Methodology

1. Principles

• Reproducibility. Each statistic is a deterministic function of the measured ring widths and the chosen master. Ring edits are stored as part of the series and covered by a SHA-256 integrity hash.
• Transparency over automation. Automated outputs (best positions, ring-error flags, wiggle-matches) are proposals. A date becomes part of the record only when the analyst commits it.
• Multiple-comparison control. Sliding a series across a master tests hundreds of positions. Significance is always reported after false-discovery-rate control, never as raw p alone.
• Provenance of reference data. Each master is labelled as synthetic demonstration data, real ITRDB data, or user-built. Demonstration data is never valid for formal dating.
• Stated limitations. Where an implementation approximates a published method, this page and the interface say so.

2. Standards alignment

The dossier formats follow these frameworks' structure and terminology. They are not certified by the named bodies; submission requirements remain with the commissioning authority.

3. Cross-dating statistics

Ring widths are in 1/100 mm. At each candidate position, only overlapping rings are compared, with a minimum overlap of 50 rings by default (adjustable per analysis).

Optional standardisation before matching

By default the statistics above are computed on raw ring widths, as in the classic definitions. The analyst can standardise sample and master identically first; t_BP, t_H, Glk and r are then computed on the result, and every dossier and audit script records the choice.

Nonlinear parameters are found by a deterministic grid search with closed-form linear solves, so results are reproducible bit-for-bit. Indices and pre-whitened values are floored at 0.02 of their mean so log transforms stay defined.

Pointer years and signature years

• Cropper (1979): Cₜ = (xₜ − mean)/sd in a centred 5-yr window; event year when |C| > 1; Neuwirth et al. (2007) classes weak / strong / extreme at 1 / 1.28 / 1.645.
• Becker (1989): relative growth change against the mean of the 4 preceding years; event year when |change| ≥ 10%.
• Pointer years across ≥ 4 series: ≥ 75% of the series present show a same-sign event (Schweingruber et al. 1990). A single mean chronology yields event years only.
• Pointer consistency: Gleichläufigkeit restricted to the master's event years at the current position, with the Glk normal approximation for significance.
• Signature years drawn on the scrubber: 536 · LALIA onset (Büntgen et al. 2016); 1540 · megadrought (Wetter et al. 2014); 1601 · Huaynaputina (Briffa et al. 1998); 1816 · Tambora (Briffa et al. 1998). They are visual references only and do not enter any statistic.

4. Confidence & multiple comparisons

For every series–master pair, all admissible positions are tested. Benjamini–Hochberg q-values are computed across that whole set. In batch matching, the q-values are pooled across every position and every master tested for the specimen, and the reporting threshold is q < 0.01.

A match is marked ambiguous when the runner-up position has t_BP ≥ 3.5 and lies within 1 t of the best. Beyond 450 km between specimen and master site, candidates are re-ranked by t·exp(−(d − 450)/450). This only reorders candidates: significant matches always rank ahead of non-significant ones, t, p and q are never altered, and distant best matches are flagged as possible imported timber.

Batch matching and pre-issue verification display each computation stage for at least 650 ms so the analyst can follow it. The true compute time of each stage is reported alongside.

5. Chronology construction

• Masters from dated series: each series is indexed as value / 31-yr centred running mean × 100; indices are combined per calendar year with Tukey's biweight robust mean (one step from the median, u = (x − median)/(9·MAD + 10⁻⁶), weights (1 − u²)² for |u| ≤ 1 — identical to dplR tbrm / chron(biweight = TRUE)), or optionally the arithmetic mean. Sample depth is recorded per year. Bundled reference chronologies were generated before method 1.16 with the arithmetic mean and are kept as generated.
• Signal-free standardisation (optional, Melvin & Briffa 2008): the simple signal-free iteration of dplR ssf(method = "Spline"). Each series is detrended with a cubic smoothing spline of 50% response at 66.67% of its length (integer stiffness, as dplR caps); the biweight chronology is divided out of the measurements, the signal-free series are rescaled to their original means (years with depth 1 keep the raw values), curves are refitted and the chronology rebuilt, until the median absolute change of the high-frequency chronology (minus a spline at the median series length, weighted by √(n−1)/√(max n−1)) is ≤ 0.0005, at most 25 iterations. It needs at least two series and continuous coverage.
• Subsample Signal Strength: SSS = n(1 + (N−1)r̄) / (N(1 + (n−1)r̄)) (Wigley, Briffa & Jones 1984; dplR sss formula), N = number of series in the chronology, n = depth, r̄ running where available; threshold 0.85. Shown on every master view; not computable for published chronologies without r̄.
• Mean sensitivity: ms = Σ|2(x_t − x_{t−1})/(x_t + x_{t−1})| / (n − 1) (dplR sens1) and Σ|Δx| / (Σx − Σx/n) (dplR sens2, Biondi & Qeadan 2008); running ms in 50-yr centred windows on every master view.
• Replication: Expressed Population Signal EPS = n·r̄ / (1 + (n−1)·r̄) (Wigley, Briffa & Jones 1984). r̄ is the running mean inter-series correlation of Baillie–Pilcher indices in 50-yr windows lagged 25 yr (pairs with ≥ 30 common years), interpolated between window centres; masters without member series use one r̄. Years with EPS < 0.85 or sample depth < 10 are exploratory: shaded on every master view and on the scrubber, counted in the live replication metric, and listed by period in the dossier.
• Long-span composites: built from openly published ITRDB raw measurements in the authors' own dating, by the same indexing, with per-year depth and running r̄. British Isles medieval oak composite (10 ITRDB sites): 812 CE–1458 CE, max depth 129; Torneträsk Scots pine, Holocene (ITRDB SWED334): 5408 BCE–2010 CE, max depth 178; Kerimäki subfossil pine composite (3 ITRDB sites): 559 CE–1732 CE, max depth 100; Abcoude subfossil oak, Iron Age–Roman (ITRDB NETH019): 325 BCE–30 CE, max depth 10.
• Site synthesis: all timber pairs are cross-matched (dated pairs at their calendar positions, floating pairs at their best relative position, overlap ≥ 40). Timbers are grouped by single linkage on pairs with t_BP ≥ 5 and Glk ≥ 65% (adjustable); each group's composite is dated as one, with sample depth and EPS from the group's mean inter-series r. Phase felling dates are the intersection of the members' felling windows; non-overlapping windows are flagged.
• Felling season: read from the final ring under waney edge — complete latewood → winter felling (shown as e.g. "Winter 1384/85", calendar window E to E+1), earlywood only → summer (calendar year E). Recorded by the analyst, not inferred.
• Floating site masters: timbers are aligned greedily against a growing composite (pairwise t_BP ≥ 4) and combined with a Tukey biweight robust mean (c = 9, MAD scale). Leave-one-out member statistics are reported.

6. Quality control

After COFECHA (Holmes 1983; Grissino-Mayer 2001).

• Filtering: log transform, then a high-pass filter, then autoregressive prewhitening (Yule–Walker, order ≤ 3 by AIC). Series and master are filtered identically. The high-pass is a Gaussian with 50% amplitude response at 32 years — an approximation of COFECHA's 32-yr cubic smoothing spline.
• Segments: 50-yr segments lagged 25 yr. Flag A: r below the 99% one-tailed critical value. Flag B: higher r at a shift within ±10 yr.
• Outlier rings: filtered values above +3 SD or below −4.5 SD.
• Missing / false rings: whole-series breakpoint search for a ±1-yr shift. A break is accepted when the tail correlation passes the 99% critical r and improves by more than 0.2; up to three breaks are reported. The reported position is an estimate with a stated ring range; corrections are applied only on analyst action and remain undoable.
• Skeleton plot: ring width relative to the ±5-ring local mean, scored 0–10 (Stokes & Smiley 1968).

7. Felling-date estimation

Each sapwood model is a log-normal distribution of total sapwood ring count, parameterised from the published median and 95% range. The interval reported is the 95% interval conditional on the sapwood rings observed and any missing outer rings estimated by the analyst. With waney edge present, the felling year is the last ring; with heartwood only, the result is a terminus post quem.

Model parameters are summary approximations of the cited distributions. Confirm them against the source publication, or replace them with a laboratory-specific model, before formal reporting.

8. Radiocarbon & cosmogenic anchors

• Calibration curve: IntCal20 (Reimer et al. 2020), 6,501 nodes, 10551 BCE – 1950 CE, linearly interpolated between nodes.
• Dendro-wiggle-matching: for each candidate calendar position of ring 1, χ² = Σ (ageᵢ − C(tᵢ))² / (σᵢ² + σ_C(tᵢ)²) with tᵢ fixed by ring count. A uniform prior over positions gives relative probabilities; the 95.4% range is the highest-probability set. Agreement test: χ² below the 5% critical value with n−1 df (Ward & Wilson 1978; Bronk Ramsey et al. 2001).
• Combined multi-proxy date: on a common grid of ring-1 positions, the ring-width proxy contributes 1 on its admissible set (committed date, else positions with BH q < 0.05; uninformative if none), ¹⁴C contributes exp(−χ²/2), and an observed Miyake onset ring contributes 1 within ±tol yr of the event. The normalised product gives the combined distribution and its 95.39999999999999% range. Pairwise agreement is tested separately: fixed-position χ² (df = n) for ring width × ¹⁴C and ¹⁴C × Miyake, and offset ≤ tol for ring width × Miyake. Verdicts: concordant, discordant, or incompatible (no common position).
• Bayesian wiggle-match (¹⁴C studio): θ = calendar year of ring 1 with a uniform prior (whole IntCal20, ring-width candidates ± 25 yr, or a custom range). Assay i on rings aᵢ…bᵢ is compared with the IntCal20 mean over that block, μᵢ(θ), with sᵢ² = σᵢ² + σ̄_C,ᵢ². Likelihood Π N(yᵢ; μᵢ, sᵢ); with the outlier model Π[(1−q)N(yᵢ; μᵢ, sᵢ) + qN(yᵢ; μᵢ, k·sᵢ)], q = 0.05, k = 10 (after Bronk Ramsey 2009), with posterior outlier probabilities per assay. The posterior is computed exactly on a 1-yr grid; 68.3% and 95.4% highest-posterior-density sets are reported as (possibly disjoint) calendar ranges for ring 1 and the last ring. Agreement: Ward & Wilson χ² at the mode against χ²(0.05, n−1).
• Convergence cross-check: an independent Metropolis–Hastings sampler (4 chains, over-dispersed starts, local ±sd and 10% global proposals) is run over the posterior support (density > 10⁻⁹ of the maximum, ± 5 sd). Reported: Gelman–Rubin R̂ (target ≤ 1.01), effective sample size (Geyer initial positive sequence, target ≥ 400) and total-variation distance to the exact posterior. Reported dates never come from the sampler.
• Miyake events (annual ¹⁴C excursions): 7176 BCE (first elevated ring 7176 BCE ±1), 5259 BCE (first elevated ring 5259 BCE ±1), ≈660 BCE (664–663 BCE) (first elevated ring 664 BCE ±1), 774/775 CE (first elevated ring 775 CE), 993/994 CE (first elevated ring 994 CE). Single-year detector: conventional ages of single-ring assays are converted to ¹⁴C concentration F = e^(−age/8033); for every split between consecutive rings the mean F of up to 4 rings after vs before gives the step in ‰ and its standard error. A step is detected at ≥ 8‰, z ≥ 4, with ≥ 60% of the rise between two consecutive rings. Each event then implies one position of ring 1; its identity is tested with the assays outside the step window only (their posterior mass at that position and χ² agreement), because IntCal20 smooths annual excursions. Locking requires exactly one consistent event; the onset tolerance of that event is stated.

9. Optical pre-flight, photo & micro-CT measurement

Checks run on a greyscale copy down-sampled to at most 1024 px.

• Sharpness: variance of the Laplacian (Pech-Pacheco et al. 2000). Sharp ≥ 150, marginal ≥ 60, otherwise blurred. Per-tile values reveal uneven focus.
• Exposure: clipping is flagged above 2% blown highlights or 5% crushed shadows.
• Scale / fiducial detection: autocorrelation of edge profiles in each border band detects a periodic tick pattern (mm ruler), confirmed by its harmonic, giving an estimate of px/mm.
• Perspective skew: estimated from the moment ellipse of strong edges, since concentric rings should appear near-circular when the camera is orthogonal. Warning above 18°, fail above 30°.

The optical checks are heuristic image-quality screens, not metrological calibration.

Field pre-flight (Field mode)

• Score (0–100, pass ≥ 80): weighted mean of sharpness (30%; Laplacian variance), lighting uniformity (20%; coefficient of variation of 4×4 tile brightness, fail above 35%), contrast (15%; 5–95% luminance range, warn below 60), exposure (15%; clipped pixels) and resolution (20%; at least 4 px across the narrowest expected ring, default 0.2 mm). Blur below the marginal threshold or very uneven light fails regardless of the score.
• Perspective correction: the analyst pins the four corners of a rectangular face; a homography is solved by the direct linear transform (Hartley & Zisserman 2003) and the image resampled bilinearly. With the physical size of the pinned rectangle, the output has a uniform, known px/mm.
• Scale validation: px/mm from a line drawn along a ruler of known length is compared with the tick period found by autocorrelation along the same line (warn above 2% disagreement); the tick period in each half of the line reveals residual perspective (warn above 3%).
• Hand-off: the rectified image and its px/mm pass to the Measurement lab in the same session; the pre-flight summary and the source image SHA-256 are written into the series notes. Images are not stored.

Photo ring measurement

• Fiducial auto-calibration: every row and column is scanned for a periodic, high-contrast tick pattern (autocorrelation of the high-passed profile, reduced to its fundamental so 5 / 10 mm marks are not mistaken for the graduation); ticks are located as darkness-weighted centroids and the longest regular run (spacing within ±35% of the period, ≥ 15 ticks) defines a ruler of known length at 1 mm per tick.
• Lens distortion: one-parameter radial model p_u = c + (p_d − c)(1 + k₁ r²), c = image centre, r normalised by the half-diagonal (Brown 1966). k₁ is the value that makes the ruler's ticks most evenly spaced (minimum coefficient of variation of tick spacing; grid search −0.3…0.3 with refinement). When the implied scale change at the corners exceeds 0.2% the analyst can resample the image (inverse mapping by Newton iteration, bilinear) and recalibrate; k₁ is recorded with the series. A single ruler constrains k₁ only along its own line, and decentring / tangential distortion is not modelled.
• Scale: a line drawn over a known length on a scale in the image plane (px/mm = pixel distance / known length), or the scan resolution (px/mm = DPI / 25.4), or an entered mm/px value. Resolution is reported, with a warning when the narrowest ring spans fewer than 4 px.
• Transect: luminance profile at 1-px steps along a pith → bark line, averaged over ±6 px perpendicular to it. Widths are boundary spacings along the transect, so it must cross the rings perpendicularly.
• Boundary suggestion: latewood → earlywood transitions, detected as positive derivative peaks of the lightly smoothed profile. Peaks must exceed median + kσ (σ from the MAD; k adjustable, default 6), with non-maximum suppression at a 0.12 mm minimum spacing and parabolic sub-pixel refinement.
• Multi-ray consensus: 3–7 rays (a fan from the pith, or parallel offsets on cores) are traced with the same boundary detector; the main transect keeps its confirmed boundaries. Ring widths are normalised by an 11-ring running mean and each ray is aligned to the ray with the modal ring count by dynamic programming with three operations: 1:1 match (cost |ln(w_ref / w_ray)|), merge — two reference rings in one ray ring, a locally absent ring on that ray — and split — one reference ring as two ray rings, a false (intra-annual) ring — each with penalty 0.9. Flags are suggestions; fixes on the main transect are analyst actions and undoable.
• Earlywood / latewood: within each ring the transition is the first point after the brightest earlywood where the lightly smoothed luminance falls below the midpoint of that maximum and the ring's latewood minimum (50% threshold, the image analogue of the density-midpoint definition; Koubaa et al. 2002). EW and LW widths are measured on the unenhanced image and stored with the series.
• Enhancement for degraded wood: CLAHE on luminance (Zuiderveld 1994; 8×8 tiles, clip limit 2.5, bilinear blending, colour scaled by the luminance ratio) and directional Gaussian smoothing along the ring tangent (perpendicular to the transect). Both feed display and boundary snapping only — never widths, EW/LW or Blue Intensity.
• Missing pith: the analyst places ≥ 3 points on the innermost visible ring; a least-squares (Kåsa algebraic) circle fit gives the estimated pith and the RMS residual. Missing inner rings = distance from the transect start to the estimated pith ÷ mean width of the innermost 5 measured rings (window adjustable).
• Analyst verification: every suggested boundary can be moved, deleted or supplemented. The series records which boundaries were auto-suggested, together with the image SHA-256, calibration and transect.

Blue Intensity (ΔBI) micro-densitometry

• Colour calibration: the analyst samples up to six neutral patches (White 9.5, Neutral 8, Neutral 6.5, Neutral 5, Neutral 3.5, Black 2; nominal ColorChecker Classic sRGB values, replace with the target's certificate values). Per channel, gain and offset are fitted by least squares in linear light (IEC 61966-2-1 sRGB transfer function); one patch gives gain only. The RMS patch error before and after is recorded with the series.
• Profile and per-ring BI: calibrated blue channel along the transect (±band px). EWB = mean of the brightest 50% of pixels in the inner 60% of each ring; LWB = mean of the darkest 25% in the outer 40%. ΔBI = EWB − LWB (Björklund et al. 2014); inverted latewood BI = 255 − LWB (McCarroll et al. 2002). Rings narrower than 8 px are not scored.
• Dual-proxy dating: t_BI from Pearson r of high-passed ΔBI (value − 5-yr centred mean) against a master ΔBI chronology. t_BP and t_BI one-sided p-values become normal scores and are combined by weighted Stouffer Z = (w·z_TRW + (1−w)·z_BI)/√(w² + (1−w)²) (Whitlock 2005), default w = 0.5, with BH-FDR over all positions. This assumes the proxies are independent.

BI is sensitive to surface preparation, resin and heartwood colour shifts that a neutral-patch calibration cannot remove; it is validated here on synthetic photographs only. The bundled ΔBI master (Alpine Spruce) is synthetic demonstration data; real ΔBI masters must be loaded for formal use.

Boundary suggestion assumes a distinct latewood → earlywood contrast (conifers, ring-porous oak). It is unreliable on diffuse-porous species, on surfaces that are not sanded or polished, and on rings narrower than a few pixels. It is validated here on synthetic photographs only.

Micro-CT virtual coring

• Volume and orthoslices: a density volume (kg/m³; imported slice stacks are mapped linearly from grey values with analyst-entered end points) shown as axial (transverse), radial (longitudinal through the pith row) and tangential sections. Window / level maps level to mid-grey over a density range width; display only, never measurement. The axial slice may be averaged over ±1 neighbouring slices.
• Ring orientation: structure tensor of the axial slice (Bigün & Granlund 1987): central-difference gradients of a lightly smoothed slice, outer products averaged by 3 box passes of radius 6 px (≈ Gaussian). The dominant eigenvector is the local ring normal, i.e. the direction of true radial growth; coherence (λ₁ − λ₂)/(λ₁ + λ₂) measures how ring-like the neighbourhood is.
• Curved rays: centripetal Catmull–Rom splines (α = 0.5; Yuksel et al. 2011) through analyst control points, resampled at uniform arc length with local tangents and normals. Auto-trace follows the orthogonal trajectory of the ring field inward from the outer point, where trajectories converge on the pith: each 1-px step blends the previous direction and the ring normal by coherence and turns at most 6° per step, so decay pockets, checks and vessels do not derail it. Control points are then placed adaptively — every 5 mm, or sooner after 4° of cumulative turning — and remain editable.
• Unrolling: density sampled every 20 µm of arc length (bilinear), median across ±0.5 mm along the local ring tangent, so inclusions, vessels and checks that touch only part of the band are rejected. Distances are arc lengths in mm from the calibrated voxel size. The profile ends where density stays below 150 kg/m³ for 0.6 mm (the timber surface), which closes the last ring.
• Ring boundaries: latewood → earlywood density drops. The profile is first contrast-normalised by its running mean and standard deviation over 2 mm, so boundaries in decayed, low-density wood count as much as in sound wood; drops are then found by the photo-measurement detector on the negated profile (median + 4 × MAD, 0.15 mm minimum spacing, sub-sample refinement). Per ring, mean, maximum (latewood, the MXD analogue) and minimum density are reported.
• Quality: the angle between the ray and the local ring normal is reported per sample (colour on the slice), as mean and 95th percentile, and compared with the straight chord between the same end points; widths are true radial widths only where the angle is small.
• Hand-off: CT-derived series are stored with format "ct", the volume SHA-256, voxel size, slice and slab, ray control points, step, band, every boundary and per-ring density. The straightened core can be verified in the Measurement lab, where the series stays CT-derived; it is rendered with the same along-ring median and a fixed 100–950 kg/m³ grey mapping (dense = dark) whatever the display window, and ends on the last latewood maximum before the surface. Volumes are not stored.

The bundled CT volumes (distorted oak, degraded spruce) are synthetic and generated from known growth-space ring widths, so ray accuracy is tested against the analytic orthogonal trajectory. Beam hardening, ring artefacts and partial-volume effects of real scanners are not modelled, and imported grey values are only as calibrated as the analyst's end points.

10. Spatial provenance

Provenance screening (dendroprovenancing; Bridge 2012) asks which reference network a timber's growth signal resembles most. Highest similarity marks the most likely growth region, not proof of origin.

• Matching: every georeferenced master is compared with the specimen at its committed date (fixed position; BH-FDR across masters), or, if undated, at each master's best position (q over that master's scan, then BH-adjusted again across masters). A master supports a source only at q < 0.05.
• Relative support: softmax of t̄ = (t_BP + t_H)/2 over supporting masters, temperature τ = 1.
• Prior: max(distance decay, trade plausibility). Distance decay from the find spot is 1 within 450 km and exp(−(d − 450) / 450) beyond. Trade plausibility = exp(−d_source / 160) · exp(−d_find / 120), where d_source is the master's distance from a corridor's loading section and d_find the find spot's distance from the downstream section, maximised over 5 corridors: Vistula rafting → Gdańsk → Øresund → Low Countries; Neman → Klaipėda → Baltic → North Sea; Daugava → Riga → Gotland → Øresund → London; Rhine timber rafts (Black Forest → Mainz → Dordrecht); Moselle floating (Trier → Koblenz).
• ⁸⁷Sr/⁸⁶Sr screen (optional): when a measured ratio ± 2σ is entered, networks whose baseline range does not overlap it are multiplied by 0.1.
• Weight: support × prior × Sr factor, normalised over candidates and summed per network. The 8 networks are schematic catchments (centre and radius); masters are assigned to the nearest centre within 1.5 radii.
• Surface: inverse-distance-weighted (power 2) interpolation of t̄ between masters, left blank beyond 650 km from any master. It visualises the spread of similarity, not a growth map.
• Map: Lambert azimuthal equal-area projection centred 52°N 10°E (the ETRS89-LAEA centre); coastlines and rivers from Natural Earth 1:50m. Exports: GeoJSON (WGS84) and SVG.

The weight ranks candidate sources for screening; it is not a calibrated probability of origin. Network extents, trade corridors (generalised from Ważny 2002 and Daly 2007) and Sr ranges are schematic. The Sr ranges are illustrative values from regional geology, not a published bioavailable isoscape, and should be replaced with a calibrated isoscape before exclusions are relied on. Distance weighting never changes t, p or q.

11. EUDR timber due diligence

The EUDR module supports operators placing wood on the EU market under Regulation (EU) 2023/1115 as amended by Regulation (EU) 2025/2650. It structures the information of Art. 9, runs a rule-based risk assessment (Art. 10), records mitigation (Art. 11) and assembles the due diligence statement content of Annex II. It is decision support; the operator remains responsible.

• Country benchmark: the Annex of Commission Implementing Regulation (EU) 2025/1093 — 140 low-risk countries, 4 high-risk (Belarus, Democratic People's Republic of Korea, Myanmar, Russian Federation); every other country is standard risk. Simplified due diligence (Art. 13) applies only when all countries of production are low risk and no unmitigated risk indicator is open. Wood from Russian Federation and Belarus is additionally blocked by EU restrictive measures.
• Geolocation: WGS-84 with at least 6 decimal digits (Art. 2(28)), counted from the entered text so trailing zeros count; a single point is accepted only for plots up to 4 ha. Polygons are closed automatically and rejected if self-crossing; area is the geodesic area on a sphere of radius 6,378,137 m (Chamberlain & Duquette 2007).
• Deforestation-free (Art. 3(a)): cut-off date 2020-12-31. Dendrotrace holds no satellite data and never infers forest status: the operator records the check, its source and date, and links the evidence document. “Not checked” is a high finding for standard/high-risk origins and medium for low-risk ones; loss or degradation after the cut-off blocks the shipment.
• Evidence: documents are identified by SHA-256 computed on the device; the files are not stored or uploaded.
• Species screening: a macroscopic key of 7 features (IAWA terminology, simplified after Ruffinatto & Crivellaro 2019) over 22 taxa. With at least 3 features recorded, any feature outside the declared taxon's profile makes the declaration inconsistent. Consistent CITES-listed taxa (status as at CoP19) are reported as look-alikes. Screening can contradict a declaration, never confirm a species.
• Felling-date check: for each linked, committed specimen the felling estimate of §7 is compared with the declared harvest window. Latest possible felling before the window means reclaimed or stored timber or a mis-declaration (high); earliest felling after the window is impossible (blocking). Wood felled before 2023-06-29 is outside the Regulation (Art. 1(2)) if documented. Heartwood-only dates can contradict but not confirm.
• Origin check: signal-only support from §10 (no distance or trade prior, so the declaration cannot bias the test). The declared plot is supported when the strongest significant network's catchment plus 150 km contains it; outside every bundled network the result is “no reference coverage”.
• Verdict: any open blocking finding → prohibited; otherwise any open high or medium finding → non-negligible; else negligible. Blocking findings cannot be mitigated, only resolved in the data.

Application dates: 2026-12-30 for large and medium operators, 2027-06-30 for micro and small operators, per Regulation (EU) 2025/2650. The HS heading list is a selection of Annex I; verify headings and operator obligations against the consolidated legal text. Reference networks cover Europe only, so tropical origins cannot be tested dendrochronologically with the bundled data.

12. Reference data & licences

13. Known limitations

• Bundled ITRDB chronologies are single-site series, not regional master chronologies. Use licensed regional references for formal dates.
• The Hohenheim, Belfast (Irish oak) and South German oak long chronologies are not openly published as raw measurements and are not bundled; load licensed copies as custom masters. Bundled long composites use the BCE year numbering of their source files.
• Batch matching on the Specimens page uses raw ring widths regardless of the standardisation chosen on the cross-dating canvas.
• Most bundled provenance references are living-tree chronologies covering roughly 1450–2020; before about 1450 only a few (e.g. Maastricht roof timbers, East Pomerania, Rostrevor) overlap, so provenance of medieval timber rests on fewer networks.
• The QC high-pass filter approximates COFECHA's spline, so segment correlations may differ slightly from COFECHA output.
• Ring-error positions are statistical estimates; in testing they are typically located within a few rings, occasionally more. Always inspect the sample surface.
• Sapwood models use summary parameters; regional and laboratory-specific models may differ.
• Wiggle-matching assumes correct ring counts between ¹⁴C samples and uses a uniform prior; it does not model reservoir or regional ¹⁴C offsets.
• Spatial re-ranking is a screening heuristic, not a statistical test.
• Provenance screening can only rank regions that have a loaded reference; a timber from an unrepresented region will appear to match whichever loaded network is most similar. Network outlines, trade corridors and Sr ranges are schematic.
• Missing-inner-ring estimates assume constant juvenile growth at the rate of the innermost measured rings; juvenile rings are often wider, so the count may be overestimated. Treat it as an estimate with that stated assumption.
• Single-linkage grouping can chain two batches through one long timber that overlaps both; inspect the matrix and raise the thresholds if a group spans more than one felling phase.
• The combined date treats proxies as independent and the ring-width proxy as a binary admissible set; it does not weight positions by t-value.
• Lens correction models radial distortion only (one k₁, centred) and is estimated from a single ruler; perspective is corrected only in Field mode. Use an orthogonal, flat-field photograph with the scale in the specimen plane. Multi-ray flags, EW/LW segmentation and enhancement are validated on synthetic photographs only.
• On real ITRDB series the probable tier is reached by about 7% of null (time-reversed) scans; only secure showed no wrong dates. See section 14.
• Published chronologies (.crn, TRiDaS derivedSeries) carry index and sample depth but no inter-series r̄. Their EPS cannot be verified, so every year is flagged exploratory; prefer the raw measurements where available.
• Negative years in Tucson files are taken as written and treated as astronomical years (0 = 1 BCE). Some laboratories write historical BCE years without a year 0; check the source documentation for BCE chronologies, as a one-year offset is possible. TRiDaS years with a BC suffix are converted explicitly (160 BC → −159).
• Private-vault match receipts commit to the reference values with a salted SHA-256 hash; they prove which reference produced a result once it is disclosed, but they do not let a third party re-run the match without it.
• Micro-CT virtual cores follow rings in one axial slice (optionally slab-averaged); strongly spiral grain or rings that change shape along the stem need rays on several slices. CT accuracy is validated on synthetic volumes only.
• Statistical significance supports, but does not replace, visual verification and analyst judgement.

14. Validation evidence

The following properties are checked by automated tests on every change to the computing code. A registry check fails if any listed claim loses its test.

The Validation lab runs the same evidence interactively: challenge series with injected faults (missing ring, false ring, decorrelated segment, frost year) and head-to-head audits against dplR 1.7.8 on real Rostrevor oak, each with its numerical divergence and declared tolerance (exact parity 1e-12, COFECHA segment correlation median |Δr| ≤ 0.1, growth curves 1% with a 5% optimiser band reported as a warning). It exports a validation certificate and an R / dplR script that re-computes every reference value.

Executable peer review (R / dplR)

Cross-dating, Quality control and the Dossier export a standalone R script. It embeds the ring widths (as a Tucson file read with read.rwl and as exact values), the master with its depth, and the standardisation parameters. Part A re-implements t_BP, t_H, Glk, r and p verbatim in base R and stops unless every value matches the app within 0.000001 (relative). Part B runs dplR's rwl.stats, caps/detrend.series, corr.rwl.seg, ccf.series.rwl and glk for comparison and prints how far dplR's own growth curves differ from the app's.

Part C of the same script asserts that dplR's sens1, sens2 and tbrm reproduce the app's values for the specimen. The Reference library's chronology builder exports a separate chronology parity script: it embeds every member series exactly, rebuilds the chronology with dplR (chron with the app's index for the biweight / arithmetic options, ssf for signal-free), checks the iteration count, the SSS formula and mean sensitivity, and stops on any mismatch (tolerance 1e-9, signal-free 1e-7). The automated suite compares tbrm, chron, sens1, sens2 and ssf with dplR 1.7.8 output stored for real Rostrevor oak (14 series).

Checked during development with R 4.5.3 and dplR 1.7.8 for all four standardisation methods with and without AR(1): Part A agreed to better than 10⁻¹¹; the chronology parity scripts agreed to 10⁻¹⁵ (biweight, mean) and 7·10⁻¹⁰ (signal-free, same 11 iterations as dplR); the smoothing spline agreed with dplR's to 10⁻¹³; dplR's nonlinear negative-exponential and Hugershoff fits differed from the app's by up to 4% of the curve, changing t by under 0.01. The R run is not part of the automated test suite; the suite checks that the script embeds the app's exact values.

15. Real-data benchmark

16. References

1. Baillie, M.G.L. & Pilcher, J.R. (1973) A simple crossdating program for tree-ring research. Tree-Ring Bulletin 33: 7–14.
2. Becker, M. (1989) The role of climate on present and past vitality of silver fir forests in the Vosges mountains of northeastern France. Canadian Journal of Forest Research 19, 1110–1117.
3. Benjamini, Y. & Hochberg, Y. (1995) Controlling the false discovery rate. Journal of the Royal Statistical Society B 57: 289–300.
4. Bigün, J. & Granlund, G.H. (1987) Optimal orientation detection of linear symmetry. Proceedings of the First International Conference on Computer Vision, London: 433–438.
5. Biondi, F. & Qeadan, F. (2008) Inequality in paleorecords. Ecology 89: 1056–1067.
6. Björklund, J.A. et al. (2014) Blue intensity and density from northern Fennoscandian tree rings, exploring the potential to improve summer temperature reconstructions with earlywood information. Climate of the Past 10, 877–885.
7. Box, G.E.P. & Jenkins, G.M. (1976) Time Series Analysis: Forecasting and Control. Holden-Day, San Francisco.
8. Brehm, N. et al. (2022) Tree-rings reveal two strong solar proton events in 7176 and 5259 BCE. Nature Communications 13: 1196.
9. Bridge, M. (2012) Locating the origins of wood resources: a review of dendroprovenancing. Journal of Archaeological Science 39, 2828–2834.
10. Briffa, K.R., Jones, P.D., Schweingruber, F.H. & Osborn, T.J. (1998) Influence of volcanic eruptions on Northern Hemisphere summer temperature over the past 600 years. Nature 393, 450–455.
11. Bronk Ramsey, C. (2009) Dealing with outliers and offsets in radiocarbon dating. Radiocarbon 51, 1023–1045.
12. Bronk Ramsey, C., van der Plicht, J. & Weninger, B. (2001) 'Wiggle matching' radiocarbon dates. Radiocarbon 43: 381–389.
13. Brown, D.C. (1966) Decentering distortion of lenses. Photogrammetric Engineering 32: 444–462.
14. Bunn, A.G. (2008) A dendrochronology program library in R (dplR). Dendrochronologia 26, 115–124.
15. Bunn, A.G. (2008) A dendrochronology program library in R (dplR). Dendrochronologia 26: 115–124.
16. Büntgen, U. et al. (2016) Cooling and societal change during the Late Antique Little Ice Age from 536 to around 660 AD. Nature Geoscience 9, 231–236.
17. Büntgen, U. et al. (2018) Tree rings reveal globally coherent signature of cosmogenic radiocarbon events in 774 and 993 CE. Nature Communications 9: 3605.
18. Chamberlain, R.G. & Duquette, W.H. (2007) Some algorithms for polygons on a sphere. JPL Publication 07-03, NASA Jet Propulsion Laboratory.
19. Commission Implementing Regulation (EU) 2025/1093 of 22 May 2025 laying down rules for the application of Regulation (EU) 2023/1115 as regards a list of countries that present a low or high risk. OJ L, 2025/1093.
20. Cook, E.R. & Peters, K. (1981) The smoothing spline: a new approach to standardizing forest interior tree-ring width series for dendroclimatic studies. Tree-Ring Bulletin 41, 45–53.
21. Cropper, J.P. (1979) Tree-ring skeleton plotting by computer. Tree-Ring Bulletin 39, 47–59.
22. Daly, A. (2007) Timber, trade and tree-rings: a dendrochronological analysis of structural oak timber in Northern Europe, c. AD 1000 to c. AD 1650. PhD thesis, University of Southern Denmark.
23. Eckstein, D. & Bauch, J. (1969) Beitrag zur Rationalisierung eines dendrochronologischen Verfahrens und zur Analyse seiner Aussagesicherheit. Forstwissenschaftliches Centralblatt 88: 230–250.
24. English Heritage (2004) Dendrochronology: guidelines on producing and interpreting dendrochronological dates. London: English Heritage.
25. Fritts, H.C., Mosimann, J.E. & Bottorff, C.P. (1969) A revised computer program for standardizing tree-ring series. Tree-Ring Bulletin 29, 15–20.
26. Gelman, A. & Rubin, D.B. (1992) Inference from iterative simulation using multiple sequences. Statistical Science 7, 457–472.
27. Grissino-Mayer, H.D. (2001) Evaluating crossdating accuracy: a manual and tutorial for the computer program COFECHA. Tree-Ring Research 57: 205–221.
28. Hartley, R. & Zisserman, A. (2003) Multiple View Geometry in Computer Vision, 2nd ed. Cambridge University Press.
29. Hollstein, E. (1980) Mitteleuropäische Eichenchronologie. Mainz: von Zabern.
30. Holmes, R.L. (1983) Computer-assisted quality control in tree-ring dating and measurement. Tree-Ring Bulletin 43: 69–78.
31. IAWA Committee (1989) IAWA list of microscopic features for hardwood identification. IAWA Bulletin n.s. 10: 219–332.
32. Jansma, E. et al. (2010) TRiDaS — the Tree-Ring Data Standard. Dendrochronologia 28: 99–130.
33. Koubaa, A., Zhang, S.Y.T. & Makni, S. (2002) Defining the transition from earlywood to latewood in black spruce based on intra-ring wood density profiles from X-ray densitometry. Annals of Forest Science 59: 511–518.
34. McCarroll, D. et al. (2002) Blue reflectance provides a surrogate for latewood density of high-latitude pine tree rings. Arctic, Antarctic, and Alpine Research 34, 450–453.
35. Melvin, T.M. & Briffa, K.R. (2008) A "signal-free" approach to dendroclimatic standardisation. Dendrochronologia 26: 71–86.
36. Miyake, F. et al. (2012) A signature of cosmic-ray increase in AD 774–775 from tree rings in Japan. Nature 486: 240–242.
37. Miyake, F., Nagaya, K., Masuda, K. & Nakamura, T. (2012) A signature of cosmic-ray increase in AD 774–775 from tree rings in Japan. Nature 486, 240–242.
38. Mosteller, F. & Tukey, J.W. (1977) Data Analysis and Regression. Addison-Wesley (biweight).
39. Neuwirth, B., Schweingruber, F.H. & Winiger, M. (2007) Spatial patterns of central European pointer years from 1901 to 1971. Dendrochronologia 24, 79–89.
40. Pech-Pacheco, J.L. et al. (2000) Diatom autofocusing in brightfield microscopy: a comparative study. Proc. 15th ICPR 3: 314–317.
41. Regulation (EU) 2023/1115 of the European Parliament and of the Council of 31 May 2023 on the making available on the Union market and the export from the Union of certain commodities and products associated with deforestation and forest degradation. OJ L 150, 9.6.2023.
42. Regulation (EU) 2025/2650 of the European Parliament and of the Council amending Regulation (EU) 2023/1115 as regards dates of application and simplification. OJ L, 2025/2650.
43. Reimer, P.J. et al. (2020) The IntCal20 Northern Hemisphere radiocarbon age calibration curve (0–55 cal kBP). Radiocarbon 62: 725–757.
44. Reinsch, C.H. (1967) Smoothing by spline functions. Numerische Mathematik 10, 177–183.
45. Ruffinatto, F. & Crivellaro, A. (2019) Atlas of Macroscopic Wood Identification. Cham: Springer.
46. Rydval, M. et al. (2014) Blue intensity for dendroclimatology: should we have the blues? Experiments from Scotland. Dendrochronologia 32, 191–204.
47. Schweingruber, F.H., Eckstein, D., Serre-Bachet, F. & Bräker, O.U. (1990) Identification, presentation and interpretation of event years and pointer years in dendrochronology. Dendrochronologia 8, 9–38.
48. Stokes, M.A. & Smiley, T.L. (1968) An Introduction to Tree-Ring Dating. Chicago: University of Chicago Press.
49. Stuiver, M. & Polach, H.A. (1977) Discussion: reporting of ¹⁴C data. Radiocarbon 19, 355–363.
50. Van den Bulcke, J., Wernersson, E.L.G., Dierick, M., Van Loo, D., Masschaele, B., Brabant, L., Boone, M.N., Van Hoorebeke, L., Haneca, K., Brun, A., Luengo Hendriks, C.L. & Van Acker, J. (2014) 3D tree-ring analysis using helical X-ray tomography. Dendrochronologia 32: 39–46.
51. Ward, G.K. & Wilson, S.R. (1978) Procedures for comparing and combining radiocarbon age determinations. Archaeometry 20: 19–31.
52. Warren, W.G. (1980) On removing the growth trend from dendrochronological data. Tree-Ring Bulletin 40, 35–44.
53. Ważny, T. (2002) Baltic timber in Western Europe — an exciting dendrochronological question. Dendrochronologia 20, 313–320.
54. Wetter, O. et al. (2014) The year-long unprecedented European heat and drought of 1540 — a worst case. Climatic Change 125, 349–363.
55. Whitlock, M.C. (2005) Combining probability from independent tests: the weighted Z-method is superior to Fisher's approach. Journal of Evolutionary Biology 18, 1368–1373.
56. Wigley, T.M.L., Briffa, K.R. & Jones, P.D. (1984) On the average value of correlated time series. Journal of Climate and Applied Meteorology 23: 201–213.
57. Yuksel, C., Schaefer, S. & Keyser, J. (2011) Parameterization and applications of Catmull–Rom curves. Computer-Aided Design 43: 747–755.
58. Zuiderveld, K. (1994) Contrast limited adaptive histogram equalization. In: Graphics Gems IV, 474–485. Academic Press.

17. Version history

Dossiers state the method version they were produced under. To reproduce an earlier report, use the TRiDaS export together with the method version recorded on it. Back to the specimen dossier.