Oracle versus index, and why Rillor uses both terms.

New partners reading Rillor's documentation hit two words for what looks like one thing. We publish the Rillor Compute Index, a daily forward reference price per SKU. We also describe an oracle that derives that price. People reasonably ask whether these are the same object under two names, whether one is the marketing word and the other the real one, or whether we are quietly running two products. The answer is cleaner than the confusion suggests, and getting it right changes how you describe Rillor to your own stakeholders.

The Index is the product. The oracle is the machine that produces it. The Index is what a fund quotes, what an exchange references in a contract specification, and what shows up in market-facing copy. The oracle is the price-derivation engine that ingests live Rillor forward contracts, applies a fixed methodology, and emits one auditable value per SKU. Same value, two vantage points. You say Index when you mean the named, branded reference price. You say oracle when you mean the integration surface and the derivation logic. Use the wrong term in the wrong room and you will either confuse a portfolio manager with plumbing or hand an engineer a brand name where they needed an endpoint.

What the Index is

A named reference price is a controlled, methodology-defined product, not a raw data dump. That is the distinction IOSCO drew when it published its Principles for Financial Benchmarks in 2013: a benchmark administrator governs methodology, quality, and accountability, and is expected to publicly disclose its compliance. The administrators that operate under those principles are exactly the names you would expect, the equity-index houses such as S&P Dow Jones, MSCI, and FTSE, and the commodity price reporting agencies such as Platts and Argus. A benchmark in that lineage is proprietary, owned, and licensed. It is a product with a name on it.

The Rillor Compute Index is that kind of object for AI compute hardware. For each SKU we maintain, it is a single forward reference price, refreshed daily, carrying a defined methodology and a Rillor name. RIL-GX-B300-2T has an Index value. So does RIL-NVL72-GB300, and RIL-H200-2T, and the rest of the catalog. When a fund hedges compute exposure or an exchange writes a contract specification that says "final settlement against the Rillor Compute Index for RIL-GX-B200-2T," the Index is the noun they are pointing at. It is the public-facing reference, and it is the thing we license.

The index is the moat, not the marketplace. That holds because the value of a benchmark is its acceptance: once funds quote it and venues settle against it, the named reference becomes the coordination point for an entire market. You can see the catalog of priced instruments on the SKU pages and the live forward tape on the marketplace. The licensing model and the audiences that reference the Index live on the for-markets page.

What the oracle is

The oracle is the engine that turns active contracts into that one number. Its job is narrow and mechanical, and the narrowness is the point. It ingests live Rillor forward contracts, the executed, deposited, delivery-intended forwards on its book. It applies the 30-day rolling-blend methodology, which weights recent executions across the trailing window so that one large print on a thin day cannot jerk the curve. And it emits exactly one auditable value per SKU at 17:00 UTC each day. One input set, one rule set, one output, on a fixed clock.

That cutoff matters as much as the math. A benchmark is only useful if every consumer reads the same value at the same moment, so the oracle stamps each SKU once per session and freezes it. Downstream consumers do not poll a moving number and hope they all caught the same tick. They reference the 17:00 UTC fix. The mechanics of the blend, field by field, are covered in how the Rillor Compute Index is computed; the curve-formation logic that feeds it is in how a forward curve forms from real contracts.

The word oracle is borrowed deliberately. In on-chain systems, an oracle is the middleware that aggregates external data into a single reference value that a smart contract can query. Chainlink's data feeds, for example, aggregate inputs across a decentralized set of independent node operators, and off-chain nodes write one aggregated value into an on-chain contract that applications read on demand. That is structurally the same shape as what Rillor's oracle does: many underlying inputs collapse to one value that downstream code consumes. Borrowing the term tells an integrating engineer exactly what role our output plays in their stack. It is the value their contract reads.

Why the distinction matters downstream

Two different audiences consume the same number, and they reach for different words because they sit at different points in the pipeline.

Funds and exchanges reference the Index. A portfolio manager hedging GPU capex, or an exchange drafting a contract spec, cares about the named, governed benchmark and its methodology. They will write "the Rillor Compute Index" into a term sheet because that is the proprietary reference they are licensing and the name their counterparties will recognize. Brand and governance are the load-bearing properties for them. A fund treating compute as an asset class needs a reference it can cite in a mandate and reconcile against at period end, and the Index is built to be exactly that citable object.

Third-party venues building cash-settled products consume an oracle value. A team building a cash-settled perpetual or futures product against Rillor's price is wiring an endpoint into a settlement contract. For them the relevant object is the feed: a value their contract can read and settle against programmatically. They will call it the oracle, because in their architecture that is literally the oracle slot. Same number, different consumer, different word.

The reason to keep both terms precise is that they map cleanly onto these two consumption patterns. Say Index and a finance reader knows you mean the governed, licensed benchmark. Say oracle and an integration engineer knows you mean the value their code reads. Collapse the two and you blur a real boundary between what is branded and what is wired.

What the oracle is not

Two misreadings come up often enough to address head on.

The oracle is not a generic third-party price oracle network. A decentralized oracle network is middleware operated by independent node operators who source and aggregate data they do not own. Rillor's oracle is the opposite arrangement. It derives its value from one proprietary source, the active forward contracts on Rillor's own book, under a methodology Rillor wrote and controls. There is no committee of anonymous nodes and no external aggregation layer. It is a single source of truth, administered the way an index provider administers a benchmark, not the way a public oracle network relays third-party feeds. The on-chain analogy explains the role our output plays in someone else's contract; it does not describe our governance.

The oracle is not a cash-settlement engine, and Rillor never cash-settles. This is the distinction worth being most careful about, because it is where the brand could be misread as something it is not. The CFTC defines cash settlement as a method of settling derivatives whereby the seller pays the buyer the cash value of the underlying, or a cash amount based on the level of an index or price, rather than delivering the commodity. CME Group draws the same line: cash-settled contracts resolve at expiry to a payment against a final settlement price, while physically delivered contracts move the actual commodity between the maker and taker of delivery. Rillor's own contracts are the physical-delivery side of that line, every time. They are bilateral OTC forwards with intent of physical delivery, which is precisely the CFTC's forward-contract definition: a commercial buyer and seller agreeing on delivery of a specified quality and quantity at a specified future date. Every Rillor contract carries a 10 percent deposit at execution, a seller performance bond, balance due at delivery, and an end-customer-of-record captured for NVIDIA channel compliance. None of that is optional, and none of it resolves to a cash payment in lieu of hardware.

So where does cash settlement happen. Not on Rillor. Third-party venues that license the Index can build cash-settled derivatives, perps and futures, that settle against the oracle value. Rillor does not operate those venues. The oracle produces a price; what a licensed venue does with that price, including settling a contract to cash, happens on that venue's own books under its own rules. The oracle is a measurement, not a settlement. Keeping that straight is the difference between describing Rillor accurately and describing a product we deliberately do not run.

Ownership is the moat

Both the Index name and the oracle methodology are owned and controlled by Rillor. That is not a legal footnote, it is the entire strategic point. Under the IOSCO framework, a benchmark is a proprietary product: the administrator owns the methodology, governs it, and licenses it to third parties. Index providers operate exactly this way, which is why a named benchmark is an asset rather than a commodity.

Rillor sits in that same position for AI compute. The methodology, the 30-day rolling blend computed from real executed forwards, is ours. The name, the Rillor Compute Index, is ours. The single underlying source, the contracts on Rillor's book, is ours. A competitor cannot reconstruct the value without the contract flow, and cannot use the name without a license. That combination, proprietary input plus owned methodology plus a recognized brand, is what makes the Index defensible. The marketplace creates the data. The oracle refines it. The Index is the licensable product that sits on top. Venues, funds, and researchers reference it precisely because they cannot replicate it.

How the two terms map to the API

If you are integrating, the cleanest way to internalize the distinction is to look at the surface itself.

The feed is the oracle output. A call such as GET /v1/forward/curve/RIL-GX-B300-2T returns the derived forward curve and the 17:00 UTC fix for that SKU. That payload is the oracle speaking. It is the value your smart contract reads, or the value your settlement script pulls. Engineers writing against this should say oracle, because that is what they are wiring: the price-derivation engine's published value.

The brand on that output is the Index. The same number, presented in a fund's quote sheet, an exchange's contract specification, or a research note, is the Rillor Compute Index. Nothing about the value changes between the endpoint and the term sheet. Only the vantage point changes, and with it the correct word.

That is the whole mapping. The endpoint is the oracle; the named reference price it carries is the Index. A licensee designing a cash-settled product builds against the feed, references the Index by name in its own contract specification, and runs its settlement on its own venue. Rillor supplies the measurement and the brand; the venue supplies everything downstream of the fix.

Guidance for partners

The rule is short. Say Index in market-facing copy. Say oracle when you describe the integration and the derivation machinery.

When you are writing for funds, exchanges, or any reader thinking about the price as a reference, use the Rillor Compute Index. That is the proper noun, the licensed benchmark, the thing with governance and a methodology behind it. When you are writing for engineers wiring a feed into a contract, or explaining how the number is produced, use the oracle. That is the engine and the endpoint.

Concretely. A B300 system such as RIL-GX-B300-2T (built on real hardware Rillor sources from the qualified OEM ecosystem, HGX B200 class platforms from vendors such as Supermicro and Gigabyte, the kind of rack that descends from NVIDIA's GB200 NVL72, 72 Blackwell GPUs and 36 Grace CPUs in one liquid-cooled rack with 130 TB/s of NVLink and 13.4 TB of HBM3e) has exactly one daily forward reference value. To a fund hedging it, that value is the Index. To a venue settling a perp against it, that same value arrives through the oracle feed. Even a previous-generation instrument like RIL-H100-2T, pricing systems built around the NVIDIA H100 SXM with its 80 GB of HBM3 and 3.35 TB/s of memory bandwidth, follows the identical pattern: one Index value, derived by the oracle, published at 17:00 UTC.

Get the two words right and your stakeholders will too. The distinction is not pedantry; it tracks a real boundary between the branded product you license and the engine you integrate against. If you are evaluating Rillor as a venue, a fund, or a research consumer, the for-markets page is where the Index and its licensing terms live.