Solsø pollen predictions under strict LegacyAge age intervals¶
We used the exact-depth LegacyAge S5 source join to check the age support of the existing Denmark country-held-out observed-pollen predictions. This is a prior-exposed development sensitivity of a closed 16-literal-label Neotoma pollen subcomposition. Its denominator is the sum of those labels' native NISP counts, not all pollen or vegetation cover. It reuses the four already frozen methods without refitting or selecting a new model.
The age rule retains a source-comparable sample only when its full closed S5 95% range fits inside its original half-open 500-cal-BP Neotoma age bin. No age interpolation, posterior draw or reassignment to a neighboring bin is made. Sample support records every core and refusal.
| Source support | Samples |
|---|---|
| Two Solsø cores with exact S5 depth matches | 92 |
| Refused Neotoma calibrated age with negative younger bound | 1 |
| Comparable samples in the parent cohort | 91 |
| Comparable intervals crossing a bin boundary | 76 |
| Full intervals inside their original bin | 15 |
Fourteen retained samples come from SOLSOE81 (Neotoma dataset 4445), and one from SOLSOE83 (dataset 4446). They occupy eight original bins: 2, 3, 4, 5, 6, 12, 15 and 16. The parent has 24 scored Solsø bins; 16 have no retained sample under this rule. All 32 reused prediction rows and their prior Denmark-excluded training identities are pinned in predictions.json and protocol.json. The freeze step opens no target pollen observations.
| Fixed method | Strict eight-bin MAE | Parent MAE, same bins | Strict Hellinger | Parent Hellinger, same bins |
|---|---|---|---|---|
| Training mean | 0.043627 | 0.042678 | 0.468875 | 0.465636 |
| Nearest site | 0.032310 | 0.032104 | 0.386071 | 0.384834 |
| Inverse distance squared | 0.038178 | 0.037560 | 0.435518 | 0.432374 |
| CLR/RBF ridge | 0.040337 | 0.040081 | 0.414663 | 0.414330 |
These are one-site, eight-bin descriptive averages, not uncertainty intervals or a model ranking. The score ledger retains every method/bin pair and its sample count. The nominal score on exactly the retained samples equals the strict-interval score because every retained sample stays in its original bin. The parent columns score the full original pollen composition in those same eight bins. Their small differences reflect sample support restriction, not an age-driven change in bin membership or a better fit.
The producer reconstructs all 24 parent bin counts from the original Neotoma observation part after the predictions have been frozen. It checks each parent bin's sample, total NISP and 16-label count denominator. The machine report retains source hashes and denominators. Replay from the repository root into an absent disposable output directory:
PYTHONDONTWRITEBYTECODE=1 \
PYTHONPATH=packages/bijux-pollenomics/src:packages/bijux-pollenomics-dev/src \
.venv/bin/python docs/report/countries/denmark/solsoe-source-lineage/legacyage-interval-sensitivity/replay.py \
freeze --repository . --output artifacts/jobs/solsoe-legacyage-pollen-check
PYTHONDONTWRITEBYTECODE=1 \
PYTHONPATH=packages/bijux-pollenomics/src:packages/bijux-pollenomics-dev/src \
.venv/bin/python docs/report/countries/denmark/solsoe-source-lineage/legacyage-interval-sensitivity/replay.py \
score --repository . --output artifacts/jobs/solsoe-legacyage-pollen-check
LegacyAge is an alternative chronology; this result does not identify the historical LandClim age model or its 79 selected Solsø samples. It does not admit LRA input or validate vegetation against independent field observations.