Reference Procurement Language¶
Drop-in contract clauses that make a VIDS PASS the acceptance condition for medical imaging dataset deliveries.
Free to use, adapt, and redistribute under CC BY 4.0. No fee, no registration, no permission required.
The problem this solves¶
Buyers of annotated medical imaging datasets currently have no objective, automatable way to verify what they paid for. Two datasets that look identical in a folder listing can differ substantially in annotation provenance, quality documentation, and downstream usability.
Without an objective acceptance criterion, dataset quality is typically assessed after integration begins — when remediation is most expensive and timelines are already committed.
VIDS provides a machine-checkable acceptance standard. The reference validator runs on the delivered dataset and produces a binary pass/fail result. There is no subjective interpretation, no checklist exchange, no negotiation about what counts as documented.
The clauses below give procurement and contract teams the language to make a VIDS PASS the acceptance condition.
Recommended action¶
Include a VIDS validation requirement as an acceptance condition in your next dataset procurement.
- Decide which Profile to require. POC for prototypes and internal research; Full for production, regulatory, or external-customer-facing use. See Profile selection below.
- Copy Clauses 1–5 into your statement of work, master service agreement, or purchase order. Clauses 6 and 7 are situational.
- Fill in the bracketed placeholders — Profile choice, remediation timeline, and any optional terms.
- Send the SOW to your vendor with a link to vidsstandard.org so they can prepare. Most vendors with mature data pipelines reach VIDS compliance within one to two weeks for POC and three to four weeks for Full.
The clauses are short enough to fit on one page for legal review and were drafted to be readable by procurement and engineering leads without specialist counsel.
Reference clauses¶
The clauses below may be used as-is, modified, or extracted individually. Bracketed placeholders should be replaced with the values appropriate to the agreement.
Clause 1 — Compliance Requirement¶
All annotated medical imaging datasets delivered under this Agreement shall conform to the Verified Imaging Dataset Standard (VIDS), version [1.0] or later, as published at https://vidsstandard.org.
The compliance Profile required for delivery is: [ POC / Full ].
Conformance shall be determined exclusively by the open-source VIDS Reference Validator. A dataset is compliant if and only if the Reference Validator reports zero failed rules at the specified Profile.
Clause 2 — Validation Report Delivery¶
Vendor shall include with each dataset delivery:
(a) The complete dataset organized per the VIDS Specification, including all required metadata files, annotation sidecars, and provenance documentation.
(b) A Validation Report generated by the Reference Validator, in JSON format, demonstrating the dataset's compliance status. The Validation Report shall be generated using the validator version current at the time of delivery and shall be included in the delivery archive.
© For Full Profile deliveries, the Validation Report shall report zero failed rules across all 21 validation rules.
Clause 3 — Acceptance Criteria¶
Buyer's acceptance of any dataset delivery is conditional upon:
(a) Successful VIDS validation as evidenced by the Vendor-supplied Validation Report.
(b) Buyer's independent re-validation of the delivered dataset using the Reference Validator, with results matching the Vendor-supplied report.
© Compliance with all other terms of this Agreement.
A dataset that fails VIDS validation does not meet acceptance criteria, regardless of any other quality attributes claimed by Vendor.
Clause 4 — Remediation¶
If a delivered dataset does not pass VIDS validation:
(a) Buyer shall provide Vendor with the Validation Report identifying failed rules and the dataset components affected.
(b) Vendor shall remediate identified failures and re-deliver the dataset within [ N ] business days at no additional cost to Buyer.
© Persistent failure of the same rules across two consecutive re-deliveries constitutes a material breach of this Agreement.
Clause 5 — Final Acceptance¶
Final acceptance of the dataset shall occur upon:
(a) Vendor's delivery of a Compliant Dataset at the required Profile;
(b) Buyer's independent re-validation confirming the same result;
© Buyer's review of provenance and quality documentation included in the delivery.
Final acceptance shall not be unreasonably withheld where the dataset meets all stated criteria.
Clause 6 — Specification Version Control (situational)¶
The applicable VIDS Specification version for this Agreement is version [ 1.0 ]. Subsequent VIDS Specification revisions shall not retroactively apply to deliveries already accepted under this Agreement unless expressly agreed in writing by both parties.
Where the Reference Validator version used by Vendor and Buyer differ, the more recent stable release shall control, provided both parties have access to the same validator version.
Clause 7 — Audit Rights (optional)¶
Buyer reserves the right to audit any dataset delivered under this Agreement for VIDS compliance at any time during the Agreement term and for [ N ] years following final acceptance. Such audit shall be limited to running the Reference Validator and reviewing the resulting report. Vendor shall provide reasonable cooperation for any such audit, including access to provenance documentation as required.
How procurement changes when VIDS is the acceptance criterion¶
| Step | Current process | With VIDS in the SOW |
|---|---|---|
| Delivery | Folder of files arrives, vendor asserts compliance | Folder of files plus Validation Report (JSON) |
| Initial review | Manual inspection, days to weeks | Validator runs, seconds to minutes |
| Acceptance basis | Subjective: does this look right? | Objective: zero failed rules at required Profile |
| Disputes | Vendor and Buyer interpretations | Reference Validator output |
| Remediation timing | After integration begins | At acceptance, before integration |
| Audit trail | Email threads, notes | Machine-readable Validation Report retained with delivery |
Definitions¶
For purposes of these clauses, the following definitions apply.
VIDS means the Verified Imaging Dataset Standard, an open specification published at https://vidsstandard.org and licensed under CC BY 4.0.
VIDS Specification means the document at https://vidsstandard.org/specification/ (or the version specified in Clause 6).
Reference Validator means the open-source software published as vids-validator on PyPI (https://pypi.org/project/vids-validator) and at https://github.com/vids-standard/vids-standard.
Profile means a defined level of VIDS compliance — either "POC" (15 rules) or "Full" (21 rules) — as defined in Section 13 of the VIDS Specification.
Validation Report means the JSON output produced by the Reference Validator when run against a dataset, indicating per-rule PASS, FAIL, WARN, or SKIP status.
Compliant Dataset means a dataset that, when evaluated by the Reference Validator at the required Profile, produces zero FAIL results.
Profile selection¶
| Use case | Recommended profile | Rationale |
|---|---|---|
| Internal R&D, prototyping | POC | Speed of iteration takes priority over full audit trail |
| External sale to AI customers | Full | Customers will demand the audit trail downstream |
| FDA / EU AI Act / CDSCO submission | Full | Provenance required for technical files |
| Academic publication dataset | Full | Inter-annotator agreement and quality docs required for credibility |
| Pilot studies | POC | Often time-constrained; can graduate to Full for production |
| Multi-vendor dataset acquisition | Full | Cross-vendor consistency requires the full standard |
POC enforces 15 of 21 rules — structure, imaging, and annotation provenance. Full adds quality documentation and ML readiness rules. Specifying POC does not preclude later upgrade to Full if requirements evolve.
Common questions¶
We already have our own dataset acceptance criteria. Why use VIDS?
VIDS does not replace your criteria. It standardizes the structural and provenance layer. Domain-specific acceptance criteria — minimum subject count, modality coverage, label accuracy — layer on top of VIDS compliance. VIDS answers "is what we received auditable?" Existing criteria answer "is what we received useful?"
What if our vendor has never heard of VIDS?
Point them to vidsstandard.org. The validator is one command: pip install vids-validator. The specification is on a single page. Most vendors with mature data pipelines reach POC compliance in one to two weeks and Full compliance in three to four weeks. Including VIDS in the SOW gives them the runway to prepare.
Does this lock us into VIDS or any single supplier?
No. VIDS is an open standard (CC BY 4.0 specification, Apache 2.0 tools) with no central licensor. The dataset structure can be exported to any downstream framework — nnU-Net, MONAI, COCO, flat NIfTI — without losing the underlying data or provenance.
What if partial compliance is acceptable for our use case?
Use the POC profile. POC enforces 15 of 21 rules. The rules that POC skips (quality documentation and ML splits) can remain buyer-discretionary in the SOW. Specific Full-profile rules can also be marked as WARN-acceptable in the contract if the use case justifies it.
Can we modify these clauses?
Yes. This document is CC BY 4.0 — adapt freely. We recommend keeping the validator-as-arbiter language intact, since that is what makes the standard enforceable rather than opinion-based. Removing the binary pass/fail mechanism converts VIDS back into a checklist exchange.
What about confidentiality, data rights, and IP?
Out of scope for this document. Those concerns are between Buyer and Vendor and should be governed by the existing master service agreement or NDAs. VIDS clauses sit alongside, not in place of, those terms.
What does it cost?
The specification, validator, examples, and reference dataset are free. Princeton Medical Systems and other implementers offer paid services around VIDS — curation, compliance auditing, training — but those are optional. Using VIDS as a procurement standard does not require purchasing anything.
References¶
- VIDS Specification v1.0 — full normative text
- Validation Rules — the 21 rules in detail
- Compliance Profiles — POC vs Full comparison
- Validator on PyPI —
pip install vids-validator - arXiv preprint (2604.17525) — academic paper
- LIDC-Hybrid-100 reference dataset — a working example of a 21/21 PASS dataset
- VIDS benchmarks — compliance scoring of major public datasets
License¶
This document — VIDS Reference Procurement Language — is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0). It may be used, adapted, and redistributed freely, including for commercial use, provided attribution is preserved.
Suggested attribution: "Adapted from the VIDS Reference Procurement Language v1.0, available at https://vidsstandard.org/for-buyers/sow-addendum/."
The VIDS Specification and Reference Validator are separately licensed (CC BY 4.0 and Apache 2.0 respectively).
Related¶
- Validation Attestation — the artifact a vendor delivers as proof of compliance under these clauses
- Compliance Evaluation — diagnostic assessment for buyers who want to evaluate a vendor's sample delivery before contracting
- Scoring Rubric — the 22-dimension framework that underlies validator results
- VIDS Specification — the technical standard referenced by Clause 1