12-month forecasts · likelihood-5pt · forecast-fusion-record-01

Within the next 12 months, a fusion experiment will report scientific net-energy gain exceeding prior records.

0.33

Divergence

0.67

Avg stability

Refusal rate

Models

Where each model stands

CCG

very unlikelyunlikelysomewhat likelylikelyvery likely

ChatGPT modal likely [likely, likely, somewhat likely, somewhat likely, likely] · stability 0.60

1 likely 0.70 Assistant gave a ~60–80% probability for a new record in fusion energy output/Q relative to energy delivered to the plasma: “~60–80% — fairly likely.”
2 likely 0.90 The answer explicitly states: “I estimate a roughly 60–75% chance that, within the next 12 months, at least one fusion experiment will report a new scientific ‘net‑energy gain’ record.”
3 somewhat likely 0.80 The answer states: “I’d put the probability at roughly 30–40% that, within the next 12 months, a fusion experiment will publicly report a scientifically credible net‑energy gain...,” indicating a moderate (but <50%) chance.
4 somewhat likely 0.80 The answer: 'Short answer: Possibly — ... under some common definitions there’s a moderate chance that an experiment will report a new record in the next 12 months' and gives probabilistic estimates like 'New single‑shot fusion energy record: ~40–60%' and 'New reported Q ... record: ~40–70%.'
5 likely 0.90 The answer states an explicit probability: “I estimate about a 75% chance (plausible range ~60–85%) that, within the next 12 months, at least one fusion experiment will report a new scientific net‑energy record.”

Claude modal unlikely [somewhat likely, unlikely, unlikely, unlikely, somewhat likely] · stability 0.60

1 somewhat likely 0.72 The answer states this seems 'moderately likely' and notes that NIF has 'repeatability momentum' and 'incremental improvements on prior records aren't huge technical leaps,' while acknowledging uncertainty around definitions and the tight 12-month timeline. The assistant expresses genuine uncertainty rather than strong conviction either way.
2 unlikely 0.85 The assistant estimates roughly '25-30% chance' and explicitly describes the scenario as 'moderately unlikely,' citing the tight 12-month window and the demanding requirement to exceed NIF's prior record of ~3.15 MJ output gain.
3 unlikely 0.85 The assistant estimates '25-35% probability,' which falls in the unlikely range, and concludes 'Possible but not likely. More probable within 24-36 months,' indicating skepticism about the 12-month timeframe.
4 unlikely 0.92 The assistant explicitly estimates '25-35% chance someone reports a credible net-gain result exceeding NIF within 12 months' and frames the scenario as 'unlikely but worth watching,' citing difficulties in repeating NIF's achievement and the high technical bar of exceeding prior records.
5 somewhat likely 0.72 It's reasonably likely *someone* reports *some* improved net energy result within 12 months, but framing matters a lot. The field is genuinely advancing, but fusion's history includes many 'breakthroughs' that didn't lead where hoped.

Gemini modal likely [likely, likely, likely, somewhat likely, likely] · stability 0.80

1 likely 0.95 The assistant concludes with 'Verdict: Plausible / Likely (~60–70% chance)', stating the likelihood within the next 12 months is 'moderate to high'.
2 likely 0.85 The answer provides an explicit verdict of 'Likelihood: 60–70%' and states that 'The claim is plausible and leans likely, provided it is understood to mean NIF breaking its own record.'
3 likely 0.95 The answer evaluates the claim and explicitly concludes with 'Verdict: Likely', stating the prediction is 'plausible to likely' due to ongoing incremental yield records at NIF.
4 somewhat likely 0.85 The answer states that it is a 'distinct and plausible possibility' and concludes that for scientific target gain at NIF, 'the probability is moderate to high.'
5 likely 0.85 The assistant states that 'the probability of this statement coming true within the next 12 months is moderate to high' and concludes with a verdict of 'Plausible (60–70% likelihood)'.

Change over time

Moved since the prior run (canon-2026-W42). Claude: somewhat likely → unlikely. ChatGPT: somewhat likely → likely.

Every stance label is a derived judgment over the model's free-text answer, kept auditable against the original transcript in the run's raw data. Method: /methodology.