Open Grant Proposal: Verifiable compute-reuse over content-addressed data

# Open Grant Proposal: Verifiable compute-reuse over content-addressed data

**Project Name:** Verifiable compute-reuse over content-addressed data

**Proposal Category:** `Research & protocols`

**Individual or Entity Name:** The UOR Foundation (U.S. 501(c)(3) non-profit organization)

**Proposer:** humuhumu33

**Project Repo(s):** https://github.com/UOR-Foundation (core repos: uor-addr, prism, UOR-Framework; project work will live in a new public repo under this organization)

**(Optional) Filecoin ecosystem affiliations:** None. We have no existing work relationships with Protocol Labs, Filecoin Foundation, FFDW, or other Protocol Labs Network organizations.

**(Optional) Technical Sponsor:** None yet; we would welcome an introduction.

**Do you agree to open source all work you do on behalf of this RFP under the MIT/Apache-2 dual-license?:** Yes.

# Project Summary

IPFS and Filecoin solved content-addressed storage: data has a permanent name, and anyone can verify they received the right bytes. Computation over that data has no equivalent. When two parties run the same analysis over the same CID-addressed dataset, each pays full compute cost, and neither can verify the other's result without re-running it. Derived results (indexes, analyses, model outputs) circulate with no verifiable link to the data and transformation that produced them.

This project builds verifiable compute-reuse for the content-addressed web. Using the open UOR standard, a computation over CID-addressed inputs receives its own content-derived identifier that binds the inputs, the transformation, and the output. Results become retrievable, verifiable objects: work computed once against a dataset serves everyone, and any result can be checked from its identifier without re-execution.

The mechanics are demonstrated today in our public repositories: repeated computations resolve in roughly 150 nanoseconds versus seconds to re-execute, with bit-identical verification, measured with a standard open benchmarking methodology and reproducible by anyone.

## Impact

Storage becomes more valuable when computation over it compounds. This project gives CID-addressed datasets a growing halo of verified, reusable derived results, all themselves content-addressed and retrievable. The pain points addressed: wasted compute on repeated analyses of public datasets, and the absence of verifiable provenance linking derived results to source data.

Getting this right strengthens the case for storing scientific and public data in the Filecoin ecosystem: a dataset whose analyses are verifiable and reusable is more useful than raw bytes. Our pilot context is high-rate experimental data at SLAC National Accelerator Laboratory, where reproducibility pressure is extreme; success there is a template for scientific data on Filecoin generally. The risk of not getting this right is that compute-over-data ecosystems develop around centralized, unverifiable services, weakening the decentralized web's position exactly where it should be strongest.

Success looks like: a published specification with test vectors, a working open-source implementation, and an end-to-end public demonstration on a real scientific dataset, with reuse and verification measurements published.

## Outcomes

Final deliverables:

1. A computation-identity specification for CID-addressed inputs and outputs: how a transformation over IPFS-retrieved data is canonically identified, how the result object binds inputs, transformation, and output, and how verification proceeds from the identifier alone. Published with test vectors.
2. An open-source reference implementation (Rust): compute a transformation over data retrieved by CID, publish the verified result as retrievable content, and verify any result from its identifier without re-execution.
3. An end-to-end demonstration on a public scientific dataset, with documentation, a tutorial, and a public report including measured reuse and verification performance.

Success metrics: specification published with passing test vectors; implementation passing conformance tests on x86, ARM, and WebAssembly; demonstration reproducible by third parties from documentation alone; verification overhead measured and reported.

## Data Onboarding

Not applicable: this project builds compute-reuse tooling over existing content-addressed data rather than onboarding storage.

## Adoption, Reach, and Growth Strategies

Target audiences: (1) scientific computing groups using content-addressed data, reached through our existing SLAC National Accelerator Laboratory collaboration and the research software engineering community; (2) IPFS/Filecoin developers building compute-over-data tooling, reached through the devgrants ecosystem, our documentation, and direct engagement with related projects; (3) our own open-source community of more than 150 contributors.

First 10 users: the SLAC pilot team plus core UOR contributors exercising the reference implementation. First 100: scientific users via the published tutorial and demonstration dataset, plus IPFS developers via ecosystem channels. All tooling is free and open source, so adoption friction is a download.

## Development Roadmap

**Milestone 1: Computation-identity specification (October 1 to November 30, 2026).** Specify canonical identity for computations over CID-addressed inputs: input binding, transformation encoding, result-object format, and verification rules. Publish with test vectors. Staff: 1 senior engineer plus specification review by a second maintainer. Funding: $15,000.

**Milestone 2: Reference implementation (December 1, 2026 to January 31, 2027).** Rust implementation: retrieve inputs by CID, execute a transformation, mint the verified result object, publish it as retrievable content, and verify any result from its identifier. Conformance tests across x86, ARM, and WebAssembly. Staff: 1 senior engineer plus review. Funding: $20,000.

**Milestone 3: Demonstration, documentation, and report (February 1 to March 31, 2027).** End-to-end demonstration on a public scientific dataset; tutorial and integration documentation; public report with reuse and verification measurements. Staff: 1 senior engineer plus documentation support. Funding: $15,000.

## Total Budget Requested

| Milestone # | Description | Deliverables | Completion Date | Funding |
| --- | --- | --- | --- | --- |
| 1 | Computation-identity specification | Published spec + test vectors | November 30, 2026 | $15,000 |
| 2 | Reference implementation | Rust implementation + cross-architecture conformance tests | January 31, 2027 | $20,000 |
| 3 | Demonstration and report | Public demo on scientific dataset + tutorial + measurements report | March 31, 2027 | $15,000 |

**Total: $50,000**

## Maintenance and Upgrade Plans

The UOR Foundation stewards the standard and its implementations long-term: published conformance suites gate releases, and maintenance and security work on the core libraries is the subject of a separate pending application to the Sovereign Tech Fund (scoped to encoder hardening and security response, distinct from this project). Deliverables from this grant become part of the Foundation's maintained repository set under the same review and conformance discipline.

# Team

## Team Members

- Alex Flom, Chief Scientist, The UOR Foundation (technical lead)
- Core UOR maintainer team (implementation and review)
- Ilya Paveliev, The UOR Foundation (coordination and reporting)

## Team Member LinkedIn Profiles

Available on request (contact below).

## Team Website

https://uor.foundation

## Relevant Experience

The team authored the UOR standard and its reference implementations: canonical content addressing conforming to RFC 8785 (JSON Canonicalization), W3C XML Canonicalization, X.690 DER, and in-toto attestation formats, with 361 passing tests, 19,074 Unicode test vectors, and conformance certification across x86, ARM, and WebAssembly. Members previously built Red Hat's air-gapped OpenShift distribution (OC Mirror), served in U.S. Navy secure communications, and co-created compliance automation with IBM Research and NIST. The affiliated open-source Hologram runtime demonstrates verified compute-reuse end to end, with independently reproducible benchmarks.

## Team code repositories

- https://github.com/UOR-Foundation (uor-addr, prism, UOR-Framework, conformance suites)
- https://github.com/Hologram-Technologies (runtime demonstrating verified compute-reuse)

# Additional Information

We learned about the Open Grants Program through the Filecoin Foundation website while researching funders whose mission aligns with content-addressed computing.

Contact for grant agreement and next steps: Ilya Paveliev, ilya@uor.foundation.

Funding disclosure: a pending NLnet application covers identifier-level interoperability bridges (IPFS CIDs and W3C linked data), and a pending Sovereign Tech Fund application covers maintenance and security of the core encoders. This proposal's scope, verifiable computation over CID-addressed data, overlaps with neither; we will disclose any awards across all three programs.

Start by reviewing the uor-addr, prism, UOR-Framework, and Hologram repositories, especially their existing compute-reuse demonstration and conformance suites. Done means delivering the published computation-identity specification with test vectors, a Rust reference implementation with cross-architecture conformance tests, and the documented scientific-dataset demonstration with measurements.

Written by the indexing model from the issue text.