The Empty Unison: Interim Record

Published October 26, 2026, by DeltaSignal (AITrailblazer). This is an interim record. Grading continues, and this page will be updated as further verdicts land.

What this is

Every day, the largest crypto information services publish a short AI-written answer to the same question: why did the price move today? Millions of readers meet that answer before they meet any primary source. The answers sound different from service to service. Underneath, they are built from the same diet of headlines, and they converge on one shared explanation of the day. We call that phenomenon the Empty Unison: many voices, one line, and the line is only as good as the record behind it.

Since October 22, 2026, we have been capturing those answers as they are published and grading each one against the primary record. This page is the public record of that grading so far.

What was measured, and how

The measured object is CoinMarketCap's AI price analysis: the dated "Why is the price up/down today?" entry for each of ten coins (BTC, ETH, SOL, XRP, BNB, DOGE, ADA, LINK, AVAX, HYPE). Entries are captured hourly while they are live, because the entry for a day can be rewritten during that day (see Revisions, below).

Each captured entry is then graded against the primary record only:

• the coin's actual price path (Coinbase),
• published spot ETF flow prints,
• liquidation data,
• and independent reporting of the session, checked on the day of grading.

The summary's own words are never used as evidence. An entry is the thing being measured, never a source for its own grade. Every verdict lists the premises it rests on, with sources and check dates. A premise that would move a verdict harsher than "confirmed" needs two independent sources. Where a claim has several parts, each part is graded separately.

The verdicts:

• Confirmed: the stated primary cause checks out against the primary record.
• Unsupported: the stated primary cause has no support in the primary record as stated: a figure that belongs somewhere else, a number with no print behind it, or a description of the price move ranked as its cause.
• Silent on the real driver: the entry names something real, but leaves out the day's largest driver.
• No cause stated: the entry itself said no clear catalyst was visible. An honest blank, recorded as such, not counted as a failure.
• Mislabeled: the entry named the wrong cause outright. No entry has earned this verdict so far.

The tally so far

Entries graded: 40. These are the dated entries for October 21, 22, 23 and 24, 2026 (ten coins per day), graded between October 22 and October 26, 2026.

• Confirmed: 26 of 40 entries (65%).
• Silent on the real driver: 3 of 40 entries.
• Unsupported: 9 of 40 entries.
• No cause stated: 2 of 40 entries.
• Mislabeled: 0 of 40 entries.

The headline is not that the summaries are usually wrong. It is that they almost never name a wrong cause outright. They fail in two quieter ways, set out below.

Verdicts by day

Entries for October 21, 2026 (graded October 22, 2026)

Day tally, October 21 entries: confirmed 5, silent on the real driver 1, unsupported 2, no cause stated 2.

Entries for October 22, 2026 (graded October 24, 2026)

Day tally, October 22 entries: confirmed 7, silent on the real driver 2, unsupported 1, no cause stated 0.

Entries for October 23, 2026 (graded October 25, 2026)

Day tally, October 23 entries: confirmed 8, silent on the real driver 0, unsupported 2, no cause stated 0.

Entries for October 24, 2026 (graded October 26, 2026)

Day tally, October 24 entries: confirmed 6, silent on the real driver 0, unsupported 4, no cause stated 0.

Revisions: the feed rewrites its own answer

A revision is a change to an entry's stated cause after it was first captured, inside the same dated entry. The answer does not just age; it is rewritten as the price moves, and the cause follows the price.

Revisions caught: 22, between October 22 and October 26, 2026.

Six of them landed on the morning of October 26, 2026, and all six were direction flips: entries that began the morning explaining a fall ended it explaining a rise, with a new cause fitted to the new direction (SOL, DOGE, ADA, LINK, AVAX, HYPE). Those six belong to the October 26 entries, which have not been graded yet; they are counted here as revisions, not verdicts.

The two failure modes

1. Specific numbers with nothing behind them (unsupported). The clearest case is Solana on October 23. The entry said Solana rose because of a $100 million J.P. Morgan real-world asset investment on Solana. The event underneath was real: J.P. Morgan's tokenized funds became available on Solana. But the $100 million figure was the seed capital from those funds' earlier launches on Ethereum, not new money on Solana, and Solana closed the graded day down. A real event, a figure that belongs to a different fact, a move that did not happen: unsupported. The same shape appears in XRP and HYPE on October 24, where headline positioning figures turned out to be market-wide totals or had no dated print behind them, and in BNB and ADA on October 24, where a "technical breakdown" was ranked as the cause of a fall it merely describes.

2. Silence about the day's real driver (silent on the real driver). On October 22, the Bitcoin entry named a liquidation cascade and the XRP entry named caution ahead of the Fed. Neither mentioned the day's two largest drivers: a $635.6 million outflow from spot Bitcoin ETFs, the largest single day since June, and a 2.6% fall in the Nasdaq. Everything each entry said was defensible. What it left out was the story. Three of the 40 graded entries failed this way, all of them in the first two days graded.

A correction of record: our error, not the feed's

The first grade this instrument issued was wrong, and the error was the grader's. On October 22, the Bitcoin entry for October 21 was first graded "mislabeled," partly on the premise that SpaceX is a private company, so its "SpaceX earnings" cause had to be an artifact. That premise was false. SpaceX has been public since its IPO on June 12, 2026 (Nasdaq: SPCX), and it reported third-quarter 2026 earnings on October 20, 2026, with revenue up 161% year over year, exactly the figure the entry cited. The grade was corrected the same morning to "silent on the real driver," on the part of the basis that survived checking: the entry named a real equity catalyst while saying nothing about the day's $541 million of spot Bitcoin ETF inflows. The correction is part of this record, kept on purpose. A grading instrument that cannot show its own corrections has not earned the right to grade anyone else. Since that morning, every world-fact used as a grading premise is checked live on the day it is used, and premises that would harshen a verdict are double-sourced.

The bar set in advance

This log was started under a proof gate written before the first grade was issued. The bar, quoted exactly from the instrument's specification:

"Subscriber-facing divergence flags ship only if logged divergences resolve the claimed way at a rate chance cannot explain."

The readout on Thursday, October 29, 2026 is judged against that bar. This interim record does not prejudge it.

Status

Interim record, published October 26, 2026. Grading continues daily. The entries for October 25 and October 26, 2026 are captured and grade on October 27 and October 28 respectively; the entries for October 27 and October 28 grade on the same 48-hour schedule. The full readout follows on October 29, 2026.

Published by DeltaSignal (AITrailblazer). Grades rest on the primary record named above; the full grading sheets, with premises, sources and falsifiers for every verdict, are kept with this record.