Lilla Gloppsjön pollen scores on contained age intervals¶
The original study scored 23 nominal 500-cal-BP bins using
point ages from Neotoma/EPD site 3234, core 4071, dataset 4200 (DOI
10.21233/vpnw-pv89). This sensitivity keeps a sample only when both source
age bounds fall inside its point-age bin. It reuses the same 69 frozen
predictions, three fixed methods, 16 literal pollen labels and count rule.
Nothing is refitted or assigned to a neighboring bin.
The sample ledger partitions all 86 source samples: 42 of the 84 selected calibrated-age samples are wholly contained, 42 cross a bin boundary, and two retain their original negative-bound age refusal. The 42 contained samples carry 61,647 native pollen NISP, of which 28,310 are on the closed 16-label response. The crossing samples carry 61,737 native and 26,619 closed-axis NISP. The two age-refused samples carry 3,643 native and 709 closed-axis NISP. All three groups sum to the original 127,027 native and 55,638 closed-axis NISP. Counts outside the response remain visible.
The coverage retains all 23 candidate bins. Twenty contain positive closed-axis counts from at least one wholly contained sample and score all three methods; bins 20–22 have no contained sample and are refused. The 69 method/bin rows preserve each attempted status, count denominator, contained score and original full-bin score. On the same 20 bins:
| Fixed method | Contained MAE | Original full-bin MAE | Contained Hellinger | Original full-bin Hellinger |
|---|---|---|---|---|
| Inverse distance squared, frozen primary | 0.019316 | 0.018367 | 0.249632 | 0.242217 |
| Nearest site | 0.020786 | 0.021199 | 0.230690 | 0.229585 |
| Training mean | 0.017497 | 0.016489 | 0.263196 | 0.257241 |
These are paired by bin and method, while the pollen composition inside a bin changes when boundary-crossing samples are withheld. The original 23-bin score is a different denominator. The source bounds have no stated confidence level or joint age posterior, so this is a conditional marginal-bound sensitivity, not a probabilistic age correction or a forward-time score. The target was already exposed in the original descriptive development study. This result does not measure vegetation, admit LRA inputs or rank methods generally.
The producer verifies the committed prediction and scoring
protocols, every declared Neotoma source part, selected source age-claim values
and bounds, and byte-reproduces the original sample, score and coverage files
before computing this sensitivity. The machine report retains
the source, freeze and original-score identities. Replay into an absent
directory under the repository's artifacts/ root and compare all four output
files byte for byte: