Download docs/VALIDATION_RECORD.md from grujdin/DISFOR-S1-Companion: direct link, hf CLI and curl.
- Browser
- Download file 6.38 kB
-
https://huggingface.co/datasets/grujdin/DISFOR-S1-Companion/resolve/main/docs/VALIDATION_RECORD.md
- Command line
-
hf download hf://datasets/grujdin/DISFOR-S1-Companion/docs/VALIDATION_RECORD.md
-
curl -L -o VALIDATION_RECORD.md https://huggingface.co/datasets/grujdin/DISFOR-S1-Companion/resolve/main/docs/VALIDATION_RECORD.md
Validation record — uploaded baseline for version 1.0.0
The full uploaded DISFOR Sentinel-1 Companion was downloaded into a separate directory and verified on 11 November 2026. This record describes that immutable baseline and distinguishes full file-integrity checks from selected reader checks and ecological validation.
Exact revision validated
| Item | Value |
|---|---|
| Repository | grujdin/DISFOR-S1-Companion |
| Validated commit | 153754644fb3e49d74a1b8176a884dd16e19310c |
| Payload baseline release ID | disfor-s1-full-cc88440e3e4fc1a5-615c5840909f |
| Metadata revision at that commit | metadata-ef6f7cff641c2be68cdaf2d0 |
| Verification completed, UTC | 2026-11-11T08:14:40+00:00 |
| Equivalent Romanian local time | 11 November 2026, 10:14:40 |
| Verification result | validated_private_revision |
| Files verified | 330 |
| Manifest entries verified | 328 |
| Bytes verified | 62,372,814,745 — 62.37 GB / 58.09 GiB |
| Radar archives | 58 |
| Terrain archives | 1 |
The immutable baseline files are identified by the full commit. Validation did not use a moving branch name. The repository was reported as private during upload and download; this is a historical access-state observation, not a statement about its current visibility.
Baseline integrity controls
These SHA-256 values identify the controls at the validated baseline commit. Later publication metadata revisions have their own control hashes.
| Baseline control | SHA-256 |
|---|---|
release.json |
13f2c30e37922d10daec968389ce9a7e7e389683e96668d9cccca354ac2b33b2 |
checksums.sha256 |
272792c5f622e34c8dfec513a01c9f13750e6a6f4d7ca433f10bf7f628d23980 |
RELEASE_COMPLETE.json |
80de10e344e9f6ce8a87c2adda402fb2a1b8f70d94f42affee74ceab53e61614 |
The manifest covers 328 files. The checksum list itself and the completion record are the two additional verified controls, giving 330 files in total. All expected file identities were checked against the finalized release contract, and the inventory contained the exact expected files after excluding download and Python caches.
Checks performed
Upload and download identity
The upload used explicit parent commits and checked remote file identities after each batch. It completed in 28 recorded commits, with RELEASE_COMPLETE.json uploaded separately after every other release file matched. Remote checks used Git blob identities for ordinary files and SHA-256 identities for LFS/Xet files.
The resulting full commit was then downloaded as a separate copy. Repository identity, private status, file inventory and total byte size were checked before transfer. The subsequent verification read this downloaded copy, rather than treating successful upload or download alone as evidence that every file hash had passed.
Full file checksums
All 330 release files were read through SHA-256 and matched their expected values. This covered every byte of the radar and terrain archives, tables, original metadata, split files, reader code, documentation and release controls. Reads were streamed in bounded chunks; image arrays were not accumulated in memory for this check.
These checks establish byte identity with the finalized release contract. They do not independently validate the scientific interpretations represented by those bytes.
Index totals and reader checks
The annual-block index was traversed and its totals matched:
| Quantity | Verified total |
|---|---|
| Annual sample/track blocks | 136,755 |
| Location-pass observation frames | 5,507,831 |
| Accepted centre observations | 5,240,543 |
| Radar archives represented by the index | 58 |
The included reader was imported only after all release file hashes had passed. For every radar archive, the annual blocks at the first and last indexed TAR byte offsets were selected. All 116 distinct block reads passed the reader's payload-hash and structure checks. The checks exercised indexed TAR access and annual NPZ loading across all 58 radar shards.
This was not a second decoding of all 136,755 annual NPZ blocks. Every exported block had already been decoded and checked during construction. The post-transfer stage combined full archive-file hashes with selected reader access at both ends of every radar archive. File fingerprints were checked again after the reader tests and remained unchanged throughout validation.
Relationship to the current metadata revision
The baseline commit above fixes the scientific payload and the publication metadata that were fully downloaded and verified. Later publication metadata changes document that outcome and supply the current publication record. They preserve the existing hashes of imagery, indices, original catalogue and annotations, quality information, splits and reader files.
The baseline's 330-file count and 62,372,814,745-byte total are therefore historical values for the identified commit. They must not be used as the expected current totals after adding documentation or regenerating controls. Use the current manifest, release description and completion record together. The publication record explains the relationship between the build, validated baseline and metadata revisions.
Scientific limits
Successful validation establishes technical consistency and preservation through transfer. It does not independently verify disturbance causes or onset dates, convert interval annotations into full-patch labels, establish physiological forest health, or remove seasonal and terrain limitations of the source observations.
The source audit and construction checks were tied to the cached acquisition inventories and consumed source products. They did not establish that those inventories include every globally available Sentinel-1 scene. The original Sentinel-2 image archive was not re-audited by this radar-transfer workflow.
No model training or model test-set evaluation occurred during upload, download or validation. Detector performance, reference construction and attribution accuracy require separately defined experiments using an identified dataset revision and split.