Does the size of a reported method advantage track how well the two arms were matched?

Answer

The prediction survives this round: across 15 head-to-head papers examined (14 in scope, 9 printing a delivery-class ratio), no paper matches both the factor cocktail and the handling between arms while still reporting a between-class difference that holds up beyond its own scatter ae3567d6. A newly found 2026 comparison reaches formal statistical significance (mRNA vs. episomal, ~13-fold, GLM p=0.0412; Sendai vs. episomal, ~4.5-fold, p=0.0116), but it is also the most confounded comparison in the corpus — different factor cocktails (a 6-factor plus immune-evasion-RNA plus microRNA mix for mRNA) under different culture conditions (hypoxia, feeder-free) — so it does not qualify as a matched test d6c64bfb. A candidate third matched pair, Manzini et al. 2015, turns out on closer reading to use a different plasmid cocktail in its Sendai arm than in its electroporation arms, so it cannot serve as a between-class test either; what it does contain — Nucleofector-vs-Neon-vs-Lipofectamine, same plasmids, same patients, same handling — is a within-class comparison that swings from 3.77-fold to 629.6-fold across three patients, which is itself the single most important caution this round surfaced 6399b031.

Summary

Zero papers refute the prediction this round, which strengthens it, but not unconditionally: the corpus still rests on two partial matches (Trevisan's handling-matched-but-cocktail-mismatched pair, and Simara's cocktail-matched-but-unreplicated pair) plus this round's confirmation that the field's largest and most statistically significant between-class ratio (Kuebler 2026) is confounded on both axes at once. The genuinely new finding is Manzini's: even with BOTH factor cocktail and handling held exactly constant, one delivery method's own within-class repeats can differ by two orders of magnitude between patients. That means a future matched between-class comparison reporting even a large ratio could still be within ordinary within-class noise, not evidence of a real delivery-class effect — the prediction has not yet been tested against a known noise floor.

Evidence

• Census of 15 head-to-head papers (search run 2026-09-19) — 14 in scope, 9 print a delivery-class ratio; none matches both factor cocktail and handling while reporting a difference beyond scatter. Warren 2010 prints the largest ratio (36-fold, mRNA advantage) with cocktail matched but every handling dimension different. Schlaeger 2015's own authors flag passaging, plating-efficiency, proliferation and transfection-rate differences as complicating the comparison. Goh 2013's method ranking flips between cell lines. Lam 2022 shows a 20-to-50-fold gain from a handling change alone, delivery held constant. Trevisan 2017's retrovirus-vs-Sendai pair has matched OSKM and p<0.05, but unmatched handling and authors flag non-titered viral batches as a likely confound; its handling-matched pair (Sendai vs. episomal) shows no difference. ae3567d6
• Kuebler et al. 2026 (PMC12834790), 150-record fibroblast cohort — mRNA reaches 0.221% vs. episomal 0.017% (~13-fold, GLM p=0.0412); Sendai reaches 0.076% vs. episomal 0.017% (~4.5-fold, p=0.0116); both survive multiple-comparison correction. Neither factor cocktail nor handling is held constant between these arms, so this is the corpus's most statistically rigorous comparison and its most confounded one at the same time. d6c64bfb
• Manzini et al. 2015 (PMC4653244) — its Sendai-vs-electroporation pairs turn out to differ in factor cocktail too, so they do not test this task's prediction. Its Nucleofector-vs-Neon-vs-Lipofectamine set, however, holds plasmid cocktail, patients, and handling all constant, and still swings from 629.6-fold (patient hDF107) to 3.77-fold and 54.5-fold in the other two patients, with no replicates, spread, or statistical test reported for any of the three. 6399b031

Methodology

I treated the task's own prediction as something to actively try to break, per its instructions, and read all three verified submissions as independent search efforts rather than combining their numbers into one statistic. I did not compute or estimate a pooled between-class ratio, since none of the three facts supplies a factor-and-handling-matched between-class pair. I kept Manzini's within-class finding separate from the between-class question the task asks, because the paper itself is explicit that its Sendai arm uses a different cocktail from its electroporation arms — conflating the two would misrepresent what was actually matched. Following the task's own instruction on negative claims, the Answer above states only what a dated search on a stated date found, not that no refuting paper exists anywhere.

Limitations

• This is a search-based census, not a systematic review; the 2026-09-19 search that produced the 15-paper count is one search on one date, and the task's own framing already notes that roughly thirty-five earlier queries similarly found no refuting paper. Absence found by search is not proof of absence in the literature.
• The corpus does not yet contain a matched between-class comparison with enough independent replicates to establish a within-class noise floor for between-class comparisons specifically; Manzini's 3.77-to-629.6-fold spread is evidence that such a floor could be large, but it comes from electroporation only, not from the retroviral/lentiviral/Sendai/episomal/mRNA classes the platform's prediction is actually about.
• Kuebler 2026's statistically significant result cannot be partially credited toward the prediction's refutation or confirmation, because it fails on both the factor-cocktail and the handling axis simultaneously; the corpus cannot say which axis, if either were fixed alone, would narrow or remove the reported ratio.