# Assessment and execution specification

17 October 2026 · First-release assessment, retained for provenance

The subsequent expanded edition contains 38 publication-checked object dossiers and a direct atlas evidence inspector. See [the expanded audit](expanded-audit.md) for current scope, corrections, and verification. The twelve-case counts below describe the first release.

## Intent, feasibility, and logic

The request seeks a tool for challenging road reconstructions through archaeological and epigraphic evidence. This is feasible as an interactive evidence atlas. Its value depends on preserving the difference between a drawn route, a catalogue location, a reported findspot, an original emplacement, and a repository. An attractive overlay without those distinctions would reproduce the problem.

Two premises need qualification. Itiner-e already supplies certainty classes and segment bibliographies, and its reconstruction can incorporate milestones. EDH also supplies findspot and repository fields, although a local derivative may contain incomplete fields or imprecise coordinates. The problem is therefore uneven, difficult-to-inspect support and heterogeneous metadata. A stone used to reconstruct a route cannot independently corroborate that same reconstruction. See the primary documentation and exact references in [the methods dossier](methods-basemaps.md).

Empire-wide visualization and metadata admission are feasible now. Individually collating every milestone against CIL, subsequent editions, excavation reports, and object histories would be a sustained research programme. The implemented scope combines the full available transport snapshot, a transparent bulk milestone selector, and twelve substantially checked biographies. The bulk import is not an individually audited physical-object census.

The reasoning must also distinguish inscription date from road use or construction date; an imperial honorand from a builder or workforce; stone erection from new-road construction; physical defacement from its proposed political explanation; and an accurately documented secondary discovery from original roadside emplacement. Unlocated evidence belongs in the dossier, not at an invented point.

## Improved prompt

> Build a reproducible Roman Roads Evidence Atlas using the  Itiner-e export and an explicitly defined milestone corpus. Audit source formats, coordinate systems, transport types, dates, duplicate identifiers, location semantics, and licences before drawing conclusions. Preserve source identifiers, original geometry, raw date values, bibliography, and uncertainty labels. Do not alter the imported originals.
>
> Admit all available land-road segments, preserving Itiner-e certainty classes. Keep rivers and sea lanes in a separate optional context layer and exclude them from milestone-to-road calculations. Label derived segment endpoints as computational geometry, without implying ancient settlements or independently verified junctions.
>
> Import records explicitly categorized as milestones in the available digital epigraphy. Remove only demonstrated exact duplicate rows; preserve catalogue identity and distinguish record totals from counts of physical objects. Enrich through explicit cross-identifiers without adding duplicate stones. Treat unverified database coordinates as catalogue locality anchors. Keep findspot assertions, secondary-use contexts, reported localities, and repository institutions separate, each with a source and precision note. Unknown locations remain unmapped and searchable.
>
> Research a bounded set of twelve cases across different regions and periods, using accessible CIL material, regional corpora, institutional editions, and milestone scholarship, both online and . For every case, assemble dated inscription phases where recoverable, exact corpus concordances, short textual evidence and a labelled translation, findspot and custody history, attested work type, actors and their roles, alterations, erasure, reuse, movement, and explicit limits. Distinguish a directly consulted CIL entry from an inherited concordance or a consulted index. Do not invent missing readings, coordinates, agents, physical road fabric, or work events. Provide a claim/source ledger and exact publication locators.
>
> Implement a usable HTML GIS atlas with switchable Esri National Geographic, Topographic, Physical, and DARE basemaps. Display attribution and service limits; retain a plain map when tiles fail. Separate layers for confidence classes, catalogue localities, published findspot localities, repository institution proxies, derived endpoints, and water transport. Provide text/ID search, date filtering with explicit treatment of incomplete dates, region presets, linked record inspection, a filterable object dossier, and downloadable data and provenance.
>
> Add an exploratory discrepancy tool reporting approximate unsnapped point-to-road distances and metadata date agreement, with source independence explicitly unassessed unless checked. Exclude repositories and water routes from the candidate comparison. Unknown positional error prevents a meaningful validation threshold; therefore do not assign a stone to a road or generate a support score. Treat absent records as a coverage limitation.
>
> Deliver the HTML build, dossier, bibliography, import manifest with hashes, schema, source-preserving scripts, audit report, and portable package. Verify actual data counts, duplicates, chronological anomalies, location roles, phase gaps, filters, linked selection, map switches, offline tile failure, exports, and desktop/mobile rendering. State the boundary between metadata checks, publication consultation, browser QA, private deployment, field verification, and independent scholarly review.

## Implemented scope

The local NDJSON contains 16,554 transport features: 15,196 land-road segments (5,929 main / 9,267 secondary), 834 rivers, and 524 sea lanes. Among land roads, source certainty is 1,082 certain / 13,309 conjectured / 805 hypothetical. All original WGS84 line coordinates are retained; an older EPSG:3395 GeoJSON was inspected but not used or mistaken for EPSG:3857. Numeric zero dates mean unspecified. Only 510 road lower dates and 214 upper dates are nonzero. Date-error zero means unspecified, not demonstrated precision.

The exact EDCS `miliaria` selector yields 8,489 source rows. Six exact duplicates are removed, leaving 8,483 unique EDCS identifiers. Of these, 7,810 have catalogue coordinates and 673 remain unmapped. Explicit partner identifiers enrich 2,140 records with EDH metadata. Cross-ID physical-object deduplication is unresolved; twelve EDH identifiers are shared by multiple EDCS IDs. One reversed source date interval is retained raw and withheld from analytical chronology.

Twelve biographies cover six British examples, Tre Fontane on the Via Traiana, Gerulata, Rabland, CIL II 4734–4735, and the La Rinconada collector-property recovery. Six RIB findspot-locality coordinates are admitted with unknown positional error; three repository institution proxies are separately sourced. The other six original-place coordinates remain null. A repository proxy establishes an institution's position, not current object presence or display.

CIL II 4734–4735 were directly consulted on printed p. 632 /  of the local 1869 edition. The official CIL XVII/4.1 index was directly consulted for Rabland; the full object entry was not. Remaining CIL references are explicitly publication-reported concordances. The dossier preserves the CIL/modern reading disagreement for 4734 and corrects an OCR error in the Tre Fontane numeral after inspecting the printed page.

The comparison is a review aid. No stone-to-road assignments, new route geometry, fabricated uncertainty radii, or road validation claims are produced. The bulk catalogue is searchable and source-linked; work type and actor roles are individually interpreted only in the twelve curated cases. Source-overlap classification remains unassessed rather than an automated bibliographic assertion.

See [the import manifest](../build.json), [local source inventory](../build.json), [publication dossier](milestone-research.md), [claim ledger](claim-ledger.jsonl), and [basemap verification](basemaps.json).