Open Grant Proposal: 402 Index on Filecoin — verifiable snapshots of the machine-payable web

# Open Grant Proposal: `402 Index on Filecoin — verifiable snapshots of the machine-payable web`

**Project Name:** 402 Index on Filecoin

**Proposal Category:** `Developer and data tooling` (with a `Storage` component)

**Individual or Entity Name:** Individual

**Proposer:** ercmine (organization account: animicaorg)

**Project Repo(s)**
- https://github.com/animicaorg/all — monorepo; the index builder lives in `apps/x402-gateway` (`bin/build-search-index.js`, `src/products/index.js`, `src/products/mesh*.js`)
- https://github.com/animicaorg/animica-core — the node this work will publish receipts from

**(Optional) Filecoin ecosystem affiliations:** None.

**(Optional) Technical Sponsor:** None yet.

**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

AI agents are starting to pay for APIs directly over HTTP using the x402 standard (HTTP 402 + a signed payment). Since November 2025 the protocol has cleared ~165M transactions, and services that accept it now number in the tens of thousands — but there is no durable, verifiable record of *what is payable, by whom, at what price*. Every directory (the Coinbase Bazaar, x402scan, the 402 Index) is a private, mutable database; when one goes away or changes its ranking, the map of the machine-payable web goes with it.

We operate the largest independent index of these services today: **17,987 endpoints**, built by crawling every service's own 402 challenge (the challenge *is* the document — it carries the price, the network, the accepted assets and the parameter schema), refreshed hourly, exposed as a free trial API, a paid x402 API and an MCP tool (`org.animica/animica` on the official MCP registry). It runs on one server and its history is discarded on every refresh.

This proposal makes that index — and the inference receipts our network produces — a public dataset on Filecoin: hourly snapshots of every 402 challenge observed, content-addressed and retrievable by anyone, with a small verifier so a third party can prove that a given service advertised a given price at a given time. The same pipeline stores the PQ-signed inference receipts that our decentralized inference network (phones, browsers and GPUs serving an OpenAI-compatible API) emits per job, giving Filecoin a real, growing, agent-generated data stream rather than a one-off archive.

## Impact

**Pain points.** (1) Agents and their builders cannot audit the market they buy from: prices, schemas and availability of x402 services change silently and no one keeps history. (2) Disputes between agents and services ("you charged X, you advertised Y") have no neutral evidence. (3) Every x402 directory is a single point of failure; ours included.

**Benefits of getting this right.** A content-addressed, append-only history of 402 challenges is the natural "block explorer" for the machine-payable web, and Filecoin is the natural place for it: the data is public, valuable to many parties, produced continuously, and needs long-term retrievability more than low latency. Builders get a dataset (tens of thousands of real API schemas and prices, with time series) that does not exist anywhere today; researchers get the first public record of agent-to-service commerce; services get provable pricing history. The inference-receipt stream adds a second, larger dataset: signed proofs of who computed what for whom, useful for reputation systems and for auditing decentralized inference.

**Risks of not doing it.** The machine-payable web stays unauditable and its history stays inside a few companies' databases. **Success looks like:** hourly snapshots published for 12 months without gaps; at least three independent consumers (a directory, a research group, an agent framework) retrieving from Filecoin rather than from our API; the verifier used in at least one real dispute or reputation product.

## Outcomes

1. **`x402-snapshot` publisher** (Node/TypeScript, in `apps/x402-gateway`): every hour, serialize the observed 402 challenges (currently ~18k records, ~40 MB uncompressed, ~6 MB compressed) as a CAR file with a deterministic layout (one block per origin, a root manifest with the crawl time, counts and per-origin CIDs), upload via Lighthouse/web3.storage-compatible tooling, and record the root CID on-chain in a small FVM contract (`SnapshotRegistry`) so the sequence of snapshots is itself tamper-evident.
2. **`x402-verify`** (Python + TypeScript libraries, CLI): given a service URL and a time, fetch the relevant snapshot from Filecoin, prove inclusion of that service's challenge in the root, and print the advertised price/network/schema at that time. Also diffs two snapshots ("what changed on the machine-payable web this week").
3. **Inference receipts to Filecoin**: our node already emits per-job receipts (job id, model, prompt/answer hashes, worker address, ML-DSA-65 signature). Batch them daily into CAR files and publish the same way; a receipt verifier joins the CLI.
4. **Public dataset page + docs**: a page listing every snapshot (CID, size, deal status, retrieval instructions), a dataset card, and an MCP tool (`filecoin_snapshot`) so agents can fetch historical challenges directly.

**Metrics:** snapshots published on schedule (target ≥ 99% of hours over 6 months); bytes onboarded; number of active storage deals; retrievals of snapshot CIDs from third-party gateways; downloads/uses of `x402-verify`; number of external projects consuming the dataset.

## Data Onboarding

Index snapshots: ~6 MB/hour compressed → ~4.3 GB/month. Inference receipts: currently ~2,000 jobs/day (~1 MB/day), growing with the network.

- Month #1: ~4.5 GB
- Month #3: ~14 GB cumulative
- Month #6: ~28 GB cumulative
- Month #12: ~60 GB cumulative (more if receipt volume grows as planned)

## Adoption, Reach, and Growth Strategies

**Audience.** (a) Builders of agents and agent frameworks who need to discover and trust paid APIs (the MCP server already sees ~2,000 tool calls/day); (b) the x402 directories and monitors that crawl us today (x402scan, 402explorer, the Coinbase Bazaar discovery crawler, several "trust monitors" — all visible in our access logs); (c) researchers studying agent commerce; (d) the x402 Foundation (Linux Foundation, launched July 2026) and its members, for whom a neutral public history is an obvious shared good.

**Onboarding.** First 10 users: the directories and monitors already consuming our live API — we will offer them the snapshot CIDs as a drop-in history feed. First 100: publish the dataset card on Hugging Face and the Filecoin dataset registries, add the `filecoin_snapshot` tool to the MCP server (installed via `pip install animica-mcp`), and write two technical posts (how the snapshots are built and verified; "what changed on the machine-payable web" weekly diffs generated from Filecoin data).

## Development Roadmap

**Milestone 1 — Publisher + registry (weeks 1–4, 2026-09-08 → 2026-10-06).** CAR serialization with a deterministic layout; hourly publisher with retries and deal-status tracking; `SnapshotRegistry` FVM contract (root CID, crawl time, record count) with tests; first 30 days of snapshots live. One engineer (proposer). **$10,000.**

**Milestone 2 — Verifier + diffs (weeks 5–8, 2026-10-07 → 2026-11-03).** `x402-verify` in Python and TypeScript (inclusion proofs against the manifest, signature checks on receipts), CLI, weekly diff generator, MCP tool, documentation. One engineer. **$10,000.**

**Milestone 3 — Inference receipts + dataset launch (weeks 9–12, 2026-11-04 → 2026-12-01).** Daily receipt batches from the node, receipt verifier, public dataset page with deal/retrieval status, dataset card, two technical posts, outreach to the directories. One engineer. **$8,000.**

## Total Budget Requested

| Milestone # | Description | Deliverables | Completion Date | Funding |
|===|===|===|===|===|
| 1 | Snapshot publisher + FVM registry | `x402-snapshot`, `SnapshotRegistry`, 30 days of hourly snapshots on Filecoin | 2026-10-06 | $10,000 |
| 2 | Verifier, diffs, MCP tool | `x402-verify` (py+ts), CLI, weekly diffs, docs | 2026-11-03 | $10,000 |
| 3 | Inference receipts + dataset launch | daily receipt CARs, receipt verifier, dataset page/card, 2 posts | 2026-12-01 | $8,000 |
| | **Total** | | | **$28,000** |

## Maintenance and Upgrade Plans

The publisher runs alongside the index builder we already operate (an hourly systemd timer); keeping it alive costs us nothing beyond deal fees, which we will fund from the index's paid tier. The code lives in the monorepo we ship from daily. Planned upgrades: incremental (delta) snapshots once the base layout is stable, and publishing the receipt stream from every node in the network rather than only ours.

# Team

## Team Members

- ercmine — sole engineer (design, consensus, cryptography, infrastructure, wallets, the x402 products and the index)

## Team Member LinkedIn Profiles

- Not available; see GitHub history at https://github.com/animicaorg

## Team Website

https://animica.dev

## Relevant Experience

Built and operate, alone, a live Layer-1 blockchain with post-quantum (ML-DSA-65) signatures, its wallets (Android, desktop, browser extension), a decentralized inference network served from phones and browsers, an MCP server on the official registry, and 70+ x402 paid APIs settling on Base — including the index this proposal is about, which already crawls and enriches ~18k services hourly and serves them to agents. Comfortable with content addressing and CAR/IPLD tooling from building the node's own snapshot import/export (canonical CBOR, chunked, self-delimiting framing).

## Team code repositories

- https://github.com/animicaorg/all
- https://github.com/animicaorg/animica-core
- https://pypi.org/project/animica/ · https://pypi.org/project/animica-mcp/

# Additional Information

Learned about the program from the devgrants repository while researching where a public, retrievable history of agent-payable services should live. Best contact: ai@3vdc.com. The live index is at https://animica.dev/x402/index (free trial: `?q=…`), and the MCP server is `org.animica/animica`.

Start in apps/x402-gateway, especially bin/build-search-index.js, src/products/index.js, and src/products/mesh*.js, to understand the existing index data and hourly workflow. The proposal is complete when the publisher, SnapshotRegistry, Python and TypeScript verifier, receipt batches, dataset page, and documentation described in the milestones are delivered and the stated snapshot and retrieval goals can be checked.

Written by the indexing model from the issue text.