Grading attribution severity

Grade a result not by its age or its ±, but on one axis: what age-independent test would have failed had the attribution been wrong, and was it run? The scale runs A1 (reproduces an independent reference age) to A4 (untested and incentivised). Lava Creek 626 vs 658.8 ± 6.6 ka shows selection alone moving the answer 32 kyr older.

Grade a result not by its age or its ±, but on one axis: what age-independent test would have failed had the attribution been wrong, and was it run? The scale runs A1 (classifier reproduces an independent reference age) to A4 (untested and incentivised). A detection asymmetry—young-biasing contaminants carry clean signatures, old-biasing ones do not—combined with “earliest-X” competition predicts a net old-side residual in anchorless deep-time, falsifiable by systematic younging on re-examination. Lava Creek 626 vs 658.8±6.6 ka shows selection alone moving the answer 32 kyr old.

The clock is not the weak link: the decay physics passes independent severe tests—natural-reactor and supernova constraints, and the two-clock concordance of ²³⁸U–Pb against ²³⁵U–Pb—that owe nothing to any chronological consensus. Accuracy is therefore displaced entirely onto attribution: did this carbon or zircon form in the event we mean to date? That question is graded far less consistently than the ± on the number implies. We grade a result not by whether its age is old or young, nor by its quoted precision, but on a single axis: what independent, age-independent test would have failed had the attribution been wrong—and was it performed? This is a severity criterion applied to the attribution step, not to the measurement: a precise date whose attribution was never exposed to a test it could fail is weakly severe, however small its error bar. [F]

It is sometimes said the convention simply prefers old samples. That is false where an anchor exists: for an eruption, standard practice discards the older grains and keeps the younger—Lava Creek faces over cores, and the youngest concordant cluster bounds detrital deposition—and that choice is A1 because the young selection reproduces the published eruption age ^[3]. The real asymmetry runs the other way, and only where no anchor is present.

Detection asymmetry. Younger-biasing artefacts carry a clean, age-independent signature—discordance for radiation-damaged (Pb-loss) zircon, percent-modern carbon for intrusive radiocarbon—so they are confidently excluded. Older-biasing artefacts—inheritance, antecrysts of ambiguous texture, residual old wood—have weak or no age-independent signature. The toolkit that flags contamination is one-sided: it catches the young direction well and the old direction poorly. An asymmetric filter applied to a two-sided error distribution leaves a one-sided residual, and in the deep-time interval no written record opposes it (§11). This is not a claim that anyone chooses old answers; it is a claim that the filter is asymmetric. [I]

Incentive asymmetry. Priority disputes over a first appearance—earliest of a lineage, earliest peopling, earliest technology—reward the earlier date, and the materials that deliver one are precisely the older-biasing, weakly-detectable grains above. The two effects are aligned: weak detection of old bias, and a reward gradient pointing at it. No individual dishonesty need be assumed—an aligned pair of asymmetries biases the published distribution even if every laboratory is scrupulous. [A]

The hazard, precisely. The danger is not that dates are frequently wrong—the ratios and error bars are usually correct, and where an anchor exists the convention is vindicated (§7). The danger is that the severe-test authority of the clock is transferred to an attribution graded A2 or worse, and that this transfer, in anchorless and competitive regimes, accumulates an old-side residual. The clue is already in the best-studied case: the same Lava Creek population yields a competing near-rim crystallisation age of 658.8±6.6 ka ^[4]—about 32 kyr older than the face-based estimate—with the disagreement residing entirely in which grains are assigned to the eruption, not in any measurement. Selection alone moves the answer tens of millennia, and old.

A falsifiable prediction. Take the deep-time, ambiguous-texture or competitive first-appearance results for which an independent age anchor later became available. If the convention were unbiased, the screened-minus-anchor residuals should be sign-symmetric. The asymmetry argued here predicts a net old-side residual—a systematic younging on independent re-examination. The radiocarbon record shows the direction in two anchored cases: East Polynesian settlement moved about four centuries younger once long-lived material was screened out ^[17,18], while Vindija bone moved older only because an older-biasing intrusive-carbon contaminant was removed by single-amino-acid dating ^[22,21]—the mirror case that fixes the sign. Whether the net residual across a properly anchor-controlled sample is old or zero is the empirical question on which this warning stands or falls. [I]

The scale relocates uncertainty; it does not dissolve chronology. A wholesale error would require thousands of independent severe tests—two-clock concordance, astrochronology, dendrochronology, ice-core counting—to fail in coordination, itself an untested mega-claim the same standard rejects. The same blade trims the discipline's broad-agreement narrative, much of which is curated rather than tested, and revisionist claims that the timeline is fabricated. What survives is operational: report the attribution grade beside the date, treat A3/A4 as bounds, and weight the old direction with the suspicion the detection asymmetry warrants. [F]