Vendor-neutral evidence for your government procurement bid

A quantum-device decision that cannot be re-derived from a recorded input + snapshot pair is, from an audit standpoint, indistinguishable from a guess. In regulated procurement, the officer signing off on the decision is often not the engineer who ran it; six months later the question "why did we pick this vendor" has to be answerable from paperwork alone.

Qlro's record format answers that question with three coupled artifacts: the frozen benchmark dataset (via snapshot_commit), the decision itself (content-hashed), and an independent archival reference (Zenodo DOI). Each is individually verifiable; together they form a reproducible evidence chain suitable for internal governance and external audit.

Three concrete workflows, each ending in a tender/grant submission packet that contains our decision record as third-party evidence. The client organisation is the counterparty to the government; Qlro is the evidence issuer.

1. 한국 정부 R&D 응찰 (KISTEP / KIAT / DAPA). Bidder runs qlro recommend on the candidate workload, generates the decision record, downloads the procurement bundle ZIP. The bundle's record.pdf is attached to the proposal package as vendor-selection evidence; the verifier-instructions.txt tells the KISTEP evaluator how to re-hash and re-derive the recommendation.
2. USA — DoE / NSF / DARPA / NQI proposals. Bidder cites the WCPP paper DOI in references, attaches the record.pdf to the technical volume, and points evaluators at the public verifier instructions. The Zenodo DOI gives the proposal a permanent academic citation that survives past project end.
3. EU — Horizon Europe / Quantum Flagship. The reproducibility-first review process is the strongest fit for this evidence model. Bidder includes record.pdf plus the GitHub URL in the Implementation section under Open Science / Reproducibility. The evaluator can replicate the recommendation independently in under 5 minutes.

Vendor-published benchmark rankings cannot serve as procurement evidence because their baseline changes when the vendor changes their methodology, and every vendor's top pick is (tautologically) their own hardware. A neutral scoring framework with a locality axiom — the property that adding or removing a vendor cannot change any other vendor's score — is the minimum bar for multi-vendor procurement comparison.

Qlro's Workload-Conditioned Physical Projection (WCPP) framework is the first published scoring system to prove this property rigorously from a single locality axiom. Full paper (peer-facing revision v1.2) at Zenodo DOI 10.5281/zenodo.19785800.

Five extensions stack on top of the base Enterprise contract. Each is contracted separately so customers pay for what they need. Indicative annual values; final pricing depends on fleet size, workload volume, and security regime.

Worked example: a Korea-Quantum government pilot taking Base Enterprise + Private Deployment + Audit + Procurement Advisory sits at $230K ACV. A multi-vendor R&D fleet taking Base + Drift Monitoring + Error Mitigation Advisory sits at $250K ACV.

Year 1 (decision-support). Today. Per-workload recommendations and audit records. ACV $50K–$80K.

Year 2 (operations layer).The CSE forward model inverts naturally into a mitigation-policy generator: given a target circuit and a target device, return the cheapest mitigation stack that clears the workload's fidelity floor. ACV $150K–$250K.

Year 3 (compliance + insurance layer). Once governments and regulated industries adopt quantum results into decision pipelines, audit-trail and accuracy-SLA become contract requirements. The r(τ) decay curve becomes an SLA primitive; insurance underwriters use it to bound expected error. ACV $300K–$500K.

Qlro is a decision-support layer. We do not execute circuits on quantum hardware, we do not guarantee the vendor of choice will meet its published performance on any specific workload, and we do not underwrite the procurement outcome. The scope of our guarantee is strict: the record is a faithful snapshot of what the model recommended under the stated inputs, and the record itself is tamper-evident.

We publish negative results about our own estimator alongside positive ones — the v1.2 paper's Appendix D.1 reports a failed extension of our calibration into zero-noise extrapolation. We believe this is the minimum epistemic standard a decision-support product should hold itself to; ask if you want the full list.