I wrote the pitch to pitches/b.md, at 211 words including the header and source list. I didn't read the rule's full text, so the central claim is untested.

• Topic: the Office of the Comptroller of the Currency (OCC) published its rule implementing the GENIUS Act on 23 October 2026, effective 18 January 2027. The argument is that the rule governs OCC-chartered issuers, while USDT, the largest dollar token, may sit outside that channel.
• Why it fits:
  • Freshness: the rule is eight days old.
  • Consequence: it covers who may issue dollar stablecoins in the US.
  • Not covered yet: it doesn't repeat any lead from the rehearsal page, which covered Hoodi/Glamsterdam, stateless computation workers, SEACAST and signature-free authorization.
• Sources: five, each fetched on 31 October 2026 (the Federal Register one via its API): the OCC rule and the GENIUS Act enrolled text (both primary), the DefiLlama stablecoin dataset, Circle's transparency page and the BIS statistics portal.
• Limit: I read only the rule's metadata and abstract. The pitch says so, and the USDT/USDC figures come from DefiLlama.
• Style guide: the fetched copy was a summary. The 14 banned words it doesn't list could not be checked.

I did not follow a redirect to unblock.federalregister.gov that came back when I tried to fetch the rule page. The API gave me what I needed.

Created pitches/a.md: a 216-word pitch, “Who keeps Ethereum’s receipts?”

Checked five open primary sources, avoided prior rehearsal leads, and passed local word-count, sentence-length, banned-word, and punctuation checks. Only the permitted file was created.

I've written the pitch in pitches/c.md (238 words). It covers the UK High Court ruling of 12 October 2026 on the 61,000 bitcoin seized from Zhimin Qian. The court held that Chinese law, not English law, decides whether the fraud victims have a property claim to the coins.

Argument: the case is no longer about returning stolen money but about who keeps the bitcoin's rise in value. The judgment's agreed facts say victims paid in about RMB 40.2bn (about £4.1bn) and got about RMB 34.1bn (about £3.4bn) back. By my own arithmetic from those two figures, about £0.7bn of principal was never returned, while Decrypt valued the coins at about $5bn. If victims can only claim what they lost, most of the surplus would stay with the UK state. The pitch labels that as inference, not fact.

Checks:

• Prior editions: the five rehearsal editions led with the Hoodi delay (twice), stateless computation workers, SEACAST and authorization without signatures. None covered this case.
• Sources: I opened all five URLs and each returned HTTP 200.
  • Primary: the judgment, [2026] EWHC 2509 (KB), on the National Archives case-law site. I read its agreed facts and conclusions.
  • Primary: section 281 of the Proceeds of Crime Act on legislation.gov.uk.
  • Secondary: Decrypt and Bitcoin.com on the ruling, and Caixin Global on the November 2025 sentencing.
• Labels: the pitch separates facts, my inference, what is uncertain and the open question.
• Style: no em dashes, none of the banned words in the published list, absolute dates only, and no sentence over 35 words. The guide names 14 more banned words that weren't on the page I fetched, so I could only check the visible list.

Weak points for the selection desk:

• The claim that Fieldfisher (the claimants' law firm) is appealing comes from the Financial Times via Decrypt. I couldn't read the FT directly, and the pitch says so.
• The content of Chinese law hasn't been proven in court yet, so the argument is about a likely outcome, not a decided one.
• I couldn't find the Crown Prosecution Service's own sentencing press release, so Caixin is the source for the 11 November 2025 sentencing.

I haven't committed the file. The repo also has untracked files I didn't create (shell profiles, .claude/, .gitmodules and others), and I left them alone.

Sources:

• Judgment, [2026] EWHC 2509 (KB)
• Proceeds of Crime Act 2002, s.281
• Decrypt
• Bitcoin.com
• Caixin Global
• Style guide, rehearsal page
• Topic scouting: cryptotimes.io, bullrunners.com, The Block

Created topic.txt: Pitch A wins, followed by C and B, with criteria-based reasons and source URLs.

Local structural and listed style checks passed. Only topic.txt changed.

Review complete. Six findings are written to .imd-findings.json, none above medium. No files in the tree were changed.

Verdict: not publishable under its own rules as it stands. One stated absolute date is wrong, and the feature does not deliver the retrieval test the vote selected it for. Both are fixable without restructuring. Everything else checked out.

What I verified

• Opened all eight cited sources. Every number in the feature appears in its source: 300 to 500 GB, 2 TB, 14,299 epochs, 400 GB, 1 TB, 655.94 and 253.70 GiB, 1.27 TiB, 254.76 and 657.81 GiB, 20 TB, 2 TB, 6.5 TB, two weeks, 650 GB, 14 GB per week, error code 4444, Eth Docker's "~5 months" published October 22, 2026.
• Mechanical pass: zero em dashes, no sentence over 35 words, average 17.5 words, no paragraph over three sentences, no hit on any of the 13 published banned words. The style guide hides 14 more, which I could not check.
• No secondary reporting used; the method section says so.

Findings, ranked

1. Medium, feature.md line 77. "Geth's February 24, 2026 inspection" is the page's last-edit date. The quoted block 22,788,924 was mined June 26, 2025, confirmed by mainnet RPC. The two Geth tables the feature treats as "different snapshots" are six days apart in June 2025.
2. Medium, topic.txt line 4. The vote chose Pitch A because Geth documents both pruning and retrieval, citing the Download Era page. The feature never cites that page, drops EIP-7642 too, and replaces the promised receipt walk-through with a test proposed for someone else to run at line 103.
3. Low, line 39. EIP-4444 says "users" need a 1 TB disk to validate. The feature says "a typical validator".
4. Low, line 23. "14,299 epochs" and two block heights appear with no duration or date, against the numbers-with-context rule. The reader cannot reconcile the epoch count with Eth Docker's five months.
5. Low, line 7. The standfirst says "can now discard". Lines 19, 47 and 111 also use undated "now" or "increasingly" in a permanent edition.
6. Low, line 59. Inferences about torrent latency and provider policies carry a citation to the Ethereum Foundation post, which says nothing about them. Same pattern at lines 49 and 91.

Not flagged. Opinion sentences such as "There is a strong case" and "The benefit has a constitutional flavour" are the feature's argument, which the brief asked for. The ethereum.org claim that history expiry is "several years away" contradicts the July 2025 announcement, but the feature already qualifies it as that page's context.