# 2025→2026 common-item drift and historical-prior audit

Created: 2026-06-15T00:47:05Z

## Subtest summary
| year_level | subtest_2026 | n_crosswalk_items | n_high_confidence_matches | n_prior_eligible_items | median_abs_logit_drift | p90_abs_logit_drift | recommendation |
|---|---|---|---|---|---|---|---|
| foundation | BNL0-20 | 10 | 10 | 10 | 0.169 | 0.43 | historical_item_priors_supported_with_drift_inflation |
| foundation | DMT10_2026 | 11 | 7 | 0 | 0.269 | 3.335 | use_2025_as_subtest_context_not_hard_item_anchor |
| foundation | MC0-20 | 60 | 0 | 0 | 6.902 | 9.5 | no_item_level_historical_prior |
| foundation | MNC0-20 | 30 | 30 | 5 | 4.845 | 8.828 | selective_item_priors_after_manual_review |
| foundation | MQ1-20 | 30 | 30 | 3 | 6.462 | 9.234 | selective_item_priors_after_manual_review |
| year1 | AAMC | 40 | 40 | 1 | 4.995 | 9.864 | selective_item_priors_after_manual_review |
| year1 | ASMC | 30 | 30 | 3 | 4.797 | 8.616 | selective_item_priors_after_manual_review |
| year1 | BNL0-100 | 13 | 13 | 13 | 0.141 | 0.285 | historical_item_priors_supported_with_drift_inflation |
| year1 | MC0-100 | 60 | 1 | 1 | 8.868 | 10.309 | no_item_level_historical_prior |
| year1 | MNC0-100 | 29 | 29 | 4 | 4.775 | 8.225 | selective_item_priors_after_manual_review |


## 2025 restricted reliability context
| year_level | term_scope | item_scope | fit_status | n_person | n_items | reliability | converged |
|---|---|---|---|---|---|---|---|
| foundation | term3 | kept_crosswalk | ok | 1440 | 93 | 0.903 | TRUE |
| foundation | term3 | all_crosswalk | ok | 1440 | 101 | 0.904 | TRUE |
| foundation | term4 | kept_crosswalk | ok | 1096 | 84 | 0.889 | TRUE |
| foundation | term4 | all_crosswalk | ok | 1096 | 90 | 0.89 | TRUE |
| foundation | term3_4_pooled | kept_crosswalk | ok | 2536 | 100 | 0.905 | TRUE |
| foundation | term3_4_pooled | all_crosswalk | ok | 2536 | 118 | 0.906 | TRUE |
| year1 | term3 | kept_crosswalk | ok | 1501 | 121 | 0.945 | TRUE |
| year1 | term3 | all_crosswalk | ok | 1501 | 134 | 0.945 | TRUE |
| year1 | term4 | kept_crosswalk | ok | 1067 | 105 | 0.916 | TRUE |
| year1 | term4 | all_crosswalk | ok | 1067 | 109 | 0.917 | TRUE |
| year1 | term3_4_pooled | kept_crosswalk | ok | 2568 | 132 | 0.949 | TRUE |
| year1 | term3_4_pooled | all_crosswalk | ok | 2568 | 148 | 0.95 | TRUE |


## Recommendation

Use 2025 as prior evidence only where item identity is strong/moderate and proxy drift is acceptable. Treat DMT10_2026 as the main caveat: it should receive at most weak subtest-level context, not hard item anchoring. For strong BNL and other exact-ID subtests, weak/commensurate priors with drift inflation are defensible if final validation supports Bayesian updating.

## Output tables
- `outputs/runs/irt-2026-boy-subtest-audit/latest/tables/model_review/historical_prior_drift/common_item_drift_prior_eligibility.csv`
- `outputs/runs/irt-2026-boy-subtest-audit/latest/tables/model_review/historical_prior_drift/common_item_drift_prior_summary.csv`
- `outputs/runs/irt-2026-boy-subtest-audit/latest/tables/model_review/historical_prior_drift/restricted_2025_t3t4_reliability_context.csv`