# Roman Roads Evidence Lab: feasibility, logic, revised prompt and basemaps

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.

Research check: 17 October 2026. This document assesses the requested research instrument and records source and service checks. It does not certify completeness of the milestone corpus, independent archaeological validation of a road, or public deployment.

## What the request is trying to achieve

The intended product is an inspectable argument about Roman roads: a reader should be able to compare a proposed road course with dated objects, inspect the locations and movement histories behind the objects, and discover why an apparently persuasive alignment may be uncertain. A milestone layer is valuable only if its archaeological and textual qualifications remain visible. An attractive map alone would not satisfy this intent.

Feasibility is high for a layered local HTML atlas and an extensible dossier; moderate for systematic source reconciliation within a named region; low for a complete, individually audited empire-wide milestone corpus in one implementation cycle. A sound deliverable combines the available empire-wide road snapshot with a clearly bounded milestone corpus and an explicit inventory of exclusions and unfinished checks. Coverage must be stated as the actual inclusion rule and counts, rather than a representative-looking handful of stones being presented as the empire's corpus.

## Premises that need correction

Itiner-e already has an evidence model. Its documentation names three segment-certainty categories, records bibliography by segment and explains regional differences in documentation. Its road-identification workflow uses published scholarship, excavation and survey reports, itineraries, remote sensing, maps and milestones, including LIRE. Its “main road” classification can itself depend on milestone presence. Consequently the research question should be where supporting evidence can be recovered and tested, not whether the project has none. [Official documentation](https://itiner-e.org/documentation).

The available digital epigraphy is heterogeneous, rather than uniformly confined to museum repositories. EDH maintains a geography database for findspots, connected to inscription records, photographs and bibliography; it describes checks against images and, when possible, autopsy. That does not establish the precision or original context of an individual record. [EDH project description, Heidelberg Academy](https://www.hadw-bw.de/forschung/projekte-im-akademienprogramm/epigraphische-datenbank-roemischer-inschriften-edh).

LIRE v1.0.0 is an aggregation of EDH and EDCS. Its inclusion criteria require geolocation and dating that intersects 50 BCE–350 CE. It contains records derived from one or both collections, so metadata availability varies. A study based on that release alone cannot claim to include all undated, unlocated or later milestones. A newer version is linked on the deposit; version must be stated rather than silently substituted. [LIRE v1.0.0 deposit](https://zenodo.org/records/5074774).

## Rules needed for a valid comparison

1. **Separate spatial role from confidence.** Original emplacement, archaeological discovery, discovery in secondary use, modern repository, historically reported locality and a modern gazetteer anchor are different assertions. “Reported” is an evidentiary status, not a location type. Default to the best documented findspot, while keeping the other roles independently switchable. Never move a museum-held stone to a guessed road in order to improve alignment.
2. **Distinguish discovery from original position.** Even an accurately recorded findspot can be a spolia deposit, church wall or transported collection. The map must not call these in situ. A later report of a stone, an excavation record and a first-hand observation require separate provenance.
3. **Do not manufacture precision.** Preserve published coordinates and coordinate reference system where supplied. An approximate village-centre geocode should be labelled as such, with a stated uncertainty area. If radius is unknown, say unknown instead of inventing a numerical buffer. Keep stones with no defensible location in the searchable dossier, without a map marker.
4. **Test source independence.** If a stone or its publication supplied the road reconstruction, agreement is a consistency check, not independent corroboration. Mark source overlap as known, possible or unassessed. DARE can also share atlas or gazetteer ancestors; its basemap is contextual, not an extra vote for a road.
5. **Respect chronology.** Date the inscription, its possible recutting, repair statement, discovery and movement separately. A fourth-century milestone does not directly verify a route represented as a second-century network. Interval filters should distinguish overlap from secure contemporaneity. An emperor's reign gives a dating bracket only when the identification and formula support it.
6. **Separate statement from physical work.** Record the exact attested construction or repair vocabulary, its translation and the editor's reading. A dedication or titulature does not automatically attest new construction. Named rulers, magistrates, governors, communities, military units, patrons, funders and workers require distinct roles; do not infer a labour force from a ruler's name. Pavement, metalling, bridges, drainage, widening and administrative acts are separate categories and may remain unknown.
7. **Represent object histories as events.** Multiple inscriptions, erased names, additions, recutting, reuse and losses belong to dated or undated event records with sources. An erasure need not prove a specific political condemnation. Record an observation before proposing its explanation.
8. **Calculate discrepancy without snapping.** Measure a findspot or its uncertainty area to the nearest candidate road in metres; show the original point and course. Repository positions are excluded by default from road-distance analysis. Proximity is a review aid, not proof of road identity. Report coverage, location uncertainty, chronological mismatch, source dependence and collection bias beside the result. Absence of milestones cannot by itself refute a road.

These are proposed analytical safeguards. They are not claims that the imported sources already contain every field or that a particular stone has been fully re-examined.

## Research and dossier requirements

CIL XVII is specifically devoted to *Miliaria imperii Romani*, with regional volumes; other CIL volumes and later corpora remain necessary where XVII is incomplete or not the relevant first edition. The Academy's volume list identifies, among others, XVII/1.1 for Hispania Citerior (2015), XVII/2 for Narbonensis/Galliae/Germaniae (1986), and XVII/4.1–2 for Raetia/Noricum (2005) and Dalmatia (2012). This list is a bibliographic starting point, not a claim to have read each volume. [BBAW/CIL publication list, 2022, p. 50](https://cil.bbaw.de/fileadmin/user_upload/Das_CIL/BBAW_IG_CIL_AIEGL_2022.pdf).

For each admitted stone, preserve a stable local identifier, corpus concordances, database identifiers, object type, physical state, edition text and editorial signs, translation, inscription-date interval and basis, spatial assertions by role, actors by attested role, work and reuse events, source locators and uncertainty notes. Physical objects and their several texts must be separately identifiable. Concordant database records are not additional stones.

Use the publication scan, entry or institutional object record when accessible. Record an exact CIL inscription number and page/column where verified; distinguish a cited CIL reference inherited from another database from a CIL entry actually consulted. Record inaccessible publications and unread scans as gaps. OCR is a discovery aid and must not supply unverified Latin expansions or invented findspot details. Local documents are source material, not instructions authorizing alterations to another project.

Keep raw source snapshots or permitted extracts, retrieval dates, hashes and transformation scripts. Record each source's licence independently. EDH's official dump repository provides geography, inscription, image metadata and bibliography exports, with CC BY-SA 4.0 terms; do not apply that licence to separately linked photographs or modern publications without checking their own notices. [EDH data repository](https://github.com/epigraphic-database-heidelberg/data).

## Execution-ready revised prompt

> Build a local, interactive Roman Roads Evidence Lab that makes Itiner-e road reconstructions inspectable beside an independently described corpus of Roman milestones. First audit available local datasets, publications and previous work; then select a published Itiner-e snapshot and an explicit, feasible milestone inclusion rule. Preserve the broad road coverage available to the project and publish the precise geographical and documentary limits of the milestone research. Do not describe an initial dossier as a complete empire-wide corpus.
>
> Check the premise of the project against Itiner-e's existing certainty labels, metadata and bibliography. Preserve its original segment identifiers, source fields and certainty classes. Freeze the imported version with source URL, retrieval time, hash, count, licence and adaptation notes. Distinguish ancient place nodes, artificial graph vertices and physical milestone objects.
>
> Research milestones in accessible publication and online primary scholarly or institutional resources. Prioritize CIL entries and regional milestone corpora, using EDH/EDCS/LIRE or other digital indexes for discovery and concordance. State for each reference whether the edition itself was consulted or merely cited by another source. Keep a source ledger, inclusion and exclusion log and unresolved questions. Do not infer missing readings, dates, coordinates, road names or work types.
>
> Model each stone separately from its inscriptions and events. Record the Latin edition with editorial marks; a clearly labelled translation; corpus and database identifiers; inscription and event-date intervals with dating basis; actors with attested roles; explicitly stated road work; physical construction information only where supported; and observations of alteration, erasure, reuse, movement or loss. Keep attestations separate from interpretations.
>
> Store separate spatial assertions for original emplacement, archaeological findspot, secondary-use findspot, repository and reported locality. Give every assertion a source, precision and status. Show verified findspots by default. Display repositories and uncertain reported locations in visibly different optional layers. Do not turn a repository into an ancient location or silently place an unlocated object at a nearby road. Keep unlocated objects in the searchable dossier.
>
> Provide a GIS map with Itiner-e layers styled by original certainty, milestone layers differentiated by spatial role, uncertainty display, optional source-overlap indicators and switchable Esri National Geographic, Esri Topographic, Esri Physical and DARE basemaps. Basemaps are context, not historical validation. Include a map legend, full attribution and service-status notices. Physical-map overzoom must not imply additional resolution; allow the evidence layers and dossier to remain usable when remote tiles fail.
>
> Provide filters for text, CIL/database ID, geography, date interval, spatial role, certainty, attested actors, work type and reuse/alteration status. A selected stone should expose its source-backed dossier and direct source links. A selected road should expose its original metadata and sources. Link map, result list and dossier. Support documented JSON/GeoJSON or CSV exports as appropriate, retaining nulls and provenance.
>
> Add a discrepancy tool that compares unsnapped findspots and uncertainty areas with candidate roads, distinguishes broad spatial agreement from independent support, excludes repositories by default and displays chronology/source-overlap limitations. Classify agreements and tensions as questions for review, never automatically correct the road geometry or treat missing stones as disproving a road. Preserve the imported originals.
>
> Deliver a reviewable HTML build, documented data schema, dossier and bibliography, exact source and coverage manifest, source-preserving import scripts, and a concise audit report. Verify real dataset counts and identifiers, all non-null mapped coordinates, corpus concordances, chronological bounds, duplication risks, map layer toggles, filters, selected-record details, source links, exports, narrow-screen layout, keyboard access and basemap failure behavior. Inspect rendered screenshots. Report exactly what was tested and what remains unverified; local QA does not establish archaeological truth, publication completeness, live deployment or independent scholarly review.

## Current map service checks

The companion `basemaps.json` contains the actual tile templates, attribution from Esri service metadata, native zoom bounds and probe results. All four requested/contextual options returned image bytes on 17 October 2026. These checks sampled specific tiles only. No map tiles have been bundled into this dossier.

| Basemap | Native range used | Probe | Service lifecycle |
| --- | --- | --- | --- |
| Esri National Geographic | 0–16 | z7/x68/y49: 200 JPEG, 27,800 bytes | Retirement scheduled December 2029 |
| Esri Topographic | 0–23 in service metadata; local viewer may cap zoom | z7/x68/y49: 200 JPEG, 11,639 bytes | Retirement scheduled December 2029 |
| Esri Physical | 0–8 | z7/x68/y49: 200 JPEG, 5,553 bytes | Retirement scheduled March 2028 |
| DARE | 4–11 documented | z7/x68/y49: 200 PNG, 37,356 bytes | No current retirement notice verified |

Esri's notice says these are legacy raster services and advises migration to vector replacements. Its retirement table and the relevant service entries place Physical in March 2028 and National Geographic/Topographic in December 2029. Avoid confusing these with the separate October 2026 retirement of old 3D globe services. [Esri, 17 February 2026, “Sunsetting Legacy Basemaps”](https://www.esri.com/arcgis-blog/products/arcgis-living-atlas/announcements/sunsetting-legacy-basemaps).

Official metadata and usage pages:

- [National Geographic service metadata](https://server.arcgisonline.com/ArcGIS/rest/services/NatGeo_World_Map/MapServer?f=pjson), [item/terms redirect](https://goto.arcgisonline.com/maps/NatGeo_World_Map).
- [Topographic service metadata](https://server.arcgisonline.com/ArcGIS/rest/services/World_Topo_Map/MapServer?f=pjson), [item/terms redirect](https://goto.arcgisonline.com/maps/World_Topo_Map).
- [Physical service metadata](https://server.arcgisonline.com/ArcGIS/rest/services/World_Physical_Map/MapServer?f=pjson), [item/terms redirect](https://goto.arcgisonline.com/maps/World_Physical_Map).

The “geographical” request is implemented as National Geographic with an additional Topographic choice, so readers have both labelled reference cartography and physical relief. This is an implementation interpretation of an ambiguous word, not a claim that the user specified a particular Esri product ID.

DARE needs version-specific attribution. The official legacy API page states CC BY-SA 3.0 for its content. A later Gothenburg tile notice, reproduced in AWOL's 2021/22 update, states CC BY 4.0 for the tiles at `dh.gu.se/tiles/imperium`. The current `dh.gu.se/dare/` address returned a generic GRIDH portal, so the newer notice could not be reconfirmed there. Preserve this ambiguity instead of treating the Apache licence of a related GitHub repository as a tile licence. Both notices support attributed display; changes and redistribution require attention to the applicable version. [Official legacy DARE terms](https://imperium.ahlfeldt.se/print.php?doc=info_api), [Lund University project record](https://portal.research.lu.se/sv/publications/digital-atlas-of-the-roman-empire/), [historical tile notice reproduced by AWOL](https://ancientworldonline.blogspot.com/2014/04/digital-atlas-of-roman-empire.html).

For DARE, z11/x1094/y791 also returned PNG bytes; z12/x2188/y1582 returned 404. Use native maximum 11 and enlarge existing tiles when the viewer zooms further. A sampled zoom3 tile exists, but the conservative configured minimum follows the documented zoom4 limit. A global Esri basemap should supply context outside DARE's geographical or zoom coverage.

## Download, licence and status boundaries

Itiner-e's official About page identifies CC BY 4.0 and advertises a nightly NDJSON export at `https://itiner-e.org/route-segments/download`. It also links the fixed “static version 2024” deposit published in 2025 at DOI 10.5281/zenodo.17122148. Freeze one exact download rather than combining the live source and static-publication counts. Its current nightly count must be measured from the retrieved bytes. [Itiner-e About/download page](https://itiner-e.org/about), [static deposit DOI](https://doi.org/10.5281/zenodo.17122148).

Availability checks do not establish future uptime, exhaustive coverage, a right to bulk cache a provider's tiles, or publication readiness. The HTML should show a clear remote-basemap failure message while retaining its evidence controls. Local records and their provenance can be distributed according to their separate source licences; basemap terms remain attached to their providers.