Validation lab
| Test | Reference | Δ / tolerance | Time | |
|---|---|---|---|---|
Control: correctly measured oak dates at its known year with no ring flags Known position (synthetic SG-OAK-014, ring 1 = 1337) | Known position (synthetic SG-OAK-014, ring 1 = 1337) challenge | — | ||
Missing ring: one ring deleted at ring 70 is located and an insert proposed Injected fault position (ring 70) | Injected fault position (ring 70) challenge | — | ||
False ring: an extra ring inserted after ring 69 is located and a merge proposed Injected fault position (ring 70) | Injected fault position (ring 70) challenge | — | ||
Low-correlation segment: 50 shuffled rings (1400–1449) are flagged, the date survives COFECHA flag A (r below 99% critical) on the damaged segment | COFECHA flag A (r below 99% critical) on the damaged segment challenge | — | ||
Frost / pointer year: a narrow ring injected into every real oak series at 1010 is detected Cropper (1979) and Becker et al. (1994) pointer-year rules, ≥ 75% of series | Cropper (1979) and Becker et al. (1994) pointer-year rules, ≥ 75% of series challenge | — |
| Test | Reference | Δ / tolerance | Time | |
|---|---|---|---|---|
Tukey biweight robust mean (C = 9) per series dplR 1.7.8 tbrm | dplR 1.7.8 tbrm parity | — | ||
Mean sensitivity sens1 and sens2 per series dplR 1.7.8 sens1 / sens2 | dplR 1.7.8 sens1 / sens2 parity | — | ||
Chronology: biweight robust mean and arithmetic mean per year dplR 1.7.8 chron (biweight = TRUE / FALSE) | dplR 1.7.8 chron (biweight = TRUE / FALSE) parity | — | ||
Signal-free standardisation (Melvin & Briffa 2008) dplR 1.7.8 ssf | dplR 1.7.8 ssf parity | — | ||
Cubic smoothing spline, 50% response at 32 years dplR 1.7.8 caps(nyrs = 32, f = 0.5) | dplR 1.7.8 caps(nyrs = 32, f = 0.5) parity | — | ||
Lag-1 autocorrelation used for AR(1) pre-whitening R 4.5.3 acf(x)$acf[2] | R 4.5.3 acf(x)$acf[2] parity | — | ||
Mean inter-series correlation r̄ and EPS (Wigley et al. 1984) dplR 1.7.8 rwi.stats (Pearson, min overlap 30) | dplR 1.7.8 rwi.stats (Pearson, min overlap 30) parity | — | ||
Subsample Signal Strength per year (Wigley et al. 1984) dplR 1.7.8 sss | dplR 1.7.8 sss parity | — |
| Test | Reference | Δ / tolerance | Time | |
|---|---|---|---|---|
99% one-tailed critical correlation for segment lengths 20–150 R qt(0.99, n − 2) and COFECHA's printed 0.3281 for 50-yr segments | R qt(0.99, n − 2) and COFECHA's printed 0.3281 for 50-yr segments parity | — | ||
Segment correlations (50-yr, lagged 25) and their flags against a leave-one-out master dplR 1.7.8 corr.rwl.seg (Pearson, prewhiten, p < 0.01) | dplR 1.7.8 corr.rwl.seg (Pearson, prewhiten, p < 0.01) approximation | — |
| Test | Reference | Δ / tolerance | Time | |
|---|---|---|---|---|
Baillie–Pilcher and Hollstein t-values for every overlapping pair Base-R reference formulas (R 4.5.3; dplR has no t_BP / t_H) | Base-R reference formulas (R 4.5.3; dplR has no t_BP / t_H) parity | — | ||
Gleichläufigkeit for every overlapping pair dplR 1.7.8 glk (Buras & Wilmking 2015 tie handling) | dplR 1.7.8 glk (Buras & Wilmking 2015 tie handling) parity | — | ||
Gleichläufigkeit significance (s = 1 / 2√n) dplR 1.7.8 glk p_mat (two-sided) | dplR 1.7.8 glk p_mat (two-sided) convention | — | ||
Student t upper-tail p-values R 4.5.3 pt(t, df, lower.tail = FALSE) | R 4.5.3 pt(t, df, lower.tail = FALSE) parity | — |
| Test | Reference | Δ / tolerance | Time | |
|---|---|---|---|---|
Modified negative exponential growth curves dplR 1.7.8 detrend.series (ModNegExp) | dplR 1.7.8 detrend.series (ModNegExp) convention | — | ||
Hugershoff growth curves dplR 1.7.8 detrend.series (ModHugershoff) | dplR 1.7.8 detrend.series (ModHugershoff) convention | — |
Real-data reference. ITRDB brit — Rostrevor oak (Quercus), Northern Ireland, 14 series 812–1458 (NOAA NCEI, public domain). Reference values were produced by R 4.5.3 with dplR 1.7.8 and are bundled, so the suite runs offline. dplR has no t_BP or t_H; those are checked against independent base-R formulas.
Challenge series. SG-OAK-014 against the Central European oak master (synthetic demonstration data), with a deleted ring, an inserted ring and 50 shuffled rings; the frost-year case injects a narrow ring into every real Rostrevor series covering one year.
Pass — within the declared tolerance. Warn — divergence explained by a documented convention or optimiser difference, inside its declared band. Fail — outside every band.
Peer review. The R audit script embeds the raw file and every value this browser computed, re-computes the references with dplR and stops on any parity failure. The certificate fingerprint is the SHA-256 of the device-independent results, so a reviewer reproducing the same values obtains the same fingerprint.