Oratis's picture
保全103个空事件流逻辑别名,逐源实际哈希核验后清理重复对象
7716cf5 verified
|
Raw History Blame Contribute Delete
4.93 kB
---
pretty_name: Diogenes Legacy Capture Metadata Archive v01
license: other
license_name: original-capture-archive-terms
license_link: https://huggingface.co/datasets/DiogenesLab/Diogenes_Legacy_Capture_Metadata_Archive_v01/blob/main/NOTICE.md
tags:
- metadata
- archival
- non-trainable
size_categories:
- 1K<n<10K
---
# Diogenes Legacy Capture Metadata Archive v01
The project has closed. This owner-authorized archive preserves **9,514 metadata/text objects / 2,123,672,551 public bytes** across 43 historical source groups: **7,793 unchanged originals and 1,721 explicit privacy-redacted copies**. **All 50 TAR shards are published and verified.** Every intended member appears in the verified receipts; `ARCHIVE_INDEX.json` records each immutable shard revision, byte size and SHA-256. The corresponding R2 source objects and staged copies have been removed after verification, while sensitive originals remain in the private audit backup.
The source material includes GTA multi-camera scene/camera/frame-event evidence, raw-capture profiles, timing and pose records, depth/telemetry experiments, QC results, manifests, original READMEs and operational provenance. It preserves failed and incomplete attempts rather than promoting them to accepted data. Referenced media and code are outside this metadata-only scope and may live in separate archive collections. The 19-game sample metadata, the complete raw-depth indices already packaged with their frames, downloaded dependency/compiler caches and previously removed extracted game-asset files are excluded from this manifest.
## Finding files
`MANIFEST.jsonl` records each public archive path, original logical source path, SHA-256, byte size and representation. Original documents are stored under `metadata/original/<opaque-id>.<extension>`. Changed documents are under `public-copies/metadata/<opaque-id>.<extension>`. These stable IDs avoid platform-specific, Unicode or overly long archive paths; use `original_logical_path` to reconstruct the historical document layout. A path inside an old document may describe the historical capture machine or a separate archive, so it must not be assumed to resolve inside this repository.
For each changed document the manifest separately lists its original byte count and SHA-256, public-copy digest and change categories. The original sensitive document is retained privately for audit and is not published here. Removing private text does not upgrade original quality gates. See [NOTICE.md](NOTICE.md) before interpreting an embedded source hash, signature or authorization statement.
## Format and review
JSON, JSONL, compressed JSON/JSONL, Markdown, plain text, CSV and original configuration formats are retained. Zstd/gzip documents keep their compression format. Historical UTF-16 or other encodings remain in unchanged originals; explicit text copies declare their public byte hashes. Arrays and nullable/absent fields remain meaningful and must not be replaced with invented defaults.
All source metadata was read completely, including every compressed JSONL line. The 1,721 public copies were independently re-read and checked: 1,490 JSON documents, 89 JSONL documents containing 233,661 rows, and 142 text documents. Their JSON container structure and every numeric/boolean/null value match the private originals. Empty `adapter_device_ids` arrays are static configuration filters rather than personal hardware identifiers. Review included privacy/credential patterns and schema/field inspection; it is not a new formal data-quality certification.
One original 179-line `issue-559-ets2-20260806/calibration/telemetry-smoke.jsonl` is non-standard or malformed JSONL. It is preserved byte-for-byte as diagnostic text and marked in the public manifest. No parse repair, numeric replacement or successful-test claim has been fabricated.
## Integrity and historical use
TAR shards retain complete documents; no file is truncated or split. Each tar ends with per-member SHA-256 and size. Source bytes and staged public-copy bytes are hashed before transfer, then pinned Hugging Face LFS hashes/sizes and anonymous downloads are checked before original storage deletion. `ARCHIVE_INDEX.json` pins all 50 completed shards.
Old scripts, endpoints, operational instructions and acceptance gates in these documents are historical context for a closed project. They should not be interpreted as instructions to recreate paid resources or as evidence that those services still exist. Capture implementation: [Diogenes-Archive](https://github.com/CuddlerAI/Diogenes-Archive).
The [historical weekly supplement](closeout-final/weekly/README.md) preserves the final reviewed source objects with their original limitations and exact member hashes.
The [empty-event alias map](closeout-final/empty-event-aliases/README.md) preserves 103 exact historical source paths and their original empty Zstandard event stream.