Method

Research design

ArchaeoLocus distinguishes discovery from testing. Exploratory work may identify candidate distances, orientations, triangles, corridors or larger configurations. Once a candidate is promoted to a formal test, the relevant point set, roles, thresholds and comparison family are frozen before the validating computation wherever the design permits.

A test frozen after a pattern has already been seen is described as post-discovery declared validation, not as an independent confirmation. This distinction is reported because it materially affects the strength of the inference.

Coordinates, identity and provenance

Every point is an archaeological claim as well as a coordinate. The preferred record therefore identifies the physical feature represented, its source, coordinate scope, uncertainty and any approved alternative representative. A city centroid, sanctuary centroid, temple footprint, survey point and modern map marker are not treated as interchangeable merely because they share a place-name.

Duplicate records and nearby records belonging to one archaeological complex are audited before statistical counting. Where a location remains uncertain, that uncertainty is retained in the evidence record rather than hidden by selecting the coordinate that best preserves a pattern.

Geodesic measurement

Distances, initial bearings, reverse bearings, cross-track offsets and along-track positions are calculated geodesically rather than from an uncalibrated map image. The computational convention is recorded for each run.

Current analyses use WGS84 geodesic calculations where implemented. Earlier frozen studies may use a stated spherical or Haversine convention. Those results are preserved under the method actually used and are not silently recomputed with a later engine. Any methodological upgrade is versioned and disclosed as a new analysis or sensitivity check.

Geometric definitions

The geometry is converted into quantities that can be audited. Depending on the project, these may include equal-leg error in isosceles structures, proportional distance families, line and finite-corridor offsets, near-parallel or near-perpendicular axes, midpoint fractions, orientation residuals and multi-condition configuration scores.

Related quantities are not automatically counted as separate evidence. For example, a near-midpoint relationship and an approximate two-to-one distance relation may describe the same underlying geometry. Where measures are mathematically dependent, the project attempts to remove or label that dependence.

Tolerances and thresholds

A spatial relationship is never described as exact merely because it looks exact at map scale. Numerical departures and the threshold used to accept them are reported. Strict and sensitivity tolerances are kept distinct.

Thresholds should be declared before the decisive run. If a tolerance is widened after inspecting an outcome, that result is labelled exploratory or sensitivity-only and does not inherit the inferential status of a predeclared test.

Null models and statistical testing

The relevant question is not simply whether a pattern is visually unusual, but how often an equal-or-better result would arise under an appropriate comparison model. ArchaeoLocus therefore uses Monte Carlo simulation, role permutations, orientation randomisation, geography-aware spatial nulls and matched comparison sets according to the research question.

No single null model is assumed to answer every question. Where reasonable alternatives exist, results are compared across them and model sensitivity is reported. A small probability under one unrealistic null is not treated as robust evidence if a more appropriate null reproduces the result readily.

Multiplicity and the look-elsewhere problem

Large archaeological datasets offer many possible points, lines, triplets, distance bands, deities and tolerances. This creates a substantial opportunity to find attractive patterns by chance. ArchaeoLocus therefore records the size of the tested family wherever possible and uses false-discovery-rate correction, family-wise searches or explicit search-budget sensitivity when appropriate.

Raw p-values are not promoted as confirmed findings when they fail the declared multiple-testing correction.

Robustness and adverse tests

A result is more informative when it survives reasonable ways of trying to break it. Depending on the study, robustness checks include coordinate jitter, approved alternative representatives, leave-one-source or leave-one-target analyses, anchor holdouts, replacement tests, matched other-cult controls and sensitivity to the assumed comparison population.

Adverse results are retained. If an anchor is replaceable, a control is infeasible, a subgroup drives the apparent excess or an uncertainty model destroys the result, that information belongs in the conclusion rather than in a discarded worksheet.

Evidence hierarchy

ArchaeoLocus separates different evidential layers rather than combining every attractive observation into one claim.

1. Primary tested result: the predeclared or frozen statistic and its stated controls.
2. Secondary measured geometry: reproducible distances, bearings or geometric relationships that are relevant but not necessarily independent components of the primary statistic.
3. Exploratory observations: post-hoc correspondences retained as possible future hypotheses, not as additions to the original p-value or score.

Interpretation boundary

Historical explanation requires separate archaeological evidence: chronology, cult association, monument phase, topography, visibility, movement, political geography, textual evidence and a plausible mechanism by which a relationship could have been conceived or transmitted.

Negative results

A statistically negative result is not treated as a failed project. It establishes a boundary on what the present data support. The theatre-orientation programme, for example, did not demonstrate a general system of theatres intentionally facing other theatres. That negative result makes the retained Epidaurus–Corinth–Delphi case more credible precisely because it is not presented as evidence for a universal network.

Reproducibility, versioning and disclosure

Publication-grade runs are intended to preserve the source data or its documented provenance, exact settings, software version, random seed, manifests, machine-readable tables, maps and statistical outputs. Corrections are versioned rather than silently overwriting the evidential record.

Where code, data or manuscript files can be released, ArchaeoLocus aims to provide repository links and stable citation information. Third-party data restrictions and archaeological sensitivity are respected.

Computational and writing tools are disclosed where relevant to a formal publication. AI assistance does not replace author responsibility for research design, source selection, interpretation or verification.

Archaeological follow-up

Computation identifies where closer archaeological work may be justified. The next stage can include coordinate audit, monument-plan comparison, chronology review, cult and textual research, terrain or visibility modelling and independently assembled replication datasets.

The purpose of the method is therefore not to convert geometry into certainty. It is to make spatial claims sufficiently explicit that they can be reproduced, criticised, falsified and, where warranted, investigated historically.

ArchaeoLocus Method · website version v001 · 17 August 2026