# Research edition audit

This document records the first release. See [the expanded audit](expanded-audit.md) for the current 38-object dossier, location corrections, direct atlas inspection, and verification.

17 October 2026

The research build passed **30 source/data admission checks** and **28 browser checks**, with no browser runtime exceptions. Desktop and narrow-screen screenshots were rendered and inspected. Detailed results are retained in `qa/data-audit.json` and `qa/browser-results.json` beside the project; the screenshot set is in `qa/`.

## Evidence covered

- Source byte identities were verified against SHA256 for the three imported local datasets. Source road geometry, identifiers, certainty labels, and bibliography were compared with the original NDJSON.
- All 16,554 source transport features are accounted for: 15,196 land roads and 1,358 water routes. Water routes never enter road-distance comparison or road-endpoint derivation.
- Exact EDCS duplicates reconcile 8,489 rows to 8,483 distinct catalogue IDs. The map admits 7,810 catalogue-locality coordinates, retaining 673 unlocated records. No museum substitution, geocoding, snapping, or silent relocation was used.
- All non-null coordinates are finite valid WGS84 coordinates. Bounds checks do not establish positional accuracy.
- Zero date placeholders, incomplete dates, a reversed source interval, and two sets of distinct inscription phases were checked. A phase gap is not rendered as continuous attestation.
- Twelve curated cases and sixty claim-ledger entries have sources, exact locators, and limits. Six source-provided findspot locality points and three separately sourced institution proxies remain role-separated. Catalogue-to-case joins use explicit identifiers.
- The browser checked catalogue search, selected-record inspection, approximate distance results, dossier filtering by work/region/alteration, repository marker selection, incomplete-date exclusion, invalid-range rejection, phase gaps, all four basemap options, plain-map mode, water-layer switching, JSON downloads, and narrow-screen overflow.
- A separate browser page blocked all HTTPS basemap requests: the bundled GIS still loaded and displayed an explicit tile-failure notice. Repositories do not expose a distance comparison action.
- Every static HTML file reference resolves locally. Modern publication PDFs and third-party map tiles are not bundled.

## Source consultation boundary

 sources include CIL II (1869), printed p. 632, entries 4734 and 4735; Kolb and Rathmann in *Roman Roads: New Evidence – New Perspectives* (2019); España Chamorro (2017); Campedelli; and Laing. The online dossier consulted RIB object editions and primary articles for the selected cases, plus the official CIL XVII/4.1 index. The full CIL XVII object entries were not comprehensively collated. Concordances inherited from another publication are marked accordingly.

The audit caught and corrected: inclusion of rivers/sea lanes among road candidates; six duplicate rows; the reversed EDCS interval; zero-error values misleadingly displayed as exact years; stale filtered search results; continuous envelopes across inscription-phase gaps; missing explicit RIB/EDCS case joins; an OCR error in the Tre Fontane numeral; and a map-script parameter initially mistaken for positional uncertainty. No numerical uncertainty radius was admitted from that parameter.

## What remains open

The old local EDCS/EDH snapshots are not a census of currently published milestones. Bulk actor/work/reuse interpretation is not supplied as individually checked scholarship. Cross-ID physical-object deduplication remains unresolved. Most catalogue coordinates have no documented error estimate or verified original emplacement. Current object presence at the repository institutions has not been confirmed. Source independence between each stone and each route remains unassessed. Source and tile licence version ambiguities are preserved in the inventory and methods dossier.

Browser and metadata checks establish a functioning research instrument. They do not establish road identity, archaeological contemporaneity, field verification, exhaustive literature coverage, or independent scholarly review. Hosted publication status is recorded separately in `deployment.json` when confirmed.