Download docs/VALIDATION.md from maxwellinked/time-lapse-artifacts: direct link, hf CLI and curl.
- Browser
- Download file 70.1 kB
-
https://huggingface.co/datasets/maxwellinked/time-lapse-artifacts/resolve/main/docs/VALIDATION.md
- Command line
-
hf download hf://datasets/maxwellinked/time-lapse-artifacts/docs/VALIDATION.md
-
curl -L -o VALIDATION.md https://huggingface.co/datasets/maxwellinked/time-lapse-artifacts/resolve/main/docs/VALIDATION.md
Validation Report
September 26–October 2 complete-drawing ingest — current candidate
The reviewed parent is 51d40de8f8facde64b1c4baa92729b7103784076. This candidate adds 116 unchanged-source silent HEVC MP4s in 15 owner-confirmed complete drawing sequences. Original source filenames and SHA-256 mappings are retained before canonical finish-label assignment.
| Check | Candidate |
|---|---|
| Standard / Pre-Standard / indexed videos | 1,039 / 128 / 1,167 |
| Earlier / prospective Standard rows | 364 / 675 |
| Active video bytes / encoded playback | 2,305,938,893,860 / 244891.269533 seconds |
| Dated fragment packet rows | 680 across 25 packets: 675 canonical and 5 historical derivatives |
| Confirmed drawing groups in dated packets | 73 across 23 calendar dates |
| Work registry / linked / unresolved indexed records | 416 / 1,039 / 128 |
| Published payloads requiring SHA256SUMS coverage | 1,172 MP4 + 1 ZIP = 1,173 |
| New source checks | 116 distinct SHA-256 values; 116 stable-source checks |
| New decode checks | 116 full plus 348 start/middle/end samples; 257,701 declared/stored packets |
The current timing view covers 675 rows, including 100 new filename-derived starts and 16 new unknown starts. The frozen 339-row Timing v2 snapshot and earlier September 26 packet remain unchanged. Prior rows, historical receipts, releases, and DOI snapshots are preserved.
September 21–26 complete-drawing ingest — current candidate
The reviewed parent is 22f6f36bed668f090c260ffadd7b9e8f60e0fe79. This candidate adds 178 unchanged-source silent HEVC MP4s in 21 owner-confirmed completed drawings. One malformed source name is preserved verbatim and has only the owner's narrow Rec18 membership correction.
| Check | Candidate |
|---|---|
| Standard / Pre-Standard / indexed videos | 923 / 128 / 1,051 |
| Earlier / prospective Standard rows | 364 / 559 |
| Active video bytes / encoded playback | 2,261,315,023,049 / 236,301.323005 seconds |
| Dated fragment packet rows | 564 across 20 packets: 559 canonical and 5 historical derivatives |
| Confirmed drawing groups in dated packets | 58 across 19 calendar dates |
| Work registry / linked / unresolved indexed records | 401 / 923 / 128 |
| Published payloads requiring SHA256SUMS coverage | 1,056 MP4 + 1 ZIP = 1,057 |
| New source checks | 178 distinct SHA-256 values; 178 stable-source checks |
| New decode checks | 178 full plus 534 start/middle/end samples; 377,788 declared/stored packets |
The current timing view covers 559 rows, including 156 new filename-derived starts and 22 new unknown starts. The frozen 339-row Timing v2 snapshot remains unchanged. Prior rows, historical receipts, releases, and DOI snapshots are preserved.
September 20 two-drawing ingest — current candidate
The reviewed parent is dataset revision
26ec194cd88ff5551777b3c637a70bcdfb55017f. Twenty-eight
unchanged-source MP4s form two creator-confirmed completed drawings:
twelve with the 07:25 BallpointPen finish label and sixteen with the
17:38 FountainPen finish label. The dated provenance packet
preserves exact source order and separates filename/device and MP4
creation observations from drawing finish labels.
| Check | September 20 candidate |
|---|---|
| Standard / Pre-Standard / indexed videos | 745 / 128 / 873 |
| Earlier / prospective Standard rows | 364 / 381 |
| Active video bytes / encoded playback | 2,196,054,134,482 / 223,708.517438 seconds |
| Dated fragment packet rows | 386 across 14 packets: 381 canonical and 5 historical derivatives |
| Confirmed drawing groups in dated packets | 37 across 13 calendar dates |
| Work registry / linked / unresolved indexed records | 380 / 745 / 128 |
| Published payloads requiring SHA256SUMS coverage | 878 MP4 + 1 ZIP = 879 |
| New source checks | 28 distinct SHA-256 values; 28 stable-source checks |
| New media checks | 28 full and 28 start/middle/end sampled decodes; 53,898 declared frames and stored packets |
Decoded frame counts were not independently measured. The living current timing view covers 381 rows, including 26 reviewed Sep20 filename-derived starts and two unavailable starts. The frozen 339-row Timing v2 snapshot remains unchanged. Remote Xet/size/index validation and hosted Viewer status are independently recorded at the exact PR and merged revisions; a pending derived Viewer cache is not a failure of byte-level repository integrity. Prior rows, historical receipts, tagged releases, and DOI snapshots are preserved.
Standard workflow selection access
Candidate parent: c3e959c8c7b4f60f19b34c5501b19459269d53a5.
The supplied workflow implementation was applied to a complete checkout of
that revision with Git LFS downloading disabled. Generated access indexes
select 364 standard_earlier and 353 standard_individual_recordings rows.
They are disjoint and together reproduce the 717-row canonical index.
The individual-recording selection includes all 175 known and 178 unknown
starts; unresolved work identities also remain included.
All 150 repository tests passed with no skips, including the full current
repository-state check and native loading of the named configurations with
datasets==4.8.5. Native loading verified exact resolved input files,
selected record IDs, every metadata field and video reference, with zero MP4
payload reads. The original 35-field schema, canonical default, and both
original configuration definitions remain unchanged. Repeated generation
reproduces identical output bytes.
The integration adds explicit UTF-8/LF fixture writes and a Windows junction for the real path-escape test, avoiding platform-specific fixture failures. It also tests full-repository native loading within the regression suite and checks the loader's resolved input paths. These extend the supplied fixture tests; they do not replace them.
The existing maintainer ingest procedure now requires regeneration after identity/timing projections and the normal count update, plus a drift/loading check before publication. There is no automated ingest runner in this repository. Source indexes, media, checksum manifest, work registry data, relationship/timing evidence, and frozen release files are unchanged; only current maintainer instructions in the timing-view README are extended.
This records local candidate validation, not publication or hosted Viewer readiness. The hosted four-configuration counts, sample IDs, and revision headers must be verified separately after a reviewed merge.
Current recording-time access extension
This extension is based on 4cc1dcd0490f150b3761765a40df522750ecd9d9.
It exports 14 retained September 19 observations and combines them with the
unchanged 339-row Timing v2 snapshot. The current view has 353 records,
175 filename-derived starts and 178 unknown starts. The two new unavailable
starts remain blank. The 492 records outside reviewed timing coverage retain
the existing not_in_timing_snapshot compatibility value.
The current-view contract describes the evidence receipts, timezone interpretation and repeatable ingest commands. Local checks passed: 116 repository tests; exact reproduction of the 14 selected observations from the retained audit; all six ingest input receipts; all 16 original Timing v2 input receipts; Timing v1 reconstruction; metadata and protocol-boundary validation; and all 851 published payload checksum entries against the pinned Hub inventory. Header-only checks verified the Xet identities of all 850 published MP4s without downloading their bodies.
The extension changes exactly 26 index cells across 14 records: 12 starts and 14 start-basis values. All 339 original Timing v2 rows retain all 21 values; both frozen timing packets, all other index values, the Pre-Standard index, work registry, relationship packets and SHA256SUMS remain byte-identical. Hosted Viewer processing is verified separately from repository content. The historical receipts below describe their named revisions and remain unchanged.
September 19 two-drawing ingest
The reviewed parent is dataset revision
5a30d310c7be0ec3fee6d00e7cd7d189a14a9b5d. Fourteen audited
unchanged-source MP4s form two creator-supplied complete drawing sets of nine
and five recordings. The exact source inventory is retained locally; the dated
provenance packet
records owner wording, ordered filenames, timing distinctions, and the
revision-binding requirements.
| Check | September 19 candidate result |
|---|---|
| Standard / Pre-Standard / indexed videos | 717 / 128 / 845 |
| Earlier / prospective Standard protocol rows | 364 / 353 |
| Active video bytes / encoded playback | 2,186,833,793,374 / 221,911.935724 seconds |
| Dated fragment packet rows | 358 across 13 packets: 353 canonical and 5 prior noncanonical derivatives |
| Confirmed drawing groups in dated packets | 35 across 12 calendar dates |
| Work registry / linked records / unresolved indexed records | 39 / 374 / 471 |
| Published payloads requiring SHA256SUMS coverage | 850 MP4 + 1 ZIP = 851 |
| New source byte-level checks | 14 unique SHA-256 values; 14 stable-source checks |
| New media checks | 14 of 14 full decodes passed; 58,350 declared frames and 58,350 stored packets |
| Remote Xet/size/metadata revision | Verified 14/14 new MP4 identities and sizes at 038c10d93d939822725fcac81903beabd529c2ad; metadata and checksum receipts at b7d446c35b5b31668f36b7c3d25fa3269983f783 |
| Final repository validators and hosted Viewer | 105/105 repository tests passed; hosted Viewer pending at final check (derived cache, not an ingest fixity gate) |
Decoded frame counts were not independently measured for this ingest, so the
declared/stored packet equality is not represented as a decoded-frame count.
The 12 literal VID_* source names provide device-clock start observations;
all 14 MP4s carry creation/finalization observations. Neither substitutes for
an individual recording stop. The current 339-row Recording Timing v2
observation table is intentionally frozen; all 14 new index rows receive
not_in_timing_snapshot until a separately reviewed timing extension.
The remote-binding and repository checks above must be verified before publication. Hosted Viewer processing is a separate derived-cache status: a pending Viewer is reported as pending, not as an ingest failure or a completed Viewer verification. No prior metadata row, historical sidecar, v0.7.0 release receipt, tag, or DOI-backed v0.5.1 snapshot is changed by this append.
Timing display extension (2026-09-19; v0.7.0 snapshot)
At the v0.7.0 preparation snapshot, the five explicit timing columns were projected from the existing indexes and Recording Timing v2. All 831 original rows retain their original 29 field values and media paths. The local VideoFolder loading check reads 703 Standard and 128 Pre-Standard rows with the declared 34-field schema; all explicit timing values, nulls and video-path joins match. It uses temporary empty video placeholders with decoding disabled, so it neither downloads nor decodes media. This local check does not establish hosted Viewer cache readiness.
The projection provides 163 filename-derived starts, 176 unknown starts within the frozen timing snapshot and 492 records outside that snapshot. Both timing packets and all prior tagged releases retain their exact bytes. The v1 validator excludes only the five named display columns when reconstructing its historical 29-column snapshot, then still requires the original exact byte length and SHA-256. Undeclared or partial extensions are rejected.
All 95 repository tests pass. Timing v2 strict validation uses the retained checkout of its declared source revision, where all 16 input receipts match.
v0.7.0 release-preparation receipt
The validated content is the exact 960-path tree at
521f4ee62ee3843cf36394499c10bf33f07dcaae. Relative to the peeled v0.6.0 commit
4fa4c34ed87ecc35747668abcdd58c91650349d0, the raw tree delta is 137 added, 12 modified,
and 12 removed paths. Twelve added/removed media pairs are documented
identity-preserving renames; the net addition is 74 recordings and 51 small files.
All previous video payload bytes and record IDs are preserved. Exactly 24
existing metadata rows change: 12 documented path/time corrections and 12
commit-history provenance backfills. Pre-Standard metadata is byte-identical.
| Check | Verified result at the content revision |
|---|---|
| Standard / Pre-Standard / active rows | 703 / 128 / 831 |
| Active video bytes / encoded playback | 2,176,788,154,698 / 219,966.955344 seconds |
| Earlier / prospective Standard protocol rows | 364 / 339 |
| Fragment packets / canonical fragments / complete groups | 12 / 339 / 33 |
| Work registry / linked records / unlinked indexed records | 37 / 360 / 471 |
| Revision-pinned work-record joins | 360 verified across 10 source revisions |
| Work-identity evidence | 23 checksum-pinned confirmation files and 27 revision-pinned evidence files |
| Recording Timing v1 | 186 fragments / 25 groups; historical packet unchanged |
| Recording Timing v2 | 339 observations / 33 works; 163 filename starts / 176 unknown starts |
| Timing v2 creation tags | 339 format tags; 316 retained video tags, all matching |
| Timing v2 evidence | 13 exact local audit receipts; 16 pinned repository input files; all media identities matched |
| Physical-reference relationships | 309; checked against their declared source revision |
| Published payloads / SHA256SUMS coverage | 836 MP4 + 1 ZIP / 837 of 837 |
| Payload inventory bytes | 2,193,547,379,973 |
| Repository tests | 86 of 86 passed, including 12 new timing-observation tests |
| Prior v0.6.0 critical-file receipts | 71 of 71 matched its immutable snapshot |
Ordinary SHA-256 and byte-size bindings were checked against the remote LFS inventory, with fresh revision-bound Xet/ETag/size HEAD checks for all MP4s. Release preparation does not download, decode, or rehash the full media corpus; the original ingest receipts retain their local source and decode evidence. Dataset Viewer row and aggregate checks are recorded separately at the exact validated content revision in the release manifest. The Viewer is a derived current-main cache, not an immutable-tag fixity service.
The release draft adds Recording Timing v2 as five packet files and two tool/test
files, alongside the release documents. Supporting documents are grouped in
docs/ and all six release manifests in releases/; the root shrinks from 22
entries to 15. Two folder indexes provide navigation, active Markdown links are
updated, and the semantic consistency test reads the relocated specification.
All five prior manifest files retain their exact bytes. Old release tags and
revision-pinned URLs retain their original paths; current-main root URLs for
the moved documents and manifests change to the listed folder paths.
The revised manifest lists all 132 non-MP4/non-ZIP files at the final
content/documentation commit with exact byte sizes and SHA-256 values. That
commit excludes the draft's earlier manifest; the regenerated
releases/RELEASE_MANIFEST_V0.7.0.json is then added alone. The final release tree
must contain 970 paths and differ from the declared critical-files revision
only by that manifest. Post-merge/tag receipts bind the actual published revisions.
The timing table is derived from the literal retained audit fields. Every one of the 339 observations joins by existing record ID, ordinary SHA-256 and size; canonical metadata, sidecars, checksum entries and revision-bound remote Xet identities agree. All seven older audit hashes still match the v1 source manifest. Twelve September 1 path corrections resolve through stable IDs and fixity. The strict timing validation checks all 16 original input-file receipts and recomputes the table. Default future validation keeps this cohort frozen and reports input-file receipts separately, permitting unrelated later index additions. No video is re-read for this projection and no exact capture-stop claim is made.
The release preparation preserves every media object, active index, checksum
entry, work assignment, confirmation, historical sidecar, canonical schema,
previous release manifest, and existing tag. The DOI 10.57967/hf/10449 remains attached
to the historical 709-record v0.5.1 snapshot. Website publication is separate.
August 20 Rec02 ingest, reviewed September 19
One owner-confirmed historical Rec02 is added under its exact filename and
linked to the existing two-recording work tlaw-2f587c509d7e6909. All 131,442 video frames
and the audio stream decoded at the source time base with zero errors or warnings; no video frames were
dropped or duplicated. The uploaded 28,666,390,097 bytes match the fresh
local SHA-256 and pinned Hub LFS/size headers; the Xet identity is recorded separately.
An initial null-output tail timestamp warning was resolved by checking the
strictly increasing source packet/decoded-tail timestamps and repeating the
full decode at the source time base.
Metadata-bearing revision: b0383afea4ceefe8c67ba5f3fc31f076c8a67858. The active indexes contain
703 Standard and 128 Pre-Standard records (831 total); the registry has 37
drawings and 360 recording links. SHA256SUMS covers 837 payloads.
All prior metadata rows, media objects, record IDs, and historical sidecars are
preserved. Only the existing work's pending-ingest note is resolved.
The canonical CSV remains sorted by filename, with every prior row value preserved. The timing validator checks this later-ingested record by its ID, path, ordinary SHA-256, Xet identity, byte size, and protocol before excluding it from reconstruction of the original 549-row timing snapshot. That snapshot still matches its original byte size and SHA-256; the historical timing packet and its receipts are unchanged. All 22 timing regression tests pass, including rejection of identity/hash drift and unregistered historical backfills. All 74 repository regression tests pass, with the historical RecNN count updated from 19 to 20.
See the ingest evidence and limits and validation receipt. Release tags and the website mirror are unchanged.
2026-09-18 two-drawing ingest preparation receipt
The media-only PR inventory resolves to immutable revision e05a264107caa380f20a8bb036d0335ba73f01c9.
The 29 new paths match the accepted ordinary SHA-256 and byte-size inventory
and have 29 nonblank, unique Xet identities. The metadata-bearing inventory
resolves to d3e4da0bb5dd3eb573f561f588f4792b93fbd88b and contains the exact 702-row root index used
below. The later relationship sidecar pins that revision, avoiding circular
self-reference.
| Check | Result |
|---|---|
| New unchanged canonical fragments | 29 |
| Complete drawings / sequence sizes | 2 / 9 + 20 |
| New media bytes / encoded seconds | 17,385,403,635 / 3387.532426 |
| Local SHA-256 / source stability / sampled decodes | 29/29 / 29/29 / 87/87 |
| Full HEVC decodes | 29/29 |
| Declared / packet / presentation-read / decoded frames | 101,627 / 101,627 / 101,627 / 101,627 |
| Frame-count or decoder exceptions | none |
Root metadata.csv |
568,419 bytes; SHA-256 a5fe13ed7b62f1eaed6658303482b7fd155ab349878370e4f7a9414bcd7da914 |
| Canonical Standard / Pre-Standard / active rows | 702 / 128 / 830 |
| Earlier / prospective Standard protocol rows | 363 / 339 |
| Relationship packets / calendar dates / complete groups | 12 / 11 / 33 |
| New relationship rows / derivatives | 29 / 0 |
| Timing-v1 rows / dated source sidecars | 186 / 6 (unchanged) |
SHA256SUMS |
121,362 bytes; SHA-256 7d05a808b5013e50ddcb48413530359f416dd7bcb29690268feacac4ccb2a7a2 |
| Manifest payload coverage | 836 / 836; 2,164,880,989,876 bytes |
| Active indexed video bytes / encoded playback | 2,148,121,764,601 / 217776.243345 seconds |
| Active videos with audio | 485 / 830 (58.4%) |
| Expected pull-request paths | 42 = 29 media + 13 non-media |
All 29 files belong to exactly one of the two owner-confirmed complete drawing groups. The 27 literal filename starts, MP4 creation/finalization observations, filesystem timestamps, and supplied 02:44/14:44 group finishes remain distinct; none is substituted for another. No existing metadata row, media path, payload, or historical relationship packet is rewritten.
September 1 finish-time correction preparation receipt
This focused candidate starts from main revision
2cd6711f0fdc7d10f2378fe0cc49825a4d815ec3, the merged result of PR #63.
It changes the active finish label for one complete 12-fragment drawing from
September 1, 2026 at 02:13 to 14:13 America/New_York. Tim directly
confirmed the whole-drawing finish, and the physical inscription reads
2:13 PM ET. The separate start-reference image displays 10:26 and is not
finish evidence. Neither photograph nor any photograph hash is published or
asserted.
The exact record-ID, old/new path, old/new finish timestamp, and unchanged
identity mapping is in
finish_time_correction.csv.
The 12 media objects are renamed server-side; the corpus is not downloaded or
re-uploaded.
| Focused check | Required/local candidate result |
|---|---|
| Correction rows / unique persistent IDs | 12 / 12 |
Old active 0213 paths / new active 1413 paths |
0 / 12 |
Corrected metadata rows at 2026-09-01 14:13 |
12 |
Active SHA256SUMS old/new paths |
0 / 12 |
| Ordinary SHA-256 / Xet identity / size preserved | 12 / 12 / 12 |
| Sequence positions / count / protocol preserved | Rec01-Rec12 / 12 / standard-2026-08-27 |
| Standard / Pre-Standard / total active records | 673 / 128 / 801 (unchanged from PR #63) |
| September 2 relationship packet | unchanged; pinned historical paths retained |
| Recording Timing v1 bundle | 9 files unchanged (8 payloads verified by bundle.sha256); 186 rows and 25 groups retained |
| Existing release receipts, tags, and DOI | unchanged |
| Historical/current final-boundary cross-check | +43,221 seconds at pinned 02:13; +21 seconds at active 14:13 |
On the local pre-publication candidate, validate_metadata.py reports zero
errors across 801 active rows, validate_recording_timing.py reports ok with
12 documented correction rows and 12 corrected checksum paths, and the
dependency-free regression suite passes 69/69 tests. Remote inventory and
Viewer results are intentionally deferred until immutable PR and merge
revisions exist.
The historical 2026-09-02 relationship packet and Recording Timing v1
bundle remain byte-identical. Their old paths, 02:13 label, and 43,221-second
conflict describe their immutable pinned snapshots. Current validation joins
those rows by record_id, requires the correction ledger's exact old/new
path-and-time pairs, and verifies that SHA-256, Xet identity, byte size,
sequence order, and protocol assignment did not change. The same retained
final MP4 boundary is 2026-09-01T14:13:21-04:00, 21 seconds after the active
corrected finish minute.
The timing validator no longer requires the current file to contain a
literally unchanged 549-row prefix. It applies two exact, declared inverses to
an in-memory copy: the 12 September 1 finish-time corrections and the separate
PR #67 provenance classification backfill for 12 exact IDs and three exact
fields. Each active backfill value must first equal the reviewed
direct_root_upload / commit_history / commit-d56edd... tuple; only then
are those 36 cells blanked in the temporary reconstruction. The reconstructed
bytes and SHA-256 must equal the original historical receipt. Rows after that
receipt remain append-only, and every undeclared difference still fails. This
preserves the original snapshot receipts and media identities instead of
disabling their checks.
Run the focused and repository-wide checks before publishing the PR, repeat
them against the immutable PR head after reconciling the latest main, and
repeat the inventory and Viewer checks after merge:
python tools/validate_recording_timing.py --root .
python tools/validate_metadata.py metadata.csv \
--pre-standard-metadata Pre_Standard_Time_Lapses/metadata.csv \
--fragment-relationships-root Standard_Time_Lapses/Provenance
python -m unittest discover -s tools -p "test_*.py" -v
python tools/capture_hub_inventory.py \
--repo-json repo_api_blobs.json \
--repo-id maxwellinked/time-lapse-artifacts \
--revision "<immutable-pr-or-merge-commit>" \
--output repo_inventory_xet.json \
--validation-file metadata.csv \
--validation-file Pre_Standard_Time_Lapses/metadata.csv
python tools/validate_sha256sums.py SHA256SUMS \
--repo-json repo_inventory_xet.json \
--revision "<same-immutable-commit>"
Post-merge acceptance additionally requires the Dataset Viewer to return the
same 12 record_id values with time = 14:13 and the corresponding
2026-09-01.1413...Rec01-Rec12 paths, with no old active 0213 path. The PR,
merge commit, immutable inventory receipts, validator totals, and Viewer result
are recorded only after those remote objects exist. No release tag or DOI is
created for this correction.
2026-09-17 11:40 ingest preparation receipt
The media-only pull-request inventory resolves to immutable revision
3843a9b320df45cd7065f5e54d0609276be0530a. The 23 new media paths match the audited ordinary SHA-256
values and byte sizes and have 23 nonblank, unique Xet identities. The
metadata-bearing inventory resolves to 21648a6a76b9174d32a4a3314cbe490219fc2151 and contains the
exact 673-row root index used below. The new sidecar pins that metadata-bearing
revision and is committed afterward, avoiding a circular self-reference.
| Check | Result |
|---|---|
| New unchanged canonical fragments | 23 |
| Complete drawings / sequence sizes | 1 / 23 |
| New media bytes / encoded seconds | 10,419,300,996 / 2010.846325 |
| Local SHA-256 / source stability / sampled decodes | 23/23 / 23/23 / 69/69 |
| Full HEVC decodes | 23/23 |
| Container-declared frames / stored packets | 60,326 / 60,326 |
| Presentation-read / decoded presentation frames | 60,325 / 60,325 |
| Frame-count exception | Rec05 only: one MP4 edit, media time 3000, time base 1/90000; 4,291 stored packets and 4,290 presented/decoded frames; zero decoder errors; integrity pass |
Root metadata.csv |
530,793 bytes; SHA-256 15dd06ebd2f9a6a09846c6a9b085a3f6b1a1a22097235a71519b65868e609a73 |
| Canonical Standard / Pre-Standard / active rows | 673 / 128 / 801 |
| Earlier / prospective Standard protocol rows | 363 / 310 |
| Relationship packets / calendar dates / complete groups | 11 / 10 / 31 |
| New relationship rows / derivatives | 23 / 0 |
| Timing-v1 rows / dated source sidecars | 186 / 6 (unchanged) |
SHA256SUMS |
117,273 bytes; SHA-256 bde3722327dd0c6e6a0ffdd936dbecbad42213ff1d1563e9d4ae0e1ea0250af2 |
| Manifest payload coverage | 807 / 807; 2,147,495,586,241 bytes |
| Active indexed video bytes / encoded playback | 2,130,736,360,966 / 214388.710919 seconds |
| Active videos with an audio stream | 485 / 801 (60.5%) |
| Expected pull-request paths | 34 = 23 media + 11 non-media |
All 23 files form one creator-confirmed complete drawing and use the
creator-supplied 2026-09-17.1140 finish label. The 22 literal VID_* starts,
MP4 creation/finalization tags, and filesystem timestamps remain separate audit
observations; none is substituted for the group finish label. No existing
canonical metadata row, media path, payload, or relationship row is rewritten
by this append. In particular, both existing 2026-09-16 packets and their
revision pins remain byte-identical.
The Rec05 count difference is a documented MP4 presentation-timeline trim, not payload corruption: the single edit begins at media time 3000 on a 1/90000 time base and omits one 1/30-second stored sample from presentation. No exception is accepted for any other file, identity, count, or edit-list shape.
2026-09-16 20:43 ingest preparation receipt
The media-only pull-request inventory resolves to immutable revision
bf4be15ac295fbf631766650660c9b441448c807. The 21 new media paths match the audited ordinary SHA-256
values and byte sizes and have 21 nonblank, unique Xet identities. The
metadata-bearing inventory resolves to ef3a865301091d42e13515d0f5c45bc323c1470e and contains the
exact 650-row root index used below. The new sidecar pins that metadata-bearing
revision and is committed afterward, avoiding a circular self-reference.
| Check | Result |
|---|---|
| New unchanged canonical fragments | 21 |
| Complete drawings / sequence sizes | 1 / 21 |
| New media bytes / encoded seconds | 11,654,268,697 / 2241.410613 |
| Local SHA-256 / source stability / sampled decodes | 21/21 / 21/21 / 63/63 |
| Full HEVC decodes | 21/21 |
| Declared / read / packet / decoded video frames | 67,243 each |
Root metadata.csv |
500,076 bytes; SHA-256 b4edbc232692c13d80f185121a7bc45a48327e57117acae7a8fa42bf61938568 |
| Canonical Standard / Pre-Standard / active rows | 650 / 128 / 778 |
| Earlier / prospective Standard protocol rows | 363 / 287 |
| Relationship packets / calendar dates / complete groups | 10 / 9 / 30 |
| New relationship rows / derivatives | 21 / 0 |
| Timing-v1 rows / dated source sidecars | 186 / 6 (unchanged) |
SHA256SUMS |
113,685 bytes; SHA-256 69b35caceb353fb6350573bdd2e20ba620d4047ae88b74bd16534b02c54653fb |
| Manifest payload coverage | 784 / 784; 2,137,076,285,245 bytes |
| Active indexed video bytes / encoded playback | 2,120,317,059,970 / 212377.864594 seconds |
| Active videos with an audio stream | 485 / 778 (62.3%) |
| Expected pull-request paths | 32 = 21 media + 11 non-media |
All 21 files form one creator-confirmed complete drawing and use the
creator-supplied 2026-09-16.2043 finish label. The 20 literal VID_* starts,
MP4 creation/finalization tags, and filesystem timestamps remain separate audit
observations; none is substituted for the group finish label. No existing
canonical metadata row, media path, payload, or relationship row is rewritten
by this append. In particular, the existing 2026-09-16 packet and its
revision pin remain byte-identical.
v0.6.0 release-preparation receipt
The release content is the exact 834-path tree merged from PR #60 at immutable
revision 0f3e0559ee3d6f5342b7043dd81c9b07ec123be1. It contains 757 active records and 763
published payloads. Relative to v0.5.1, the content revision adds 48 unchanged
canonical Standard MP4s, two dated relationship layers, and two helper-test
modules; it modifies 11 existing non-payload paths and deletes or renames none.
Release preparation uses a separate two-commit pull request. Commit A changes
only CHANGELOG.md, README.md, and VALIDATION.md. After Commit A exists,
RELEASE_MANIFEST_V0.6.0.json binds that exact critical-file revision and is
added alone in Commit B. Commit A has 834 paths and 71 non-MP4/non-ZIP critical
files; the final release tree has 835 paths. Release preparation does not alter
an active index, relationship sidecar, checksum entry, media path, or payload.
| Check | Result |
|---|---|
| Validated content revision | 0f3e0559ee3d6f5342b7043dd81c9b07ec123be1 |
| Canonical Standard / Pre-Standard / active rows | 629 / 128 / 757 |
| Active video bytes / encoded playback | 2,108,662,791,273 / 210136.453981 seconds |
| Earlier / prospective Standard protocol rows | 363 / 266 |
| Dated relationship sidecars / canonical fragments / complete groups | 9 / 266 / 29 |
| Recording Timing v1 | 186 fragments / 25 groups (unchanged) |
| Physical-reference relationships | 309 (unchanged) |
| Published MP4 / ZIP payloads | 762 / 1 |
SHA256SUMS payload coverage |
763 / 763; 2,125,422,016,548 bytes |
| Critical-file receipts after Commit A | 71 / 71 expected |
| Dependency-free repository tests | 62 / 62 passed |
| Dataset Viewer | Derived cache pending at release preparation; not used as a repository-byte fixity claim |
The DOI 10.57967/hf/10449 and immutable v0.5.1 tag continue to identify the
historical 709-record release. They are not moved or reinterpreted as covering
v0.6.0.
v0.6.0 content revision: 2026-09-16 ingest
The media-only pull-request inventory resolves to immutable revision
d365fd2dceecd06338c4f5fad823e47b2b830ef1. The 25 new media paths match the audited ordinary SHA-256
values and byte sizes and have 25 nonblank, unique Xet identities. The
metadata-bearing inventory resolves to a2e0d8a25d1ba02336ff8260dac18822ece7cfaa and contains the
exact 629-row root index used below. The new sidecar pins that metadata-bearing
revision and is committed afterward, avoiding a circular self-reference.
| Check | Result |
|---|---|
| New unchanged canonical fragments | 25 |
| Complete drawings / sequence sizes | 1 / 25 |
| New media bytes / encoded seconds | 6,287,296,691 / 1203.554337 |
| Local SHA-256 / source stability / sampled decodes | 25/25 / 25/25 / 75/75 |
| Full HEVC decodes | 25/25 |
| Declared / read / packet / decoded video frames | 36,107 each |
Root metadata.csv |
472,814 bytes; SHA-256 79a19ef7bcfb5f90235ab532da649cec6cc3e454fa8a09715c5ecd69640661bc |
| Canonical Standard / Pre-Standard / active rows | 629 / 128 / 757 |
| Earlier / prospective Standard protocol rows | 363 / 266 |
| Dated relationship sidecars / complete groups | 9 / 29 |
| New relationship rows / derivatives | 25 / 0 |
| Timing-v1 rows / dated source sidecars | 186 / 6 (unchanged) |
SHA256SUMS |
110,388 bytes; SHA-256 a241207d496969cecef298b877a2de4597978e982e3f442e5267bcb318bbf4a3 |
| Manifest payload coverage | 763 / 763; 2,125,422,016,548 bytes |
| Active indexed video bytes / encoded playback | 2,108,662,791,273 / 210136.453981 seconds |
| Active videos with an audio stream | 485 / 757 (64.1%) |
| Expected pull-request paths | 36 = 25 media + 11 non-media |
All 25 files form one creator-confirmed complete drawing and use the
creator-supplied 2026-09-16.0122 finish label. The 24 literal VID_* starts,
MP4 creation/finalization tags, and filesystem timestamps remain separate audit
observations; none is substituted for the group finish label. No existing
canonical metadata row, media path, payload, or relationship row is rewritten
by this append.
v0.6.0 content revision: dataset-card and helper maintenance
The maintenance candidate is based on immutable Hub revision
ced495111f937bb913fb3dae2b6c84cfa56f1c17. Its intended seven-path delta is
limited to the dataset card, changelog, validation report, two helper scripts,
and two new offline test modules. It does not change an active index, media
path, media payload, checksum entry, relationship or timing sidecar, release
receipt, tag, or DOI registration.
| Check | Result |
|---|---|
| Repository regression tests | 62 / 62 passed |
| New helper regression tests | 9 / 9 passed |
| Active metadata rows | 604 Standard + 128 Pre-Standard = 732 |
| Metadata identities / inventory mismatches | 732 unique IDs and Xet hashes / 0 |
| Payload checksum manifest | 738 / 738 inventory entries verified |
| Recording Timing v1 | 186 rows; 8 / 8 bundle files verified |
| Physical-reference layer | 309 / 309 relationship rows validated |
| Verified-download live header checks | 1 Standard + 1 Pre-Standard; no video downloaded |
| Dataset-card front matter | Byte-identical to the base revision |
| Dataset-card local links | 18 unique targets / 0 missing |
The live Pre-Standard check confirms the repaired path contract: a basename in
Pre_Standard_Time_Lapses/metadata.csv resolves below that index directory.
The Standard check confirms that already repository-relative paths in root
metadata.csv remain unchanged. Enrichment tests confirm that valid existing
record_id values survive reruns, only blank IDs are generated, invalid or
duplicate IDs fail before probing, legacy unbound caches are treated as misses,
and cache hits require matching repository, path, revision, Xet identity, and
byte size.
v0.6.0 content revision: 2026-09-15 ingest
The media-only pull-request inventory resolves to immutable revision
491fc388b858ef7a247e531970a639c717cc8f9d. The 23 new media paths match the audited ordinary
SHA-256 values and byte sizes and have 23 nonblank, unique Xet identities. The
metadata-bearing inventory resolves to 65f7a124c6f6d68f79d6cd89f9ef1b40707929b4 and contains the exact
604-row root index used below. The new sidecar pins that metadata-bearing
revision and is committed afterward, avoiding a circular self-reference.
| Check | Result |
|---|---|
| New unchanged canonical fragments | 23 |
| Complete drawings / sequence sizes | 1 / 23 |
| New media bytes / encoded seconds | 9,488,375,282 / 1,853.681426 |
| Local SHA-256 / full demux / sampled decodes | 23/23 / 23/23 / 69/69 |
| Full HEVC decodes | 23/23 |
| Declared / counted video frames | 55,611 / 55,611 |
Root metadata.csv |
440,444 bytes; SHA-256 0dfeb4e008e26b2bcda7df3112e814e677a6f5197bd69ab25d25ed888651974a |
| Canonical Standard / Pre-Standard / active rows | 604 / 128 / 732 |
| Earlier / prospective Standard protocol rows | 363 / 241 |
| Dated relationship sidecars / complete groups | 8 / 28 |
| New relationship rows / derivatives | 23 / 0 |
| Timing-v1 rows / dated source sidecars | 186 / 6 (unchanged) |
SHA256SUMS |
106,488 bytes; SHA-256 9034342cf693f31564be5b9c6d90bf07b9ecbe49440a06726edad90fc95f7542 |
| Manifest payload coverage | 738 / 738; 2,119,134,719,857 bytes |
| Active indexed video bytes / encoded playback | 2,102,375,494,582 / 208,932.899644 seconds |
| Active videos with an audio stream | 485 / 732 (66.3%) |
| Expected pull-request paths | 34 = 23 media + 11 non-media |
All 23 files form one creator-confirmed complete drawing and use the
creator-supplied 2026-09-15.1236 finish label. Filename, MP4-container, and
filesystem times remain separate audit observations; none is substituted for
the group finish label. No existing canonical metadata row, media path,
payload, or relationship row is rewritten by this append.
v0.5.1 semantic-metadata and documentation maintenance
Relative to tagged commit
886ee9522f99e7176285f589014a874e4dab32e2, this release changes exactly one
active-index value: the tool field for Pre-Standard record
tla-34d3d5f4de884c6b changes from FountainPen to Brush;FountainPen,
matching its literal recognized filename tokens. No record identity, media
path, media hash, byte size, encoded technical field, index membership,
relationship sidecar, checksum entry, or media payload changes.
The resulting Pre-Standard CSV is 38,175 bytes with ordinary SHA-256
f46a7344a7fc66803f3a75d4546cf3d9b91419a5c3c44d301059fe4e207047dd.
All 53 dependency-free regression tests pass at merged content revision
c839fade7308aa78231803221514fa5c06f34aa2, and the metadata validator reports
zero errors against the 779-path inventory resolved at that revision.
RELEASE_MANIFEST_V0.5.1.json records
c839fade7308aa78231803221514fa5c06f34aa2 as the validated content revision
and separately binds critical-file receipts to the release-documentation
revision immediately preceding manifest addition. The final tagged tree adds
only that manifest to the critical-file revision.
v0.5.0 content revision: 2026-09-14 ingest
The media-only pull-request inventory resolves to immutable revision
0ef0f61cd0c54232baa58774713d57168e5e90de. The 32 new media paths match the audited ordinary
SHA-256 values and byte sizes and have 32 nonblank, unique Xet identities. The
metadata-bearing inventory resolves to db1d515c9626adf25207c2e844c913d74dba438f and contains the exact
581-row root index used below. The new sidecar pins that metadata-bearing
revision and is committed afterward, avoiding a circular self-reference.
| Check | Result |
|---|---|
| New unchanged canonical fragments | 32 |
| Complete drawings / sequence sizes | 2 / 4 and 28 |
| New media bytes / encoded seconds | 9,568,508,054 / 1,859.281028 |
| Local SHA-256 / full demux / sampled decodes | 32/32 / 32/32 / 96/96 |
| Declared / counted video frames | 55,779 / 55,779 |
Root metadata.csv |
411,583 bytes; SHA-256 f493638a947ec3a053845a02c4f64e6d57e3bd852948805c886b678adbdd7479 |
| Canonical Standard / Pre-Standard / active rows | 581 / 128 / 709 |
| Earlier / prospective Standard protocol rows | 363 / 218 |
| Dated relationship sidecars / complete groups | 7 / 27 |
| New relationship rows / derivatives | 32 / 0 |
| Timing-v1 rows / dated source sidecars | 186 / 6 (unchanged) |
SHA256SUMS |
103,245 bytes; SHA-256 09ff9cbb6722886b882c35aed2e2247944d7ec960508aa4504e6e226f90bbf17 |
| Manifest payload coverage | 715 / 715; 2,109,646,344,575 bytes |
| Active indexed video bytes / encoded playback | 2,092,887,119,300 / 207,079.218218 seconds |
| Active videos with an audio stream | 485 / 709 (68.4%) |
| Expected pull-request paths | 45 = 32 media + 13 non-media |
The first four files form the first creator-confirmed complete drawing and keep
the 2026-09-02.2358 finish label. The remaining 28 form the second complete
drawing and use the creator-supplied 2026-09-14.1635 finish label. Filename,
MP4-container, and filesystem times remain separate audit observations; none is
substituted for the group finish label. No existing canonical metadata row,
media path, payload, or relationship row is rewritten by this append.
v0.5.0 content revision: exploratory recording-timing layer
The recording-timing sidecar is pinned to source revision
b4b58f0ce0effb31ebf7671fbb0bd9c11765db4e and covers exactly the 186
canonical fragments in the six dated relationship sidecars. The included
validator passed with repository-packet self-consistency scope.
Its 549-row, 370,833-byte canonical metadata receipt remains a historical
snapshot. The validator applies the exact inverse path/time mapping for the 12
IDs in the documented September 1 correction CSV. It also verifies and then
reverses the separate PR #67 provenance backfill for exactly 12 other IDs and
exactly three fields per row; that metadata-only change was commit
57a465ea69a4a6444e668f3267095a1b9fdfff6b, merged as
6758d92f07f233166a1fa8a5c04e46c6cd96382e. The reconstructed 549-row bytes
must equal the original receipt, including SHA-256
d516da7a4341dddf0b2b5ad50f6200f283ea0cdefc6de465882fbc021abfba8a.
Rows after that reconstructed receipt must retain the same 29-field header and
width and nonblank, globally unique record_id, file_name, and
hub_xet_hash identities. The root metadata validator remains authoritative
for the rest of the schema. Appended rows remain outside timing v1 coverage
until a separately reviewed, versioned timing layer includes them.
The preserved v1 bundle's validation_summary.json still records
metadata_csv_sha256_unchanged: true. That is a historical result from the
pinned-revision publication check, not a claim that the current whole
metadata.csv keeps the old hash after append. Current compatibility is
established by exact inverse reconstruction of the historical receipt, the
declared 12-row finish-time correction round-trip, the exact 12-record/36-cell
provenance-backfill inverse, and current identity joins for all 186 covered
records.
| Check | Result |
|---|---|
| Timing rows | 186 |
| Direct device-filename start observations | 22 |
| Profile-matched model applications | 164 |
| Independent holdout observations | 0 |
| Maximum absolute in-sample model residual | 0.833333 seconds |
| Owner-confirmed complete source sequences | 25 |
| Within-sequence transitions | 161 |
| Positive beyond uncertainty / within uncertainty / negative beyond uncertainty | 159 / 2 / 0 |
| Final boundaries near documented finish minute | 24 / 25 |
| Preserved documented-finish conflicts | 1 |
| Timing bundle files verified | 8 |
| Repository regression tests | 46 / 46 passed |
The one retained conflict is
2026-09-01.0213.BallpointPen.Ink.Paper.9x12, whose final MP4 boundary is
43,221 seconds after the documented finish minute in the immutable timing-v1
snapshot. That historical path, time, conflict state, and note are not
rewritten. The active correction maps those 12 IDs to the 14:13 finish label;
the same retained final MP4 boundary is then 21 seconds after the corrected
finish minute. The 109,973.465-second serialized aggregate mixes 22 observed
and 164 estimated start-to-finalization intervals; it is neither drawing-active
time nor corpus calendar span. Calculation before three-decimal row
serialization differs by 0.001667 seconds.
The validator recomputes the inverse-reconstructed historical metadata receipt, the exact 12-row finish-time correction round-trip, the exact 12-record/36-cell provenance-backfill inverse, current identity joins for all 186 covered records, timestamp arithmetic, row formulas, full-precision calibration metrics, sequence chronology, group summaries, validation summaries, source receipts, and bundle hashes:
python tools/validate_recording_timing.py --root .
python -m unittest discover -s tools -p "test_*.py" -v
This repository-only run does not re-probe source-media tags or frame counts,
rehash the 186 source videos, verify the retained local audit bytes, or
re-observe NTFS timestamps. Publication of the timing layer changed no MP4,
canonical metadata row or field, relationship sidecar, Viewer configuration,
or root SHA256SUMS payload entry.
v0.5.0 pre-tag content receipt
RELEASE_MANIFEST_V0.5.0.json evaluates every listed critical_files byte
size and SHA-256 digest at validated_content_revision
9089dd9447672866a8a7cb9b97ebf28cefd85e87. It is not a checksum of every
file at the final v0.5.0 tag.
The final annotated v0.5.0 tag resolves to commit
886ee9522f99e7176285f589014a874e4dab32e2. At that commit, 19 of the 22 listed
files retain the receipt bytes. README.md, CHANGELOG.md, and
VALIDATION.md changed during release preparation; the active indexes,
relationship sidecars, checksum manifest, and media payloads did not change.
Release preparation started from exact Hub main revision
9089dd9447672866a8a7cb9b97ebf28cefd85e87, the verified merged result of
Hugging Face PR #52. It changed only README.md, CHANGELOG.md, and
VALIDATION.md, and added RELEASE_MANIFEST_V0.5.0.json. It did not change an
active index, relationship sidecar, validator, checksum entry, media path, or
media payload.
The pre-tag manifest binds the following content-revision facts:
| Check | Result |
|---|---|
| Canonical Standard / Pre-Standard / active rows | 581 / 128 / 709 |
| Earlier / prospective Standard protocol rows | 363 / 218 |
| Dated fragment sidecars / canonical fragments / complete groups | 7 / 218 / 27 |
| Timing-v1 rows / groups | 186 / 25 (unchanged) |
| Published MP4 / ZIP payloads | 714 / 1 |
SHA256SUMS coverage |
715 / 715; 2,109,646,344,575 bytes |
| Active indexed bytes / encoded playback | 2,092,887,119,300 / 207,079.218218 seconds |
| Active videos with an audio stream | 485 / 709 |
| Critical-file receipts at the validated content revision | 22 / 22 |
| Repository regression tests | 46 / 46 passed |
The content revision's full Hub inventory contains 778 paths and 715 LFS/Xet payloads. Metadata, relationship, timing, checksum, and exact Viewer-row validation passed; the Dataset Viewer settled at 581 canonical plus 128 Pre-Standard rows with zero pending or failed jobs. The release-preparation delta was documentation-only, and the resulting annotated tag was created only after that delta was merged and its exact tree was verified.
v0.4.0 pre-tag content receipt
Branch codex/prepare-v0.4.0-release started at exact Hub main revision
442b4786afbdd31bd16cd0a1488f09c07f956316, the merged result of Hugging Face
PR #46. Release preparation changed only release documentation and added a
pre-tag content manifest. It did not change an active index,
relationship sidecar, validator, checksum entry, or media path or payload. No
MP4 was materialized in the working checkout.
The underlying 29-field snapshot assigns protocol versions from documented relationship membership rather than a date cutoff or filename inference:
| Check | Result |
|---|---|
| Canonical Standard rows | 549 |
Earlier workflow, standard-2025-07-13 |
363 |
Prospective workflow, standard-2026-08-27 |
186 |
| Dated fragment sidecars | 6 |
| Prospective groupings with one shared finish timestamp | 25 / 25 |
Historical RecNN rows outside the sidecars, retained on the earlier protocol |
19 / 19 |
| Metadata columns before / after | 29 / 29 |
| Non-protocol metadata field changes | 0 |
| MP4 path or payload changes | 0 |
| Relationship-sidecar changes | 0 |
SHA256SUMS byte changes |
0 |
| Regression tests | 35 / 35 passed |
The validated metadata.csv remains 370,833 bytes and has ordinary
SHA-256 d516da7a4341dddf0b2b5ad50f6200f283ea0cdefc6de465882fbc021abfba8a.
Relative to the pre-boundary revision, exactly 186
acquisition_protocol_version values changed; every other field in every row
is byte-for-byte equivalent after CSV parsing. In particular, all record_id,
file_name, hub_xet_hash, size_bytes, encoded-video fields, and provenance
fields are unchanged. SHA256SUMS remains byte-identical at SHA-256
21e9c6dd4cb976754a73201cf5a5dcab8b4402a03781b9afe9ae8ea5f2c4b351.
The metadata validator passed against the base revision's 732-file Hub
inventory with zero path, Xet-identity, or size mismatches. All six fragment
sidecars passed both current structural/metadata joins and their original
pinned-revision inventory checks using the metadata bytes from each declared
source_hub_revision. Checksum coverage passed for all 683 payloads (682 MP4
and one ZIP), totaling 2,100,077,836,521 bytes. The 309-row physical-reference
relationship validation also passed against its pinned source metadata.
Scope and publication boundary
Hugging Face PR #45 begins at exact Hub main revision
81d5675113fcd5e26eac0289c228f7ccfe552e42. It adds 29 unchanged canonical
Framelapse source fragments grouped into two creator-confirmed complete
RecNN sequences. It does not change the 29-field active schema,
standard-2025-07-13, any existing media payload, or any existing
record_id.
The media-only PR revision is
9bfce508b6c6bdca12f475deb9b540f05754d8fd. The metadata-bearing revision is
7fca6fcc36d873b92a6c64fff5980e8d5afde5a6; it contains the exact 549-row root index
validated below. The expanded 61-row relationship sidecar is committed afterward
and pins that metadata-bearing revision, avoiding a circular self-reference.
This report validates the pull-request state; merge status and Viewer
settlement are separate deployment checks.
PR #45 immutable active-index receipt
PINNED REMOTE-INVENTORY PASS — zero validation errors
The repository inventory was captured from the Hub API and immutable resolve headers at the metadata-bearing revision. The validators checked both active indexes, every indexed media identity and byte size, the repository-wide payload checksum manifest, and the new relationship sidecar against that inventory.
| Check | Result |
|---|---|
| Hub source revision | 7fca6fcc36d873b92a6c64fff5980e8d5afde5a6 |
Root metadata.csv |
370,833 bytes; SHA-256 9456adc5922c3dd6a3b511ecdc4ef97fa803900e9e2a1f028ad76ffe45d2d5b2 |
Pre-Standard metadata.csv |
38,169 bytes; SHA-256 7ef4b21645760b2aa2047ad1b548cb341c97c288378498f135cbbecd82c4f4b9 |
| Active indexed rows | 677 |
| Canonical Standard / Pre-Standard | 549 / 128 |
| Metadata columns per index | 29 |
Unique record_id / nonblank Xet identities |
677 / 677 |
| Direct Standard / Pre-Standard MP4 files | 549 / 128 |
| Nested noncanonical provenance MP4 files | 5 |
| Missing indexed targets | 0 |
| Unindexed direct active-tier MP4 files | 0 |
| Metadata-to-Hub size mismatches | 0 |
| Metadata-to-Hub Xet mismatches | 0 |
SHA256SUMS |
98,313 bytes; SHA-256 21e9c6dd4cb976754a73201cf5a5dcab8b4402a03781b9afe9ae8ea5f2c4b351 |
| Manifest payload coverage | 683 / 683; 2,100,077,836,521 bytes |
| Manifest-to-Hub identity errors | 0 |
| Regression tests | 30 / 30 passed |
Indexed video totals are 2,083,318,611,246 bytes (2.083 TB decimal) and 57.006 encoded playback hours. Encoded playback time is not real drawing time.
PR #43 semantic-correction receipt
The content-bearing PR revision
4d6c6098c7428d77e7c04771a2aabcb46918345f is based directly on Hub main
revision 9ba68359e4886d12ad865bd0a5cd00515a78ede6. Remote validation at that
immutable revision produced zero errors.
| Check | Result |
|---|---|
Root metadata.csv |
311,869 bytes; SHA-256 00bc268f0267b09f2cd18e8b1ac8751f45e7d2b2232d240b438857f82558a161 |
Pre-Standard metadata.csv |
38,169 bytes; SHA-256 7ef4b21645760b2aa2047ad1b548cb341c97c288378498f135cbbecd82c4f4b9 |
| Active indexed rows | 616 |
| Canonical Standard / Pre-Standard | 488 / 128 |
Unique record_id / nonblank Xet identities |
616 / 616 |
| Missing indexed targets | 0 |
| Unindexed direct active-tier MP4 files | 0 |
| Metadata-to-Hub size / Xet mismatches | 0 / 0 |
SHA256SUMS |
90,319 bytes; SHA-256 5e86108f0323c4a5c0c6e01636529ecc185252d5cc1eb23e5b46ab7a0ccd8e3f |
| Manifest payload coverage | 622 / 622; 2,079,242,936,421 bytes |
| Manifest-to-Hub identity errors | 0 |
| Regression tests | 30 / 30 passed |
All fifteen renamed June 30 × 40 in paths have the same Hub LFS SHA-256 and
byte-size receipts as their Paper-named source paths at the base revision.
The PR head contains all fifteen IllustrationBoard paths and none of the old
Paper paths. This verifies a path-and-metadata correction without a media
payload rewrite. Viewer settlement remains a post-merge check and has not been
claimed here.
Historical pinned active-index receipt
This receipt records the exact PR #42 validation event described above; it is
not a digest of current main.
PINNED REMOTE-INVENTORY PASS — zero validation errors
The repository inventory was captured from the Hub API and immutable resolve
headers at the metadata-bearing revision. tools/validate_metadata.py then
checked both active indexes and every indexed media identity and byte size
against that inventory.
| Check | Result |
|---|---|
| Hub source revision | fe66d1dbb117486c2046376423316fd2cc25d488 |
Root metadata.csv |
309,400 bytes; SHA-256 1a20df86d44dde90ed89bda230666772f24b04ca0dbd2db8fba5749fe1da1308 |
Pre-Standard metadata.csv |
37,361 bytes; SHA-256 15de8b2c97a8c463dd279054ab0ce3894caa6602f3b72f8025d32511746c3630 |
| Active indexed rows | 616 |
| Canonical Standard / Pre-Standard | 488 / 128 |
| Metadata columns per index | 29 |
Unique record_id / nonblank Xet identities |
616 / 616 |
| Direct Standard / Pre-Standard MP4 files | 488 / 128 |
| Nested noncanonical provenance MP4 files | 5 |
| Missing indexed targets | 0 |
| Unindexed direct active-tier MP4 files | 0 |
| Metadata-to-Hub size mismatches | 0 |
| Metadata-to-Hub Xet mismatches | 0 |
Indexed video totals are 2,062,483,711,146 bytes (2.062 TB decimal) and 55.878 encoded playback hours. Encoded playback time is not real drawing time.
Viewer settlement is checked after merge but is not used as a substitute for the revision-pinned row, media, size, and identity checks above.
Historical repository-wide payload checksum receipt
This receipt likewise records the exact PR #42 validation event rather than
the current SHA256SUMS file.
SHA256SUMS was generated from the LFS-compatible content identities at exact
Hub revision e2094ae582c057df9d40e9bfe4dc00afed0bad2a. Every identity was also
matched to the immutable resolve response's X-Linked-ETag and byte size. The
Xet file ID was retained separately and was not treated as a SHA-256.
| Check | Result |
|---|---|
| Manifest entries | 622 |
| Published MP4 payloads | 621 / 621 |
| Published ZIP payloads | 1 / 1 |
| Payload bytes covered | 2,079,242,936,421 |
| Missing / extra manifest paths | 0 / 0 |
| Invalid / duplicate SHA-256 values | 0 / 0 |
| LFS SHA-256 / resolve-header mismatches | 0 / 0 |
| Manifest byte size | 89,532 |
| Manifest SHA-256 | 968e885399584ff6532af7fcb0c23602e6eb27d1f9ff7525b3083ca6f584ac93 |
These are ordinary byte-level SHA-256 values for the payload contents. The
cross-check did not transfer and rehash all 2.079 TB of payload bytes; a full
download can independently reproduce every digest with
tools/validate_sha256sums.py --verify-root.
PR #45 new-source media verification
Before upload, all 29 sources were SHA-256 hashed, probed, fully demuxed, and sample-decoded as raw pixels at start, middle, and near-end checkpoints. An independent uploader preflight rehash confirmed that all 29 sources remained byte-identical. Results were 29/29 hashes, 29/29 complete packet demuxes, 87/87 decoded frame samples, 60,900 container frames matching 60,900 counted video packets, and zero errors. The local source directory was not modified.
Relationship validation
Fragment sequences
All twelve relationship packet sidecars spanning eleven calendar dates pass current structural and metadata-join validation:
| Sidecar | Canonical fragments | Derivatives | Sets | Result |
|---|---|---|---|---|
2026-08-27 |
22 | 5 | 5 | zero errors |
2026-08-29 |
19 | 0 | 4 | zero errors |
2026-08-30 |
39 | 0 | 5 | zero errors |
2026-08-31 |
32 | 0 | 4 | zero errors |
2026-09-01 |
13 | 0 | 2 | zero errors |
2026-09-02 |
61 | 0 | 5 | zero errors |
2026-09-14 |
32 | 0 | 2 | zero errors |
2026-09-15 |
23 | 0 | 1 | zero errors |
2026-09-16 |
25 | 0 | 1 | zero errors |
2026-09-16.2043 |
21 | 0 | 1 | zero errors |
2026-09-17 |
23 | 0 | 1 | zero errors |
2026-09-18 |
29 | 0 | 2 | zero errors |
The validator compares canonical byte sizes with joined metadata and emits an
explicit verification level. The 2026-09-15 sidecar was checked against the
exact metadata-bearing remote inventory at 65f7a124c6f6d68f79d6cd89f9ef1b40707929b4 and reports
pinned_repository_inventory and inventory_binding_checked: true. The
2026-09-16 sidecar is independently bound to metadata-bearing revision
a2e0d8a25d1ba02336ff8260dac18822ece7cfaa; the separate 2026-09-16.2043
packet is independently bound to ef3a865301091d42e13515d0f5c45bc323c1470e;
the 2026-09-17 packet is independently bound to 21648a6a76b9174d32a4a3314cbe490219fc2151;
the 2026-09-18 packet is independently bound to d3e4da0bb5dd3eb573f561f588f4792b93fbd88b with the
same inventory-binding checks.
Historical sidecars pass current structural and metadata joins; their own
pinned-inventory receipts remain in the dated READMEs.
owner_confirmed_complete describes the retained canonical source sequence's
membership and order. It does not independently establish gapless real-world
capture, real drawing time, or permanent physical-work completion.
Physical references
The physical-reference sidecar passes its dedicated validator against the
exact 363-row source metadata at revision
ba7e7903901767f2d5a4367d7ae22c2546366d6d:
| Check | Result |
|---|---|
| Source metadata receipt | 190,250 bytes; SHA-256 2062d1ed0ae83a54cac501e1dfcc23bc6cff087d43fd1242fdd19f3028f3d30d |
| Sidecar receipt | 346,794 bytes; SHA-256 5f3e1ed6c484853cd6df8722e3916b111fc280a1e02a4f9272c0566172821c3a |
| Relationship rows / columns | 309 / 32 |
Unique reference_id / record_id |
309 / 309 |
| Endpoint photographs | 52 |
| Owner-confirmed record references | 232 |
| Two-sided work references | 24 |
| Longitudinal continuations | 1 |
| Structural, join, hash-state, and manifest errors | 0 |
Nine rows previously said front_and_paired_hashes_verified while their paired
filename and paired hash were blank. They now use the mechanically accurate
front-only ledger state. No photograph pairing, endpoint, date, material, or
owner evidence was reinterpreted.
The photographs remain local evidence and are not bundled. This validation checks internal consistency and exact historical metadata joins; it is not visual authentication of absent media.
Regression tests
The dependency-free suites currently run 69 tests covering repository receipt binding, fragment-size joins and verification levels, semantic documentation, recognized filename-tool coverage, derived headline facts, archive-origin framing, the exact protocol-version boundary and prospective timestamp scope, verified-download and Pre-Standard enrichment behavior, and physical-reference identity, hash-state, pairing, manifest, metadata-join failures, and append-compatible recording-timing receipts. Focused timing tests also reject correction-row count or identity drift, active metadata drift, historical relationship-packet mutation, non-LF metadata serialization, altered active checksum values, and retained old checksum paths.
python -m unittest discover -s tools -p "test_*.py"
Result: 69/69 passed.
Outstanding curator questions and limitations at the v0.7.0 data baseline
These remain explicit; validation must not guess their answers:
- One Pre-Standard record has no support token or independent support evidence;
its
supportremains blank and itsnotesstate the limitation. - The earlier Standard and Pre-Standard workflows do not normalize whether a
documented finish label applies to a recording, session, reviewed fragment
sequence, or physical work. Under
standard-2026-08-27, the scope is the complete prospective grouping/physical work. - Twenty active Standard
RecNNrows are outside the twelve relationship packet sidecars and retainstandard-2025-07-13. Their existing notes and changelog evidence remain authoritative; sidecar absence does not establish no relationship or sequence completeness. - Twenty-seven physical-reference rows retain explicit conflict evidence. Conflict language is an adjudication queue, not proof that 27 canonical timestamps are wrong.
- Three Standard records are explicitly documented shortened derivatives, not full-duration masters.
- Most active indexed videos at this revision contain an audio stream (486 of 831). Stream presence does not establish meaningful signal; rights and privacy review remain necessary.
- Encoded frame rate and playback duration cannot by themselves recover wall-clock or real drawing time. Recording Timing v1 covers 186 matching-profile fragments; Recording Timing v2 documents 339 prospective fragments, including 163 filename-derived starts and 176 unknown starts. Neither layer supplies drawing-active time or calibrated per-frame timestamps.
- The 2012–2016 archive has no public video payload in this release.
- No official train/validation/test partition exists. The partial global
work_idregistry has 360 positive recording links across 37 works; 471 active recordings remain unresolved. Distinct IDs are not verified negative pairs for work-disjoint evaluation.
Reproduction
Validate both active indexes locally:
python tools/validate_metadata.py metadata.csv \
--pre-standard-metadata Pre_Standard_Time_Lapses/metadata.csv \
--fragment-relationships-root Standard_Time_Lapses/Provenance
Validate the payload checksum manifest against a blob-enabled repository inventory captured from one immutable revision:
python tools/validate_sha256sums.py SHA256SUMS \
--repo-json repo_api_blobs.json \
--revision "<40-hex-commit>"
For the strongest active-index check, capture an inventory at one immutable revision, including exact receipts for both metadata files:
python tools/capture_hub_inventory.py \
--repo-json repo_api_blobs.json \
--repo-id maxwellinked/time-lapse-artifacts \
--revision "<40-hex-commit>" \
--output repo_inventory_xet.json \
--validation-file metadata.csv \
--validation-file Pre_Standard_Time_Lapses/metadata.csv
python tools/validate_metadata.py metadata.csv \
--pre-standard-metadata Pre_Standard_Time_Lapses/metadata.csv \
--fragment-relationships-root Standard_Time_Lapses/Provenance \
--repo-inventory-json repo_inventory_xet.json
Validate the dated fragment sidecars against current metadata for structural
joins. For a pinned-inventory check, use the exact historical metadata.csv
downloaded from that sidecar's source_hub_revision; substituting the current
metadata bytes would invalidate the receipt:
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-08-27/fragment_sequence_relationships.csv \
--metadata metadata.csv
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-08-29/fragment_sequence_relationships.csv \
--metadata metadata.csv
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-08-30/fragment_sequence_relationships.csv \
--metadata pinned/7293edccff883b4c604d710addbf26c85910cfcc/metadata.csv \
--repo-inventory-json repo_inventory_7293edcc.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-08-31/fragment_sequence_relationships.csv \
--metadata pinned/480bc7e24261e6d3bbcacd261a00782b7e40211a/metadata.csv \
--repo-inventory-json repo_inventory_480bc7e2.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-09-01/fragment_sequence_relationships.csv \
--metadata pinned/fe66d1dbb117486c2046376423316fd2cc25d488/metadata.csv \
--repo-inventory-json repo_inventory_fe66d1db.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-09-02/fragment_sequence_relationships.csv \
--metadata pinned/7fca6fcc36d873b92a6c64fff5980e8d5afde5a6/metadata.csv \
--repo-inventory-json repo_inventory_7fca6fcc.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-09-14/fragment_sequence_relationships.csv \
--metadata pinned/db1d515c9626adf25207c2e844c913d74dba438f/metadata.csv \
--repo-inventory-json repo_inventory_db1d515c.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-09-15/fragment_sequence_relationships.csv \
--metadata pinned/65f7a124c6f6d68f79d6cd89f9ef1b40707929b4/metadata.csv \
--repo-inventory-json repo_inventory_65f7a124.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-09-16/fragment_sequence_relationships.csv \
--metadata pinned/a2e0d8a25d1ba02336ff8260dac18822ece7cfaa/metadata.csv \
--repo-inventory-json repo_inventory_a2e0d8a2.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-09-16.2043/fragment_sequence_relationships.csv \
--metadata pinned/ef3a865301091d42e13515d0f5c45bc323c1470e/metadata.csv \
--repo-inventory-json repo_inventory_ef3a8653.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-09-17/fragment_sequence_relationships.csv \
--metadata pinned/21648a6a76b9174d32a4a3314cbe490219fc2151/metadata.csv \
--repo-inventory-json repo_inventory_21648a6a.json
python tools/validate_fragment_relationships.py \
--relationships Standard_Time_Lapses/Provenance/2026-09-18/fragment_sequence_relationships.csv \
--metadata pinned/d3e4da0bb5dd3eb573f561f588f4792b93fbd88b/metadata.csv \
--repo-inventory-json repo_inventory_d3e4da0b.json
Retrieve the exact historical metadata bound by the physical-reference manifest, then validate the sidecar:
hf download maxwellinked/time-lapse-artifacts metadata.csv \
--type dataset \
--revision ba7e7903901767f2d5a4367d7ae22c2546366d6d \
--local-dir physical-reference-source
python tools/validate_physical_references.py \
--references Physical_References/endpoint_references.csv \
--metadata physical-reference-source/metadata.csv \
--manifest Physical_References/MANIFEST.json
Always record the exact repository revision. A moving branch is not a reproducible identity receipt.
Confirmed work-ID display projection (2026-09-19; v0.7.0 snapshot)
At the tagged v0.7.0 snapshot, the active indexes and both Viewer configurations
expose work_id immediately after record_id, for 35 fields per tier. The
generated column copies exactly
360 positive owner-confirmed registry links covering 37 active drawings, leaving
471 recording identities unresolved. The registry tables and their historical
evidence remain unchanged. All previous 34 field values on 831 rows, row order,
media paths, media identities, release tags and preview bindings are preserved.
python tools/project_work_ids.py --root .
python tools/project_recording_times.py --root .
python -m unittest discover -s tools -p 'test_*.py'
Regression coverage includes duplicate/conflicting assignments, unsupported owner evidence, stale generated IDs, inactive work references, missing recordings, confirmation-checksum drift, malformed rows, supersession cycles and independent timing/work rebuild order. The historical Timing v1 validator still reconstructs its exact original 29-field checksum after removing only named generated fields. Local VideoFolder loading verifies all 831 rows, work-ID nulls and existing timing values using placeholders with decoding disabled. This metadata display change neither downloads nor re-decodes source media. Hosted Dataset Viewer cache readiness is checked separately after publication.