
The public index currently records — corpora across — canonical shards—— documents and — reference tokens under — asserted license identifiers.
Why use OpenWALDO
Every model needs
a bill of materials.
OpenWALDO starts that record with the corpus. A resolved selection pins the exact data meaning and object identity, then crosses the boundary into exports and training runs so provenance does not disappear when data becomes weights.
01 / ORIGIN
Trace every corpus
Source provenance
Trace every corpus
to its source.
OpenWALDO records where the material came from, which upstream object or collection was acquired, and which version was used. Reviewers get a concrete origin trail instead of relying on a dataset title alone.
Each claim is attributable and correctable. Missing evidence remains visible, helping teams make informed decisions about what they use.
OPENWALDO BOM LIVE INDEX
- Revision
- Loading…
- Selection
- Loading live index…
- Objects
- —
- Contents
- —
- Assertions
- —
Current complete public-index selection · Loading live index…
02 / RIGHTS
See the terms
License accountability
See the terms
before you train.
OpenWALDO carries asserted license identifiers with the corpus inventory, helping teams understand declared terms before selecting data. When a corpus contains multiple assertions, their scope remains visible instead of being flattened into “mixed.”
Each assertion is attributable, reviewable, and correctable in public. It informs responsible selection without pretending to be a legal opinion.
ASSERTIONNamed license identifier
Recorded as a claim, not silently promoted to fact.
SCOPECorpus and shard inventory
Counts show which recorded objects the assertion covers.
OPTIMIZATIONPublic review path
Disagreement becomes a change to the record, not private metadata.
03 / IDENTITY
Know exactly
Reproducible objects
Know exactly
the contents of the model.
WALDO records the converter identity, canonical format, exact document and reference-token counts, encoded size, and content-addressed shard inventory. A readable corpus record is therefore bound to specific bytes.
Teams can reproduce the selection and verify its identity and integrity. License, quality, safety, and fitness remain separate questions with their own evidence.
-
01
CONVERSIONNamed tool and format
-
02
CANONICAL OBJECTSSelf-contained Parquet shards
-
03
CONTENT IDENTITYSHA-256 object references
-
04
EXACT INVENTORYDocuments · tokens · bytes
04 / HISTORY
Accountability needs
Public governance
Accountability needs
a place to live.
Reviewable meaning stays in Git: descriptions, source claims, asserted licenses, conversion identity, counts, and object references. Large canonical bytes stay in lookaside storage.
Every accepted contribution retains its public diff, review, and Developer Certificate of Origin sign-off. The DCO makes responsibility attributable; it does not turn a rights assertion into legal proof. Corrections become part of the same durable record.
GIT / INDEX
- Source and license claims
- Conversion identity
- Counts and object references
- Diff, review, and correction history
LOOKASIDE / DATA
- Canonical Parquet objects
- Content-addressed identity
- Independent retrieval
- No private catalog required
Read the evidence precisely
Recorded.
Identified.
Verified.
OpenWALDO does not collapse different kinds of evidence into a trust score. Each word has a narrower meaning—and visible limits.
Identified
A content hash binds the record to specific bytes. Identity and integrity do not establish rights, quality, or safety.
Verified
A named machine check passed against a defined selection. Verification proves only what that check actually examined.
Choose your proof
Inspect lightly.
Audit completely.
WALDO separates quick structural checks from full object and record validation, so the cost of verification is explicit.
Verify metadata
Validate index structure and check that referenced objects are reachable with their declared sizes.
waldo index verify core/books
Verify objects
Download selected objects and prove every content hash.
waldo index verify core/books --objects
Audit records
Validate canonical records, identities, text hashes, duplicates, and manifest totals.
waldo index audit core/books