Methodology

---

What this site is

A record of what fusion and advanced-nuclear devices have actually achieved, with the source for every number and an explicit statement of what each number means.

That second part is the reason the site exists. The fusion field does not lack for numbers. It lacks numbers you can compare.

---

The gain problem

In December 2022 the National Ignition Facility reported a fusion gain of 1.5. The figure was accurate. It was also, in almost every account of it, ambiguous in a way that mattered by two orders of magnitude.

Gain is a ratio: energy out over energy in. Everything depends on what you count as *in*.

None of these is dishonest. They answer different questions — *did the physics work?* versus *did the plant produce net energy?* — and only the last one is the question a power plant has to answer. Reporting slides between them constantly, usually without anyone intending to mislead.

So we never publish a bare gain figure. Every gain on this site names its denominator:

• Target gain — fusion energy ÷ laser energy delivered to the target *(inertial confinement)*
• Capsule gain — fusion energy ÷ energy absorbed by the capsule *(always flatters)*
• Plasma gain — fusion power ÷ heating power coupled into the plasma *(magnetic confinement)*
• Equivalent gain — what a deuterium-only shot *would* have reached on D-T. A calculation,

never a measurement, and always labelled as such

• Engineering gain — net electrical output ÷ total electrical input, recirculating power

included. No device has exceeded zero

• Wall-plug gain — fusion energy ÷ total facility grid draw

This is enforced in the data model rather than by editorial vigilance. There is no field in which an undenominated gain figure could be stored. It is not that we decline to publish one — we cannot represent one.

---

Achieved, planned, claimed

Three words that get used interchangeably and shouldn't.

Achieved. Measured on a shot, with a date. Eligible for every leaderboard.

Planned. A design target, roadmap milestone or stated goal. Shown on the relevant page, never ranked against achieved results. A company's 2030 target and another's 2024 measurement are not comparable quantities, and putting them in one column implies they are.

Claimed. Asserted by the operator and not independently confirmed. Published, labelled, and not treated as established.

We also keep disputed (a credible challenge exists), retracted and superseded. A later shot that underperforms an earlier one is normal physics; we show both rather than overwriting the record, because hiding regression is a form of falsification.

---

Sources

Every claim points to a source. We fetch each source once, store the bytes, and record a cryptographic hash of them. Verification happens against that stored copy, not against a live URL — so a page that changes or disappears later cannot silently alter what we reported.

We classify each source by its independence:

• Independent — could not have come from us or from the subject: a peer-reviewed paper, a

regulatory docket, a company registry, a patent, a named journalist's original reporting

• Operator — the subject's own publication. Legitimate, and not confirmation
• Aggregator — a directory that may be recycling someone else's copy
• Circular — provably reproduces our own material. Never cited as evidence

Why repetition is not corroboration

A single press release propagates to a dozen trade outlets within a day, each rewording the others. Directories then republish those, and a figure that traces to one company statement acquires the appearance of broad confirmation.

Worse, aggregators republish databases like this one. Searching for a company can return our own summary quoted back at us, which looks exactly like independent verification and is worth nothing.

So we trace claims to their origin rather than counting how many places carry them, and for non-English organizations we search the native-language name — frequently the only route to anything genuinely independent.

---

What we reject, and why we publish the rejections

Some claims do not meet the bar. We publish those with our reasoning, at /benchmark/rejected/.

A published rejection is as informative as a leaderboard entry. It is also the clearest evidence that a standard is being applied rather than announcements being transcribed. Each entry states the claim, who made it, the specific reason it does not qualify, and what evidence would change the verdict — written so the claimant would recognise it as a fair statement of their position.

Recurring reasons a figure is rejected or downgraded:

• Gain quoted with no stated denominator
• "Net energy" that means net-to-plasma
• Triple product extrapolated from simulation rather than measured
• Peak ion temperature presented as sustained
• Plasma duration reported as energy confinement time — they differ by orders of magnitude
• A deuterium result reported as though it were deuterium–tritium
• Records with no shot number or date
• Roadmap dates presented as achievements

---

Nuclear fission is tracked differently

Gain is meaningless for fission, so the comparable question is regulatory position — the thing that actually gates deployment.

We track each developer's stage against its regulator's own records, anchored to a docket or application number a reader can look up. Stage inflation is the dominant error in this sector: "NRC approved" frequently describes a pre-application meeting or acceptance for docketing, and design certification is not permission to build anything anywhere. Where the evidence is ambiguous we assign the lower stage and say so.

Denials and withdrawals are recorded as milestones. They are routinely absent from company timelines, which is precisely why tracking them is worth doing.

We also keep every previously stated target date. A first-criticality target that has moved four times in six years tells you something the current target conceals on its own.

---

Confidence tiers

Not every fact here carries the same weight, and pretending otherwise would be the dishonest choice. Every fact is labelled:

A tier C entry is a pointer, not an assertion. Treating it as one would be our error, not yours — so we say which is which on the page.

---

We will be wrong sometimes

Every change to this dataset is recorded in an append-only log: when a claim entered, when it was promoted, downgraded, superseded or corrected, and why. That log is published at /changelog/.

This is deliberate. A site that never appears to correct itself is not more accurate than one that does — it is less observable. If you find something wrong here, the correction and its reasoning will be visible in the record, not quietly edited into place.

---

What we don't do

• We do not rank targets against results.
• We do not sum grants, milestone awards, in-kind contributions and equity into one "total

raised" figure. They mean different things; we segment them.

• We do not treat an announced order, MOU or letter of intent as a binding commitment.
• We do not report a modelled result as a measurement.
• We do not cite a source we have not opened.