Benchmark exchange · method 1.0.0

The same case. Explicit comparison rules.

The exchange compares submitted predictions with fixed versions of published experimental cases. The initial cases are three nominally monotonic, solid CLT wall tests from one UNBC / Fast + Epp campaign. These published measurements are openly accessible, so every starter case is an open benchmark.

What is fixed

Each case preserves specimen and campaign metadata, source locators, reuse licences, force and displacement definitions, a fixed comparison domain, a processed comparison curve and its processing description. Case versions identify their frozen contents. Every submitted report keeps its exact case snapshot, prediction points, scoring method and content hashes.

How the comparison trace is prepared

The full measured actuator history remains available through the original source and the linked full-resolution CSV. The comparison trace starts from the atlas display reduction, which retains original measured samples and their peaks. The first-excursion rule then retains the first sample and every later sample whose displacement exceeds all previously retained displacements. This removes repeated excursions and sensor backsteps; it is not a complete measurement history.

Force and displacement pairs are not smoothed, shifted, sign-flipped or fitted to a prediction. The fixed domain starts at zero when the source trace brackets zero, otherwise at its first nonnegative displacement. A comparison boundary is interpolated linearly when needed. No extrapolation is used. Reports state raw, reduced and retained point counts and distinguish the full-history peak from the processed-curve peak.

Peak force

The measured reference peak comes from the complete original acquisition history. The predicted peak is the maximum absolute force in the submitted record. A partial record cannot establish that a model reached its full capacity; its coverage and limitations remain visible.

Stiffness

The existing curve method attempts a secant stiffness between 10% and 40% of ascending-branch peak load for explicitly monotonic records. It withholds a stiffness error when reversals, missing crossings or other ambiguity prevent a reliable value. An unavailable value remains unavailable. This diagnostic stiffness is not a material property or a design resistance.

Curve error and coverage

Piecewise-linear interpolation is integrated exactly over the union of curve knots. The prediction must cover the complete fixed displacement domain; otherwise full-domain RMSE is withheld. Coverage describes displacement span, not the number of points. All submitted coordinates are explicitly converted to mm and kN. The server recalculates the result from submitted points and never accepts client-supplied scores.

Calibration history and comparing methods

Contributors declare whether they used this case, a related campaign, neither, or have not disclosed the history. These are self-reported declarations. Public tables keep those groups and case versions separate. Results assess specific records and measurement definitions, with no global best-model ranking or implied independent validation.

Publication and contributor credit

An active academic account may submit predictions from any modelling software. Submissions are private until the owner reviews them and separately chooses to publish. Publication consent explicitly covers the prediction curve, model methods, software version and contributor credit. Account identifiers, email addresses and private review notes are excluded from public reports. Team and institution names are contributor supplied and do not establish institutional endorsement.

Program compatibility

The CSV interface accepts results from the researcher’s preferred software. LignoScope Research Studio is currently promoted as coming soon. An automatic LignoScope connection is planned and is not provided by the exchange release.

Future blind challenges

A genuine blind challenge needs unreleased measurements supplied by a partner laboratory and a controlled release process. Hiding measurements that are already public does not make a benchmark blind. No blind challenge is currently offered.

Browse the exchange · Submit an academic prediction · Inspect source specimens