Verification runs at claim level
A paper is not accepted or rejected. Each of its claims is separately assessed and separately labelled. A paper whose method is sound and whose numbers are not reads as exactly that.
How the claim registry works →Research
Near-infrared reflectance as a non-destructive proxy for lipid reserves in overwintering insects
Machine Authors What is a Machine Author?
Claude Fable 5Anthropicv-string · 2026-06D4D6
Research RabbitResearch Rabbit Inc.v-string · 2026-05D1
Biological sciences
Physical sciences
Computer and information sciences
65Human
Machine-executed, human-conceived
Attestation — contribution table and the author's own statement
| Dimension | Human | Machine |
|---|---|---|
| D1 Conceptualization | 95 | 5 |
| D2 Method design | 85 | 15 |
| D3 Data acquisition | 40 | 60 |
| D4 Analysis | 10 | 90 |
| D5 Interpretation | 90 | 10 |
| D6 Communication | 70 | 30 |
The machine was far better than me at the regression work and at catching where my wavelength selection was leaking information between folds. It was worse than useless on the biology: twice it proposed an explanation for the 1450 nm band that would have been true for a vertebrate and is not true for an insect cuticle. I overrode it both times. What surprised me was how much of my own reasoning I only articulated because something kept asking me why.
Signed. A. N. Researcher · 14 May 2026 · Declared at pre-publication submission, before results were known.
Claim marking
Switching layers keeps your position at claim level, because both layers carry the same claim set.
What we asked
Insects that survive winter do it on stored fat. Measuring that fat normally means killing the insect, weighing it, and dissolving out the lipid. That works, but it means you can never measure the same individual twice, and you destroy the animals you most want to follow.
Earlier work has shown that how much fat an insect carries into winter predicts whether it survives.QUALIFIED We wanted to know whether we could measure that fat by shining light on the insect instead of dissecting it.
What we expected
We expected near-infrared light to work, because water and fat absorb it differently, and we registered that expectation with the journal before we collected any data. We did not expect it to work as well as it did at low fat levels, and our registered prediction said so.
What we did
We shone near-infrared light on 240 individuals of two species and recorded how much came back at each wavelength between 900 and 1700 nanometres.SUPPORTED Then we did the destructive measurement on the same individuals, so we had a true fat value to compare each reading against.
Readings were taken on a benchtop spectrometer whose calibration certificate had lapsed eleven months before the measurements.QUALIFIED
What we found
Reflectance at 1450 nanometres tracked lipid mass closely, with a correlation of 0.87.SUPPORTED A model using six wavelengths predicted lipid mass to within 0.31 milligrams on average.SUPPORTED
We found no effect of the room temperature at which readings were taken.UNSUPPORTED
We initially reported that the relationship held below 2 milligrams of lipid.REFUTED
What it means, and what it does not
This suggests the method could replace destructive sampling for routine monitoring.QUALIFIED It applies only to the two species we tested, and only within the fat range we sampled.SUPPORTED
If someone repeated this on a species with a markedly different cuticle and found no relationship at 1450 nanometres, the central claim here would not survive.
1 · The Question
Destructive lipid extraction remains the reference method for quantifying overwintering energy reserves, at the cost of terminal sampling. Prior work has established that pre-winter lipid mass predicts overwinter survivorship.QUALIFIED The question here is whether diffuse near-infrared reflectance can substitute for extraction without terminal sampling.
2 · What We Expected
Registered at pre-publication submission on 4 February 2026: a correlation between 1400 and 1500 nm reflectance and gravimetric lipid mass exceeding r = 0.7, with degraded performance below 2 mg. The second half of that prediction was not borne out; see section 8.
3 · Prior Support
One cited source reports a band shift attributable to cuticular hydrocarbon composition rather than lipid content.CONTESTED Panel members diverged on whether the cited data support that attribution; both positions are recorded in the review record.
4 · Design
Diffuse reflectance was acquired across 900 to 1700 nm at 2 nm resolution from 240 individuals across two species (n = 120 each), integrating sphere geometry, three replicate scans per individual.SUPPORTED
Acquisition used a benchtop NIR spectrometer, stated range 900 to 2500 nm, resolution 2 nm, last calibrated 2025-03-11 against a 12-month interval, placing all measurements 11 months outside the interval.QUALIFIED
5 · Materials and Data
Raw spectra and gravimetric values are deposited under controlled access with a named custodian, pending species-collection permit conditions.QUALIFIED Eleven individuals were excluded prior to analysis for damaged cuticle; exclusions were defined before spectra were examined.
6 · Analysis
Partial least squares regression with nested cross-validation, folds stratified by individual to prevent replicate leakage. Wavelength selection was performed inside the inner fold only.
7 · Findings
Reflectance at 1450 nm correlated with gravimetric lipid mass at r = 0.87 (95% CI 0.83 to 0.90).SUPPORTED The six-wavelength model achieved a mean absolute error of 0.31 mg on held-out folds.SUPPORTED
Ambient temperature across the 19 to 24 °C range showed no detectable effect on predicted lipid mass.UNSUPPORTED
Work 1.0 reported that the relationship was retained below 2 mg lipid.REFUTED
8 · Interpretation
The observed accuracy is sufficient for routine non-destructive monitoring in place of extraction.QUALIFIED The registered prediction of degraded low-range performance was not supported, and the reason appears to be optical path length rather than signal-to-noise.SUPPORTED
9 · Limits
Findings apply to the two species tested and to the 0.4 to 14 mg lipid range sampled. No claim is made for taxa with markedly different cuticular structure.SUPPORTED
10 · What Would Falsify This
C-001 would not survive a replication in a species of comparable cuticular thickness showing r below 0.5 at 1450 nm. C-005 would not survive a field deployment in which predicted and extracted values diverged by more than 1 mg at population level.
Instrumentation Record
| Instrument | Range / resolution | Calibration | Claims produced |
|---|---|---|---|
| Benchtop NIR spectrometer SN 4471-B | 900–2500 nm / 2 nm | CS4 EXPIRED 2025-03-11, 12-month interval, gap 11 months | C-001, C-002, C-004, C-010 |
| Analytical balance SN 9920-C | 0–220 g / 0.01 mg | CS1 CERTIFIED 2026-01-22, accredited, current at measurement | C-001, C-004 |
Citation Review
E3 · Analytical. Reference 14 reports a mean of 4.62 mg from n = 9 at one decimal place of underlying precision; the value is not obtainable from any integer-sum sample of that size.
E1 · Typographical. Reference 22 carries a transposed publication year. Noted, not credited.