A submission is a dated price observation for a defined unit of compute — the fields of the specification template: accelerator class and memory configuration, interconnect and node size, contract basis (on-demand · reserved · committed-term · spot marketplace), term, minimum commitment, region, what the rate includes, currency and unit, and the date the price was bookable or transacted. A concluded transaction (tier 1) is the most valuable input; a firm, bookable ask or bid (tier 2) is next; commentary is not a submission.
What it is not.Not a view on where the market is going, not a marketing rate card, not a request for The Grid to publish anything on the submitter’s behalf. A submitted price is an input to a provider-pooled level, never a quote published in its own name.
Most compute supply agreements carry confidentiality terms written before an independent price-reporting publication for compute existed. The clause below is the fix: model language either party can put in front of counsel so the agreement itself permits confidential disclosure to an independent assessor — on terms that require exactly the discipline The Grid publishes (no venue, no counterparty interest, aggregates only, minimum-observation withholding rather than exposure).
Benchmark Submission Carve-out. Notwithstanding any confidentiality obligation in this Agreement, either party may disclose the commercial terms of this Agreement (including price, quantity, term, and delivery characteristics) on a confidential basis to an independent price-reporting publication for the sole purpose of contributing to aggregated market price assessments, provided that: (i) the publication operates no venue for, and is not a counterparty or affiliate of a counterparty to, any transaction in the products it assesses; (ii) disclosure is made under a written confidentiality undertaking no less protective than this Agreement, which prohibits attribution of any submitted term to either party; and (iii) submitted terms appear only in aggregates from which neither party's individual terms can be derived, subject to minimum-observation rules under which the publication withholds publication rather than expose an identifiable submission.
Every condition the clause imposes on the publication is one The Grid already operates in public: it runs no venue and holds no position (/standards/terms); submissions are held under the confidentiality model in §Ⅳ below; and the minimum-observation rules that withhold publication rather than expose a thin panel are live on the daily print today (§Ⅴ). A counterparty’s counsel can verify each condition against the published record before any term is disclosed.
Submissions go to deals@thegridco.ai. On receipt, a submission is timestamped (received-at, UTC), hashed (the payload as received, so it can later be shown unaltered), assigned a pseudonymous submitter identifier, and acknowledged to the sender with the timestamp and hash. The receipt record is append-only: it is never edited, and a withdrawn or amended submission is a new record referencing the old one. This is the first of the audit artifacts the capability specification names — the chain from every published number back to its inputs begins here.
The Grid follows the panel model of the Baltic Exchange rather than the open-window model: submissions are confidential. A submitted price is never attributed to its submitter, never published individually, and never disclosed except by order of a court or regulator. Panellists are never named— before or after a panel exists. What is published is the aggregate: the level, the panel size, the count of contributors and their data tier on the provenance stamp, and — once a panel exists — its composition by class (buy-side · sell-side · intermediary), never by name.
The Grid will never require a submitter to change a submitted price, and never impose one. Validation, once a panel exists, is a recorded disposition — included, excluded, or benched — with its reason code, and a submitter is told when their data is excluded and why (the validation log is specified in §Ⅵ below and is not yet built).
The assessment specification (Part C4) adopts the thresholds the Baltic Exchange publishes for its own panel-based benchmarks. They are the numbers that decide when a level is created, when it is published, and what a thin panel must disclose:
| Threshold | Rule | Source |
|---|---|---|
| Create an index | Do not create a route index without a panel of at least five panellists — three in the Asia variant. | Baltic Exchange, Guide to Market Benchmarks / Guide to BEA Indices |
| Publish | Do not publish where fewer than four can contribute — two in the Asia variant. | same |
| Forward curves | May run on as few as two, provided the less reliable nature of the data is drawn to the attention of the market. | same |
| Calculation default | Arithmetic average of validated input; market-share weighting is the exception. | same |
| Graduated sanction | Operational benching — a panellist keeps contributing but their data is excluded from the calculation. | same |
Where GAP-H100 stands against them. The panel holds two nominal contributors, one static for its entire stored history — below every variant of the floor, which is why the series is an indicator today. The adopted minimum-observation rule (binding from the 2026-08-22 print) applies the Asia-variant publication floor: fewer than two contributors, or every contributor unchanged for 7 or more days, and the print is withheld — disclosed with its provenance stamp, never a silent gap. The path back to assessment class is at least two independently movingcontributors; a target of three submitters is the Baltic’s own Asian minimum for creating an index, not an arbitrary number.
The rules below are drafted in the assessment specification and are being settled with the first market participants before they are published as binding. Until each is published here, it is not a rule The Grid claims to operate: