Skip to content

VISUS source-resolution status

GazeForge treats locating the VISUS benchmark as a separate scientific-provenance task from validating a local copy. The repository therefore records what the public literature and current institutional indexes establish, while refusing to turn historical availability into a current source or licensing claim.

The baseline machine-readable checkpoint is validation/protocols/visus-source-resolution-2026-09-04.json. A dated institutional recheck is frozen separately at validation/evidence/visus-source-recheck/visus-authoritative-source-recheck-2026-09-09.json so it does not replace or duplicate the reviewed dashboard checkpoint. Both are source-resolution status records, not source-audit specifications and not empirical evidence.

What is established

The original 2014 benchmark publication is:

Kuno Kurzhals, Cyrill Fabian Bopp, Jochen Bässler, Felix Ebinger, and Daniel Weiskopf. Benchmark Data for Evaluating Visualization and Analysis Techniques for Eye Tracking for Video Stimuli. BELIV 2014. DOI: 10.1145/2669557.2669558.

The paper describes the benchmark as publicly available and names the original distribution endpoint:

http://go.visus.uni-stuttgart.de/eyetrackingBenchmark

It describes 11 video scenarios and 25 participants, with gaze acquired using a Tobii T60 XL at 60 Hz. The videos are 1920×1080 and normalized to 25 fps. The distributed benchmark is described as containing video stimuli, exported eye-tracking data, and dynamic AOI annotations in ViPER-compatible XML.

A 2021 Sensors paper independently reports that the VISUS dataset was downloadable from:

https://www.visus.uni-stuttgart.de/publikationen/benchmark-eyetracking

and records an access date of 12 April 2021. This is useful historical distribution evidence; it is not proof that the same endpoint or terms remain current.

Current resolution result

A public-source resolution pass on 2026-09-04 checked the current VISUS institutional site/indexed publication pages, the historical endpoint identifiers, the University of Stuttgart DaRUS VISUS dataverse, the ACM publication record, and general public indexing.

The current VISUS site still lists the 2014 publication on Kuno Kurzhals's institutional profile. DaRUS also contains current VISUS eye-tracking datasets. However, the search did not locate a current authoritative distribution for this specific 2014 benchmark, a matching DaRUS record, a separate dataset DOI, or explicit current dataset reuse terms.

Accordingly, GazeForge records:

  • current_authoritative_distribution_unresolved;
  • current_authoritative_download_found=false;
  • source_audit_ready=false;
  • analysis-use terms unresolved;
  • raw-data redistribution terms unresolved.

Failure to locate a current distribution is not evidence that no authoritative copy exists. It means the public evidence available in this resolution pass is insufficient for GazeForge's empirical source gate.

2026-09-09 institutional recheck

The 2026-09-09 recheck repeated the authoritative-source search instead of treating the 2026-09-04 checkpoint as permanently current. It records the same unresolved scientific result with additional present-day evidence:

  • the live Kuno Kurzhals VISUS profile still lists the 2014 benchmark publication;
  • the live VISUS DaRUS dataverse reports 18 records, but exact-title, publication-DOI, and author-targeted searches in this recheck did not locate a record matching the 2014 25-participant × 11-stimulus benchmark;
  • modern VISUS DaRUS datasets such as 10.18419/DARUS-4023 and 10.18419/DARUS-4141 expose explicit CC BY 4.0 dataset licenses, but those licenses apply to those deposits only and are not transferred backward to the 2014 benchmark;
  • a 2026 dissertation still cites the historical VISUS distribution page with an access date of 12 April 2021, so it corroborates historical access rather than current availability;
  • current public code/derivative searches still recover VISUS-derived material, not an authoritative replacement for the complete benchmark.

The recheck is frozen as:

validation/evidence/visus-source-recheck/visus-authoritative-source-recheck-2026-09-09.json

with canonical fingerprint:

97bd19b892032a663b7707235504a7ba886a86d6c3caab4966fa6c5ee96a2571

Its strict validator additionally binds the two already-reviewed derivative evidence records to their real repository bodies:

  • public partial dynamic-AOI evidence: 80e008228e39c2b17bae99a526e2a0157c2c850ebe803c5a370abe9167efde14 (3 participants × 1 stimulus);
  • public event-extension evidence: 2f12bd83d71786bfae7101dec6515c49c5ff4e696df8675b3955300e5e5e6dfd (2 additional complete Tobii exports, with stimulus identity still not file-bound).

Neither derivative is promoted to the full VISUS source. The strict layer rejects even correctly refingerprinted attempts to claim full-corpus recovery, a matching DaRUS record, a separate dataset DOI, an explicit current benchmark license, transfer of modern DaRUS licenses, derivative source authority, independent annotation streams, source-audit authorization, human-human agreement, model-human validation, Frozen Evidence eligibility, or raw-source redistribution.

The 2026-09-09 result therefore does not satisfy the roadmap item “Obtain/verify an authoritative current VISUS copy and reuse/distribution terms.” That item remains open.

2026-09-10 Osnabrück derivative recovery

A new authenticated lead was pursued after the 2021 supplementary-material recovery search was frozen. The University of Osnabrück Institute of Cognitive Science multimedia-container page explicitly lists the Kurzhals et al. [K] dataset — 11 videos among converted datasets provided for research purposes, with ASS and USF variants and a citation requirement for both the converter work and the original dataset.

The page's current individual-file links no longer resolve to the historical MKVs, but its legacy institutional bundle endpoint still serves a real ZIP object. The host's TLS certificate chain is currently invalid, so recovery was deliberately treated as a historical derivative-source operation rather than current source authority.

A bounded GitHub Actions reconnaissance recovered the ZIP central directory without downloading the full 2.6 GB archive and then reconstructed each of the 11 standard USF members independently by exact byte range. Every reconstructed member passed its ZIP CRC, decompressed-size check, Matroska/EBML check, and structural inspection before the source bytes were deleted from the ephemeral runner.

Across K1–K11 the recovered USF derivative contains:

  • exactly 25 participant tracks per scenario with participant numbers 1..25;
  • the same A=13, B=12 participant-suffix distribution on every scenario;
  • gaze, fixation, timestamp, and point metadata in every participant payload;
  • 1920×1080 video at 25/1 fps on every scenario;
  • one to six semantically named rectangular AOI tracks per scenario.

Schöning et al. (10.16910/jemr.10.5.4) describe their USF representation as preserving complete gaze metadata without loss, whereas ASS retains selected metadata only. That statement supports selecting USF for recovery, but GazeForge does not infer that this specific conversion is field-identical to the unavailable original VISUS files.

The reviewed derivative-recovery record is:

validation/evidence/visus-source-recheck/
visus-osnabrueck-derivative-recovery-evidence-v1.json

with canonical fingerprint:

78b538ad35e411fe9e7020d47c759d2d6d246c96887bf76da979e23f35335d32

See VISUS Osnabrück derivative recovery for the exact run/artifact binding, per-scenario fingerprints, AOI inventory, and claim boundaries.

This substantially improves derivative recovery: the repository now has evidence of a historically institutional 11-scenario × 25-participant converted corpus, not merely small public fragments. It still does not satisfy the authoritative-current-copy roadmap item. The research-purpose statement is not treated as a formal dataset license; the converter article's CC BY 4.0 license is not projected onto dataset files; exact original-file identity, formal analysis rights, redistribution rights, AOI identity to original ViPER XML, and annotation independence remain unresolved.

Validate the checkpoint

The baseline checkpoint has a dedicated JSON-only validator:

gazeforge-visus-source-resolution \
  validation/protocols/visus-source-resolution-2026-09-04.json

The validator checks the record type, benchmark identity, ISO check date, current-source/audit/empirical flags, publication DOI, rights fields, annotation-independence gate, explicit claim limits, and a deterministic canonical SHA-256 fingerprint. The same generic validator also applies to the 2026-09-09 recheck file, while gazeforge.visus_authoritative_source_recheck.validate_visus_authoritative_source_recheck adds the frozen current-search and derivative-binding contract.

The Osnabrück derivative checkpoint has its own strict validator, gazeforge.visus_osnabrueck_derivative_recovery.validate_visus_osnabrueck_derivative_recovery. It revalidates the existing authoritative-source and supplement-recovery checkpoints, binds the exact successful reconnaissance artifact and four probe hashes, recomputes every scenario structural fingerprint, and keeps authority/rights/empirical promotion fail-closed.

The current unresolved status remains fail-closed: structural recovery of a historical derivative cannot simultaneously claim that a current authoritative original download was found, that the source is audit-ready, that empirical evidence was created, that analysis or redistribution rights are resolved, or that independent human annotation streams are verified.

This validation is still not a source audit. It verifies the internal integrity and conservative semantics of the source-resolution and derivative-recovery records only.

The ACM paper contains its publication copyright/permissions notice. GazeForge does not reinterpret that notice as a license covering the benchmark's raw video, gaze, or AOI files. Similarly, the paper's description of the dataset as publicly available establishes historical availability, not unrestricted redistribution.

The Osnabrück page's research-purpose statement likewise documents distribution intent rather than a complete modern license grant. The JEMR converter article is CC BY 4.0, but that article license is not treated as licensing the converted VISUS files.

Analysis-use permission and raw-file redistribution remain separate evidence fields and must be established explicitly for the exact copy used in an empirical run.

Annotation independence remains unverified

The paper reports that dynamic AOI annotation was performed manually by two contributors to improve annotation quality. It further explains that the first contributor performed the main annotation and the second made additional annotations and refinements.

That workflow is not evidence of two independently produced annotation streams. GazeForge therefore keeps:

  • independent_annotation_streams_verified=false;
  • human_human_agreement_ready=false.

The recovered USF derivative contains a single semantic AOI-track set per standard scenario. That does not create a second independent human annotation stream. Some derivative track titles also differ textually from labels shown in the original paper, so derivative AOI identity to the original ViPER XML remains unproven.

Only an authoritative obtained copy that contains separately recoverable streams, together with evidence that they were produced independently, can open the human-human agreement gate.

What must happen next

The next empirical step is not another model run. The Osnabrück recovery makes the remaining source task more precise:

  1. obtain the benchmark from a current VISUS/author-verified institutional source, receive an author-verified original copy, or recover an authoritative redeposit tied to the 2014 benchmark;
  2. document current analysis-use and raw-data redistribution terms separately for that exact copy;
  3. inventory and hash every file in the authoritative copy;
  4. compare its stimuli, gaze exports, participant identities, and AOI annotations with the recovered Osnabrück USF derivative;
  5. determine whether independent annotation streams actually exist;
  6. only then execute the canonical human-AOI intake, documented model prediction intake, externally supplied evaluation grid, and Frozen Evidence workflow.

Until those steps are complete, the recovered Osnabrück corpus is strong derivative evidence and the existing VISUS software remains validated infrastructure with empirical execution pending.

Public sources used for the resolution checkpoints

These references document the resolution decision. They do not replace the exact-file source audit required for empirical evidence.