- Metric
- Share of the tracks delivered to Deezer each day that its detector classifies as fully AI-generated, as disclosed by Deezer. Baseline: about 10% in January 2025, 39% in January 2026, 44% in April 2026, above 50% at the June 2026 peak.
- Threshold
- At least 45% in the first figure Deezer publishes in 2027.
- Deadline
- Deezer’s first such disclosure dated in 2027, no later than 30 June 2027. If Deezer publishes no figure by then, the entry is VOID, not confirmed.
- Source
- Deezer newsroom or an investor communication quoting Deezer’s own detection figures. If both a peak and an average are given, the average decides.
- Refuted if
- Below 40% refutes it. 40 to 45% is MIXED.
- Confidence
- 80%
- Status
- OPEN
ACCOUNTABILITY_INSTRUMENT()
Preregistered predictions.
An opinion asks you to agree. A signal shows you where its author bleeds. This page is where the claims made in my essays become falsifiable commitments: registered before the results arrive, scored in public, with null results published as prominently as confirmations.
The rules
- Every prediction states five things: the metric, the threshold, the deadline, the data source, and the decision criterion. If any of the five is missing, it is an essay, not a prediction.
- Registered before results. Each entry is committed to a public git repository before the measurement window opens; the commit hash is the timestamp.
- Third-party data first. Where possible, the resolving series is published by someone else (official statistics, public price trackers), so the only discretionary act is choosing the series, made once, in advance.
- Resolvable in both directions. Each entry states what outcome refutes it. CONFIRMED, MIXED and WRONG get the same font size.
- Nullius in verba. Anyone can recompute every score from the cited sources.
- Calibrated, not brave. Each entry carries a stated confidence, so the register can be scored as a whole and not only entry by entry. An 80 percent claim is supposed to fail one time in five.
The register
BATCH 01 · REGISTERED 2026-11-09 · COMMIT 612d0dedcb27
- Metric
- Combined global share of white, black, grey and silver in Axalta’s annual Global Automotive Color Popularity Report. Baseline: 81% in the 2025 report (white 29, black 23, grey 22, silver 7).
- Threshold
- At least 78% in the 2027 edition.
- Deadline
- Publication of the 2027 edition; scored within 30 days. If none is published by 31 March 2028, the entry is VOID.
- Source
- Axalta. A paint supplier is an interested party; the series is used because no neutral global series exists. This entry tests whether the convergence is structural in a market with no model in the loop.
- Refuted if
- Below 75% refutes it. 75 to 78% is MIXED.
- Confidence
- 85%
- Status
- OPEN
- Metric
- Whether the name of an AI model, model family or AI product comes into ordinary use as the name of a style, the way Victorian, Bauhaus, Kodachrome and Instagram Face name a look. Three public markers: (a) a general English dictionary (Oxford, Merriam-Webster, Collins, Cambridge, Macquarie or Dictionary.com) adds an entry or sense, or names a word-of-the-year finalist, in which such a name denotes a style or quality of work; (b) English Wikipedia carries a stand-alone article, not a redirect or a section, about a visual, musical or prose style named after such a model or family; (c) a museum or public gallery uses one in the title of an exhibition.
- Threshold
- At least one of the three markers.
- Deadline
- 31 December 2030, checked on that date.
- Source
- The dictionaries’ own sites and word-of-the-year announcements; Wikipedia article history; the institution’s exhibition page. Marketing by the model’s maker and articles about the product itself do not count. At registration I know of no example of any of the three.
- Refuted if
- None of the three exists on the deadline. No MIXED outcome.
- Confidence
- 65%
- Status
- OPEN
STILL TO COME
Entries for the generation-versus-verification cost ratio and the Originality Premium accompany the next measurement posts in The Scarcity Ladder. Per rule 2, nothing is claimed as predicted until it appears here with a verifiable commit hash.