Solsø age support on the 31 author pollen targets¶
We applied the primary LegacyAge exact-depth join to the separate Nordic author-target observed-pollen study. This is a prior-exposed development sensitivity of a closed subcomposition of 31 author taxonomy targets, with author-assigned native pollen NISP as its denominator. The 16-literal-label sensitivity has a different response and denominator; its scores are not combined here.
The original 31-target Danish cohort selects exactly the 91 Solsø samples
with comparable calibrated Neotoma ages from both cores. The S5 join has 92
exact sample-depth matches; the remaining sample, 352996, retains its
negative-bound age refusal. We kept a comparable sample only when its entire
closed S5 95% interval lies inside its original half-open 500-cal-BP bin.
There is no age interpolation, posterior draw or bin reassignment.
| Source support | Samples |
|---|---|
| Exact Solsø sample-depth joins, both cores | 92 |
| Refused calibrated source age | 1 |
| Comparable samples in the 31-target parent | 91 |
| Intervals crossing a bin boundary | 76 |
| Full intervals within the original bin | 15 |
The sample ledger preserves all 92 decisions. The 15 retained samples include 14 from SOLSOE81 and one from SOLSOE83. They occupy eight nominal bins: 2, 3, 4, 5, 6, 12, 15 and 16. The parent scores 24 Solsø bins; 16 have no sample under this strict rule. The protocol pins the author-target axis, source age join, Denmark-excluded training input and prediction hashes. All 24 reused prediction rows for the three fixed methods are in predictions.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 | 0.034878 | 0.035023 | 0.491619 | 0.490921 |
| Nearest site | 0.025673 | 0.025513 | 0.357172 | 0.356094 |
| Inverse distance squared | 0.027231 | 0.027082 | 0.405523 | 0.405143 |
These are one-site, eight-bin descriptive averages, not independent-site uncertainty intervals or a model ranking. The score ledger shows every method/bin pair. Scores on the same retained samples using their original nominal bins exactly match the strict-interval scores, because no retained sample changes bins. The parent columns use complete original pollen compositions in those same eight bins. Their differences reflect sample support restriction, not age reassignment or a new fit.
The producer maps the original Neotoma observation counts through the exact author taxonomy lookup after predictions are frozen. It checks all 24 reconstructed parent bins against the sealed author-target counts and assigned, excluded and unmapped mass. The machine report records source identities, coverage and results. 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/author-target-age-interval/replay.py \
freeze --repository . --output artifacts/jobs/solsoe-author-age-check
PYTHONDONTWRITEBYTECODE=1 \
PYTHONPATH=packages/bijux-pollenomics/src:packages/bijux-pollenomics-dev/src \
.venv/bin/python docs/report/countries/denmark/solsoe-source-lineage/author-target-age-interval/replay.py \
score --repository . \
--author-source-root ../bijux-pollenomics-data/landclim/author-reference \
--output artifacts/jobs/solsoe-author-age-check
LegacyAge is an alternative chronology. This result does not establish the historical LandClim age model or its 79 selected Solsø samples, supply basin geometry, admit LRA input or validate measured vegetation.