Falsification Criteria for Quantum‑Advantage Claims: An Effective‑Criterion Audit Framework

In the paper's words

Quantum‑advantage claims are increasingly used by hardware developers, funding agencies, and journal editors to justify large expenditures and publication decisions. Existing audits are ad‑hoc, allowing authors to revise task definitions, noise models, or baseline algorithms after a classical rebuttal appears, which renders many claims effectively unfalsifiable. This paper fills the gap by proposing a five‑criterion falsification framework that is effective in the algebraic sense: each criterion is designed to hold for a dense, open subset of well‑formed claims, and degenerate cases are explicitly identified.

The abstract, verbatim; every line below is quoted from the text.

What it claims

“A semigroup mass‑ratio claim scores F=0.70 and was refuted by a statistical audit.”

Numbers it reports

1. Projected storage lies between $30$ TB (1 byte/token) and $150$ TB (4 bytes/token plus $25\%$ overhead), central estimate $120$ TB.

   from 5. Results

2. The HPLT 3.0 collection contains $T_{\text{tot}} = 30$ trillion tokens, i.e. $T_{\text{tot}} = 30\times10^{12}=3.0\times10^{13}$ tokens, covering “almost 200 languages.” Using $L=200$ as an upper‑bound language count, $$ \bar{t} = \frac{T_{\text{tot}}}{L} = \frac{3.0\times10^{13}}{200} = 1.5\times10^{11} \text{ tokens per language.} $$ Input 5 (storage projection assumptions, this paper).

   from 4. Analysis

3. Assuming $b$ bytes per token: Lower bound $b=1$: $3.0\times10^{13}\times1 = 3.0\times10^{13}$ bytes $=30$ TB.

   from 4. Analysis

4. Upper bound with $25\%$ annotation overhead: $1.2\times10^{14}\times1.25 = 1.5\times10^{14}$ bytes $=150$ TB.

   from 4. Analysis

5. For $n=20$ audit checks, the unpruned pair count is $$ P_{\text{unpruned}} = \frac{n(n-1)}{2} = \frac{20\times19}{2} = 190. $$ Assuming a pruning fraction $p=0.9$, the pruned cost is $$ P_{\text{pruned}} = 190\times(1-p) = 190\times0.1 = 19. $$ Derivation 7 (worked example of the index).

   from 4. Analysis

6. Gap density: $42$ gaps across $5$ categories [10] yield $\bar{g}=8.4$ gaps per category; at least one category deviates from the mean by $\ge0.4$ gaps.

   from 5. Results

How it is done

1. Using inputs from the QNFO gap synthesis (42 gaps, 5 categories) and the HPLT 3.0 corpus (30 trillion tokens), we derive quantitative anchors: $8.4$ gaps per category, $2.0$ weeks per roadmap transition, $1.5\times10^{11}$ tokens per language, and worked index values $F=0.675$ and $F=0.10$.

   from Abstract

2. Mirroring the finite‑step decision procedure of [7], each $s_i$ is decided by inspecting a finite document set (the claim, its supplement, and the baseline code), so the total procedure is a finite composition of finite checks.

   from 3.4 Termination

3. Additionally, following the Gebauer‑Moeller strategy of eliminating unnecessary critical pairs before they arise [2], the protocol must specify which checks are not required, so that audit cost remains finite.

   from 3.5 Genericity caveat and pruning requirement

4. Input 1 (weights, Section 3.3). $w_1 = 0.30$, $w_2 = 0.25$, $w_3 = 0.20$, $w_4 = 0.15$, $w_5 = 0.10$.

   from 4. Analysis

What would falsify it

1. The falsification index is $$ F(\mathcal{K}) = \sum_{i=1}^{5} w_i\,s_i. $$ Decision rule: $F\ge0.8$ → falsifiable‑ready ; $0.5\le F\lt 0.8$ → conditionally falsifiable ; $F\lt 0.5$ → structurally unfalsifiable .

   from 3.3 The falsification index

2. This paper fills the gap by proposing a five‑criterion falsification framework that is effective in the algebraic sense: each criterion is designed to hold for a dense, open subset of well‑formed claims, and degenerate cases are explicitly identified.

   from Abstract

3. The framework assigns fixed weights $w_1=0.30$, $w_2=0.25$, $w_3=0.20$, $w_4=0.15$, $w_5=0.10$ to criteria $C_1$–$C_5$, computes a composite index $F=\sum w_i s_i$ with scores $s_i\in\{0,0.5,1\}$, and defines decision thresholds $F\ge0.8$ (falsifiable‑ready), $0.5\le F\lt 0.8$ (conditionally falsifiable), $F\lt 0.5$ (structurally unfalsifiable).

   from Abstract

Where the claim stops

1. However, “dense open’’ is a topological notion with no canonical measure on claim space; we therefore use it only heuristically and require the criteria to hold for all well‑formed claims, not merely generically.

   from 3.5 Genericity caveat and pruning requirement

2. (v) Does the Monte‑Carlo indicator structure of unravelings [4] introduce irreducible statistical uncertainty into $C_1$’s noise‑model specification?

   from 6. Discussion

3. The claim is specification‑complete if each component is fixed in advance and cannot be revised after classical rebuttals appear.

   from 3.1 Claim formalization

Where to read

1. Abstract196 words · 1 min
2. 1. Introduction183 words · 1 min
3. 2. Background and Related Work188 words · 1 min
4. 3. Methods473 words · 2 min
5. 4. Analysis490 words · 2 min
6. 5. Results217 words · 1 min
7. 6. Discussion433 words · 2 min
8. 7. Practical Implications and Engineering Pathways389 words · 2 min
9. 8. Conclusion160 words · 1 min
10. Correction Notice304 words · 1 min
11. Provenance and AI Involvement231 words · 1 min
12. Appendix A. Divergence report117 words · 1 min