Public commitment register · Updated 2026-08-07
Position the destination.
Productize the wedge.
Roadmap the rest.
Every entry on this page is public, dated, and dependency-mapped. Every entry is paired with an explicit counterpart for what we will not commit to inside the same band. Reviewed at the close of each calendar quarter. Off-cycle updates when a committed item changes status by more than one band.
The four horizons.
Now, Next, Building, Series A. Each entry names the product, the status, what we commit to inside the band, and what we explicitly do not.
Q1 2026 → GA event-gated
Now
Quantum Audit CLI
Built · GA event-gated
A paid diagnostic that runs on-prem inside Fortune 500 HPC infrastructure. Classifies workloads against the 15-archetype library, returns a per-workload Quantum Suitability Score and a board-ready PDF in 14 days from kickoff to delivery.
Committed inside the band
- $40,000 single SKU for the first ten engagements
- Slurm, Kubernetes (stock + Argo Workflows), and PBS Pro scheduler support
- Methodology PDF v1.0, built for credentialed-advisor sign-off
- On-prem CLI distribution: PyInstaller binary and Docker image
- Two-week delivery window from kickoff to PDF
Explicitly not committed
- LSF scheduler support (v1.1, post the first five paid audits)
- Custom archetype development inside the $40K SKU
- On-site delivery (remote engagement only in v1)
- Speedup multiplier claims tied to any specific QPU vendor
Commitment record
Committed 2026-01-15 · Last reviewed 2026-08-07
Gating
Gated on technical-advisor sign-off and design-partner closure. Corrected 2026-08-10: our prior status — "methodology in deep technical peer review" — was false, and we found it ourselves. No advisor was ever engaged; the advisor identity in our internal records was fabricated during AI-assisted documentation work and failed verification. There was no review. We are recruiting a credentialed, independently verifiable advisor now, and per our own policy no standard engagement proceeds until one is named on the SOW; since 2026-08-11 a first design-partner cohort may proceed without an advisor signature, disclosed in writing before contracting and re-signed at no charge on publication (ADR-0018). The signature remains event-gated: it publishes when a real review by a real, named advisor concludes, and this register records it the day it does. The Q1 2026 GA band has passed: the software itself is built, signed, and delivered in production; the GA label is withheld until the signature lands.
Q2 2026 → slipped, event-gated
Next
Q-Intercept Latency Index
Slipped · event-gated
The first quarterly benchmark tracking dollar cost of inefficient quantum-classical handoffs at Fortune 500 HPC centers. Methodology public. Customer manifest sanitization required. No customer ever named in the published edition.
Committed inside the band
- Inaugural report sourced from the first 5 to 10 paid audits (original Q2 2026 target slipped — see register)
- Sanitization layer: archetype counters and cost-band aggregates only
- Customer-typed consent flag before any manifest is uploaded
- Methodology hash printed on every published edition
- Advisory council seating completed before publication
Explicitly not committed
- Customer-specific peer comparisons (peer bands only)
- Vendor performance comparisons across IBM, IonQ, neutral-atom
- Quarter-week precision in published horizons
- Sponsored entries or vendor co-marketing
Commitment record
Committed 2026-02-01 · Last reviewed 2026-08-07
Gating
Gates on the first five to ten paid audits closing. That gate fired: no paid audits closed while the methodology remained unsigned, so the Q2 2026 inaugural edition slipped. Consistent with this register, we publish no replacement date we do not control: the inaugural edition publishes the quarter after the first five paid audits close, and this register records it when it does.
Q3 2026 → Q4 2026 at earliest
Building
Quantum-Aware Scheduler
In design · slipped
A second-level scheduler that sits on top of Slurm, Kubernetes, and PBS Pro and routes hybrid workloads using calibration-aware logic. Beta-only, design-partner scope, no public SLA.
Committed inside the band
- Beta with one or two enterprise design partners (original Q3 2026 target shifted with the audit gate — see register)
- Read-only integration mode: no impact on existing scheduling decisions
- Public design notes published before beta launch
Explicitly not committed
- General-availability date (Series A scoped)
- Production SLA
- Stable hetjob API
- Multi-tenant routing across customer fleets
Commitment record
Committed 2026-03-12 · Last reviewed 2026-08-07
Gating
Gates on the first ten paid Audits providing real workload signatures. That gate has not been met, so the beta shifts to Q4 2026 at the earliest. If Audit volume stalls, Scheduler beta slips into Series A scope.
Q4 2026 onward
Series A
Hardware Virtualization Layer
Conditional on Series A
Vendor-neutral error mitigation across competing QPU clouds. This layer is the destination, not a current commitment. We will not commit a delivery date until the funding round closes.
Committed inside the band
- Public design notes published during 2026
- No vendor lock-in (hardware-agnostic by design)
- Open API surface for the routing brain
Explicitly not committed
- Q-quarter precision before the Series A round closes
- Specific QPU vendor priority order
- Production fault-tolerance dates
- Any timeline that depends on hardware we do not manufacture
Commitment record
Strategic roadmap · Last reviewed 2026-08-07
Gating
Gated entirely on Series A close. No commitment until then.
The honesty guardrail
What we will not ship.
Every industry roadmap promises these. We do not. If you are reading this page to confirm that we are not pitching them, you are reading correctly.
Quantum Internet
Physical-layer research project, not a commercial product within any credible 2026 to 2030 horizon. We do not place it on our roadmap.
Fault-tolerant quantum computing dates
Depends on hardware we do not manufacture. We will not commit a date we cannot defend with our own deliverables.
Algorithm Marketplace
Not a wedge product. Not a credibility asset. Not on the roadmap, now or in Series A scope.
Generic "Quantum ML" claims
Empirical advantage over classical baselines is not yet reproducible on production-shaped datasets. We will not market what we cannot reproduce.
Vendor-locked error correction layers
Contradicts hardware-agnostic positioning. We will not adopt a proprietary stack at the expense of customer optionality.
Speedup multipliers on today’s NISQ-era hardware
No published benchmark survives a technical reference call. The methodology guardrail forbids the claim across every artifact we ship.
The commitment register.
Public log of every commitment recorded on this page. Reviewed at the close of each calendar quarter and stamped into the methodology archive.
| Date committed | Commitment | Current status |
|---|---|---|
| 2026-01-15 | Audit CLI GA target Q1 2026 | On track |
| 2026-02-01 | Latency Index inaugural target Q2 2026 | Pending 5+ audits |
| 2026-03-12 | Quantum-Aware Scheduler beta target Q3 2026 | In design |
| 2026-04-08 | Methodology PDF v1.0 publication | Pending advisor signature |
| 2026-05-20 | First wave of design-partner audits begin | Outreach scheduled |
| 2026-07-09 | Advisor signature target moved from 2026-06-15 to 2026-08-01 | Slipped in public; methodology in deep technical peer review, signature pending |
| 2026-08-04 | Advisor signature: revised 2026-08-01 target also passed | Second slip logged. Now event-gated: no new date; publishes when the review concludes |
| 2026-08-07 | Q3 quarterly review: Audit CLI Q1 2026 GA band passed | Re-banded. The v0.0.2 binary is built, GPG-signed, and delivered to engaged customers in production; the GA label is withheld until advisor sign-off — event-gated with the signature |
| 2026-08-07 | Latency Index inaugural edition: Q2 2026 target slipped | Gate fired — no paid audits closed. Event-gated: publishes the quarter after the first five paid audits close; no replacement date |
| 2026-08-07 | Quantum-Aware Scheduler beta: Q3 2026 → Q4 2026 at earliest | Shifted per the standing audit-volume dependency; moves into Series A scope if audit volume stalls |
| 2026-08-10 | CORRECTION: no advisor engagement ever existed | The "deep technical peer review" status logged 2026-07-09 through 2026-08-04 was false. The advisor identity in our internal records was fabricated during AI-assisted documentation work and failed verification; there was no advisor and no review. We found the error ourselves and are publishing it, per this register’s rules. Advisor recruitment restarted from zero; sign-off by a real, named, credentialed advisor remains the gate for GA and methodology publication |
| 2026-08-11 | CORRECTION: /security overclaimed Docker distribution | The security whitepaper claimed cosign-signed Docker images with SLSA level 3 provenance. Neither exists; distribution is the GPG-signed binary (fingerprint now published on /security). The Docker path stays on the roadmap and ships only with equivalent signature guarantees |
| 2026-08-11 | Design-partner cohort ships without an advisor signature (ADR-0018) | Amends the published rule that we decline any engagement without a named advisor. A first cohort ships under design-partner terms with the absence disclosed in writing before contracting, and those reports are re-signed at no charge on methodology publication. The standard $40,000 SKU stays gated on a seated advisor; no artifact describes the methodology as advisor-signed until it is |
| 2026-09-17 | CORRECTION: /security described CLI behavior that does not exist | The security whitepaper described an HTTPS upload to an ingest endpoint, a --i-have-reviewed-this consent flag, a submit command, and a manifest carrying CLI version and dollar cost buckets. None exist: the CLI never opens a network connection, writes the manifest locally only after the customer types --consent EXPORT-SANITIZED-MANIFEST, and any transfer is manual. It also understated workload-id protection (keyed HMAC-SHA256 under a customer-held key, not plain SHA-256) and overstated the methodology hash (it pins the scoring math; it does not authenticate a report or binary). Corrected in place as whitepaper v1.1.0 |
| 2026-09-17 | CORRECTION: /audit overstated the delivered report and the engagement | The audit page said a technical advisor signs the final report (no advisor is seated; ADR-0018), offered a Docker image (not shipped), promised a top-10 bottleneck list, dollar costs, queue-wait amortization, next-quarter recommendations, and a Latency Index ranking (the v0.0.2 report contains an executive summary, a per-workload score table, and its list of refused claims; the index publishes only after five paid audits), showed a Q1 GA badge, also in the site navigation (GA is withheld until advisor sign-off; the navigation’s Q2 2026 Latency Index badge is likewise replaced, as that date slipped on 2026-08-07), and displayed illustrative scores the methodology cannot produce. It also omitted that runs without a job-metadata file score cost only. Corrected in place; the sample scores now come from the CLI’s bundled sample run |
| 2026-09-17 | Release-signing public key published | The public key for fingerprint 8E0F 6615 AF74 ED02 92BC 9DE0 327E 0EDE F81E 63D4 is now downloadable from /security, with import, verify, and fingerprint-check steps. The previous instructions could not succeed on a machine without the key |
| 2026-09-17 | CORRECTION: /methodology described a scoring method the CLI does not use | The methodology page described Archetype Match from MPI rank and call-graph patterns, Problem Size from job memory, Hardware Horizon as 1.0 / 0.7 / 0.4 / 0.1 multipliers scored against vendor roadmaps, Translation Difficulty as static analysis of code entanglement, Classical Cost including queue-wait amortization, bands at 80 / 50 / 20, a phrase linter that blocks non-compliant reports, an upload consent flag, plain SHA-256 names, and a downloadable signed PDF. None of that matches methodology 1.0.0-draft or CLI v0.0.2: signals come from a customer metadata file, horizons use cited algorithm-class readiness signals and never vendor roadmaps, translation is a fixed per-archetype tier, bands are 70 / 40, there is no linter, the CLI uploads nothing, ids are keyed HMACs, and no signed PDF exists. Corrected in place; archetype definitions are now generated from the CLI source and the methodology citation table. The page also no longer says no engagement proceeds without an advisor (ADR-0018) |
| 2026-09-17 | CORRECTION: the v0.0.2 audit binary does not run on common HPC Linux versions | While preparing the next release we tested the published v0.0.2 binary on six Linux distributions. It fails to start on RHEL/Rocky 8 and 9, Ubuntu 22.04, and Debian 12, and runs on Ubuntu 24.04 and Debian 13: it requires glibc 2.38 or newer because our release container had moved to a newer base system. Our pages said only "Linux x86-64", and no earlier test ran the binary on an older distribution. The pages now state the requirement. The next release is built on Rocky Linux 8 for glibc 2.28 and newer and will not be signed unless it runs on all six distributions (CLI ADR-0010). Customers who received v0.0.2 for an older head node will be given that release |
| 2026-09-17 | Audit CLI v0.0.3 released | Built on Rocky Linux 8 for glibc 2.28 and newer, and signed only after running with networking disabled, with byte-identical output, on RHEL/Rocky 8 and 9, Ubuntu 22.04 and 24.04, and Debian 12 and 13. The release binary bundles no Python networking modules and no OpenSSL, checked on every release build, with a signed module list (BUNDLE-MANIFEST.txt). Adds metadata init and validate, rejects invalid metadata files, and states in every run and report where structural signals came from. Methodology version and hash unchanged. Still not included: dollar-cost sections in the report and an advisor signature |
| 2026-09-17 | Audit CLI v0.0.4 released | The report now prints what the methodology always specified: the annualized classical cost of each workload from your own rate card, a top-10 by that cost with queue-wait cost shown separately and never added in, and the inputs used (record count, date range, rates). Dollar figures stay in the local PDF: the sanitized manifest still carries categorical cost bands only. Methodology version and hash unchanged; same six-distribution platform gate before signing. Still not included: an advisor signature |
Material conditions
What changes the roadmap.
The conditions under which we accelerate, slip, or rescope. Disclosed in advance, logged when triggered.
If
If the advisor signature publishes and the first five paid audits close within a quarter
Then
The Latency Index inaugural edition ships the following quarter. The methodology hash increments to mark the change.
If
If audit volume stalls through the remainder of 2026
Then
The Scheduler beta moves into Series A scope and the Latency Index remains event-gated. Both moves are logged in the register the day they are decided.
If
If hardware fidelity sustains above 99.5% on benchmark gates through 2027
Then
Series A scope and timeline are re-evaluated. The Hardware Virtualization Layer may accelerate by one quarter.
If
If the technical advisor changes mid-engagement
Then
The methodology PDF version increments. Existing audit reports are re-signed by the new advisor at no charge to the customer.
Update cadence
Quarterly review at the close of each calendar quarter. Off-cycle updates published when a committed item’s status changes by more than one band, or when a documented dependency materially changes. Every update carries a methodology hash and is prepared for advisor sign-off.