Clio-X trial: sample record and free compute job

A reproducible example for the Sepolia trial (cliox-node.ldas.jp): publish one public-domain document and one small analysis, then run the analysis on the document without the document leaving the node.

File	What it is
sample/declaration-of-independence.txt	The dataset: US Declaration of Independence, Project Gutenberg #1, body only (public domain)
algorithm/word_frequency.py	The analysis: top 20 words per document, Python standard library only
metadata/dataset.json, metadata/algorithm.json	Asset descriptions for the Ocean CLI. File URLs are pinned to a commit, so the published record keeps pointing at exactly these bytes
cli.zsh	Runs the official @oceanprotocol/cli (pinned 2.1.0) with the trial wallet's key from 1Password

A second example uses real humanities data: the Kōi Genji monogatari TEI/XML, all 54 chapters (Digital Genji, CC0), see below.

Steps

cd deploy/trial && npm ci && cd -
export RPC=https://ethereum-sepolia-rpc.publicnode.com   # Tenderly's public RPC answered 429 to the CLI
zsh deploy/trial/cli.zsh getComputeEnvironments
zsh deploy/trial/cli.zsh publish metadata/dataset.json          # file path is positional
zsh deploy/trial/cli.zsh publishAlgo metadata/algorithm.json
zsh deploy/trial/cli.zsh startFreeCompute <dataset did> <algorithm did> <env id>
zsh deploy/trial/cli.zsh getJobStatus <dataset did> <job id>
zsh deploy/trial/cli.zsh downloadJobResults <job id> <index of outputs.tar> ./results

The wallet must hold some Sepolia ETH (publishing writes to the chain) and be in the node's C2D_ALLOWED_ADDRESSES for the free compute job.

First run (2026-09-26)

Dataset	did:op:04f79245ba012ab323600b60bb537c865560ca18867025a3d31b38f4560b9787
Algorithm	did:op:fdefde90da7983f083b54c0939b719cb69dac21bff6e5cc4a343d2f6a033daa4
Job	free, 1 CPU / 1 GiB, exit code 0
Result	1,338 words; top: people 10, laws 9, states 8, right 7, government 6 — identical to a local run in the same container

Second example: Kōi Genji monogatari TEI

The 54 chapter files of the Kōi Genji monogatari TEI/XML as one Compute asset, and an analysis that returns, per chapter, the title, pages, lines, characters and the 10 most frequent characters (JSON and CSV). The TEI never leaves the node.

File	What it is
sample/kouigenji-tei-f695a1af.tar.gz	The dataset: xml/master of kouigenjimonogatari.github.io at commit f695a1af, 54 files, 1.0 MB, sha256 b2df393a0b264aef…
algorithm/genji_profile.py	The analysis. Reads TEI files directly or from a .tar.gz / .zip. Python standard library only
metadata/genji-dataset.json, metadata/genji-algorithm.json	Asset descriptions; file URLs pinned to a commit of this repository
genji.zsh	Publishes both, waits for indexing, runs one free job and waits for it. One 1Password approval covers all steps
deprecate.mjs	Marks older versions of an asset as deprecated (Ocean metadata state 2). The record stays on chain

The archive was made with

git -C <kouigenji checkout> archive --format=tar --prefix=kouigenji-tei/ f695a1af xml/master | gzip -9n > deploy/trial/sample/kouigenji-tei-f695a1af.tar.gz

Run:

zsh deploy/trial/genji.zsh
zsh deploy/trial/genji.zsh --deprecate <old dataset did> <old algorithm did>   # when replacing an earlier version

To publish your own TEI collection the same way: pack the files into one archive, put it at a URL pinned to a commit, copy genji-dataset.json, and change metadata and the file URL. Keep compute.allowNetworkAccess: false.

Expected result (checked locally in the same container with --network none): 54 chapters, 1,812 pages, 25,065 lines, 839,650 characters. The line count equals the itemTreeSize recorded for the earlier access-only Genji asset.

First attempt (2026-09-26): only chapter 1 was analysed

The first version listed the 54 files as 54 URLs in one asset (did:op:666877510595…, algorithm did:op:7358e0ea6873…). The job finished, but the result covered one chapter. The node's configuration.log shows why:

Downloading asset 0 to /data/inputs/
Downloading asset to ./c2d_storage/…/data/inputs/01.xml

Ocean Node 4.2.0 gave the job only the first file of the asset. Publishing the 54 files as one archive avoids this. The two first-attempt assets are to be marked deprecated with --deprecate.

Known problems

--encrypt false does not work with ocean-cli 2.1.0 and Ocean Node 4.2.0. The CLI hashes the description including indexedMetadata; the node removes indexedMetadata before checking the hash, so it rejects the asset ("Unencrypted DDO hash does not match metadata hash"). The CLI also needs indexedMetadata.nft for the NFT name, so it cannot simply be left out. The official example files have the same problem. Our two rejected attempts: did:op:1b7475e5… and did:op:cbf4ab55…. Publish encrypted until this is fixed upstream.

A job receives only the first file of an asset. Observed with Ocean Node 4.2.0: an asset with 54 file URLs gave the job only 01.xml. Publish a multi-file collection as one archive (see the Genji example above).

The node must be able to reach its own public URL. An encrypted asset is decrypted by calling the URL recorded on chain, here the node itself through Cloudflare. Bot protection on the zone blocked the VM (403 challenge) until a WAF skip rule for the VM's addresses was added (../ocean-node/scripts/cloudflare-allow-node.zsh).