Within-cell Rpr correlation verification

These records verify the existing Table S15 correlation summary. They do not
replace the published fitting, modelling or prediction outputs.

Files
- Rpr_Correlation_Check.json: population counts, independent recomputation,
  checkup counts, observed range and cell-level coefficients.
- Rpr_Correlation_Check.csv: cell-level rounded-input recomputation.
- Rpr_Precision_and_Ties_Check.json: full resistance precision, tie exclusion
  and checkup-count sensitivity, with all 30 cell records.
- Rpr_Precision_and_Ties_Check.csv: the same per-cell precision comparison.

Inputs and method
Paths below are relative to Battery_Analysis_Code.
1. Select primary_eligible rows from data/processed_features.csv (12,643 rows).
2. Group by (cell, checkup, source_order), and calculate the median of R_DCIR
   and SOH. R_DCIR is the stored name of the pulse-to-rest indicator Rpr.
   There are 352 checkups in 30 cells.
3. For each cell, assign average ranks for tied values, then calculate the
   Pearson correlation of the two rank vectors. Compare the resulting
   Spearman coefficients with the stored fit_diagnostics/Cell_Aging_Trends.csv.
4. For the resistance-precision check, join the same observations to
   data/extracted_features.csv by their measurement IDs and repeat with
   the resistance values before six-decimal rounding. Keep the SOH targets
   unchanged. Do not reinterpret this as recomputing raw capacity integration.
5. Summarize coefficients with equal weight per cell. Repeat after excluding
   all cells with any tied checkup medians and after requiring >=10 checkups.

Results
The published-input median is -1.000; 18 of 30 cells have exactly reversed
average-rank vectors. Cells have 5-17 checkups (median 13). The maximum absolute
difference from the stored coefficients is 3.33e-16.
Five cells contain ties in at least one variable. Excluding them leaves
25 cells, 16 with perfectly reversed ranks, and the median remains -1.000.
With resistance values before rounding, 19 cells have perfectly reversed
ranks and the median remains -1.000. Only HE4C22B02_T45_S000_100 changes beyond
numerical precision, from -0.9989004944448661 to -1.0.
The 19 cells with >=10 checkups (13-17 observed) have median -0.9991067439039163.

Scope
These are descriptive correlations of checkup medians, with variable
observation counts and unmatched instantaneous SOC and temperature. Perfect
rank ordering does not imply a perfectly linear relationship at the individual
measurement level or a causal aging effect. SOH retains the precision of the
published target values; the original capacity integration was not repeated.
