> Historical 39-case release. Current counts and subsequent source corrections are in [primary-context-audit.md](primary-context-audit.md).

# Spatial evidence hierarchy release audit

17 October 2026. This supersedes the previous 38-case release for current scope; earlier release reports remain historical records.

## Feasibility and intent

Prioritising fixed rupestral mileage markers is feasible: continuous original bedrock can strongly constrain the local relation between a record and a surviving road. The atlas now separates monument form from the strength of that relation. Original installation, precise coordinates, date, milestone function and source independence are separate claims. The tiers are editorial review priorities, not probabilities or an automatic verdict on an Itiner-e polyline. A well-excavated primary milestone station can be as useful as a bedrock marker.

## Execution-ready improved prompt

Extend the existing Roman Roads Evidence Atlas with a publication-audited spatial evidence hierarchy. Create a separate rupestral-bedrock milestone facet. Admit T1 only when an inspected publication establishes Roman mileage-marker function, continuous original bedrock, and a direct local Roman-road relation, without an unresolved displacement contradiction. Create T2 for documented association with physical Roman-road remains, distinguishing T2a primary installation/socket from T2b qualified local association with original installation unproved. Explicit displaced spolia, secondary fill or transport cannot supply a primary road anchor. On-site later reuse must remain flagged and cannot prove an unchanged original station. Cascade through T3 published conditional road association, T4 specific discovery context, T5 broad locality, and T6 custody/proxy. Keep U unreviewed and X excluded/disputed visible. Require a rationale, exact publication locator, source IDs, limitations, approved historical-location assertions and assessment date for every reviewed case. Never promote a bulk record by material words, proximity, inferred mileage or an isolated in-situ field. Preserve chronological phases, movement history and source geometry. Add shared tier/form filters, source-linked case assessments and typed map markers; each map assertion shows its own tier so custody or approximate locators cannot impersonate the object's strongest ancient context. Retain unlocated high-tier evidence in the dossier. Test dates, filters, source access, marker selection and responsive layouts, and preserve the private audience.

## Reviewed admissions

| Class | Objects | Meaning |
|---|---:|---|
| T1 | 1 | Donnas XXXVI; same-bedrock mileage marker and road |
| T2a | 0 | No primary socket/installation admitted in this bounded corpus |
| T2b | 1 | Casa de Campo CIL II 4644; qualified publication-documented archaeological road association |
| T3 | 7 | Published conditional road/original-setting argument |
| T4 | 17 | Specific discovery context; original road setting unproved |
| T5 | 9 | Broad locality or corridor |
| T6 | 3 | Collection/custody/proxy only; ancient provenance unestablished |
| X | 1 | Cesiomaggiore retained as a contested comparator, excluded from alignment inference |
| U | 0 reviewed cases | Bulk catalogue records remain unreviewed; U is not a claim of poor ancient context |

The dossier now has 39 physical-object cases, 47 inscription phases, 74 separately retained object-history events, five competing date assignments, and 115 typed location assertions, 68 mapped. The claim/source ledger contains 475 entries, including 39 explicit editorial spatial assessments. The existing 8,483 catalogue IDs and 15,196 land-road segments are unchanged. Catalogue IDs are not a count of deduplicated physical stones.

Donnas's exact original marker coordinate is null. The official map coordinate locates the 221/222m complex and appears separately as T5, with an editorial 250m envelope rather than a surveyed positional error. The catalogue point remains an optional T6 proxy. Donnas's inscription phase is undated; a possible Augustan attribution appears separately and does not place it into the active chronology filter when unknown dates are excluded. CIL V 8074 is a scholarly concordance; its original CIL page was not newly collated. See [rupestral research](rupestral-research.md) and its machine-readable source register.

Casa de Campo is a borderline T2b admission, not a claim of an excavated original socket. Paredes Martín records fence-post reuse, short modern movement and a surviving excavated/restored Roman road at the same locality. The assessment records both the association and the unresolved original station. The 2020 observation/display coordinate is T4 and excluded from original-position distance comparisons. Full excavation/discovery records are still needed. [Paredes Martín 2023, p.41 / ](https://revistas.uned.es/index.php/ETFII/article/download/36712/28064#page=11). The earlier CIL entry is directly available through the existing public-domain page viewer; CIL alone does not supply the modern road-context argument.

Methodological controls retain relocated Tabula Traiana, disputed milestone function for CIL II 6211, unresolved classification/dating of Gospodjin Vir CIL III 13813d, and milestones reused in the Rantis roadway as research controls rather than invented new milestone admissions. [Feasibility and sources](hierarchy-feasibility.md), [criteria](hierarchy-criteria.json), [38-case reassessment](spatial-tier-audit.md).

## Implementation and verification

- Shared tier/form filters apply to atlas case pins, case search, checked locator, fitting and dossier. Chronology selects inscription phases separately from history events. Explicit selections outside filters remain readable with a warning.
- Pin labels describe the selected location assertion. T1/T2 object associations are not inherited by a museum, catalogue coordinate or approximate complex locator. Source buttons open the exact consultation record inside the atlas; existing CIL page images remain inspectable.
- The longer assessment is inside Object so Timeline, Locations and Sources remain readily accessible. Source-locator actions refresh the enabled-layer counts.
- 60 semantic/source-byte checks passed, including the four T1 gates, source-reference resolution, typed-location ceilings, chronology separation, unchanged original geometry and rejection of a failed bedrock-continuity gate.
- 50 browser interaction checks passed, including tier/form filtering, source access, undated exclusion, disputed comparators, date gaps, point selection, baseline basemaps, mobile overflow and failed-tile behaviour. After visual refinement, ten targeted desktop/mobile layout and locator checks passed. No runtime exceptions were observed.
- Both browser JavaScript files passed syntax checking. Desktop and mobile screenshots were inspected. These are local browser and data checks, not field verification, current custody confirmation, comprehensive CIL collation or independent archaeological certification.

Verification receipts are in `qa/data-audit.json`, `qa/browser-results.json`, and `qa/hierarchy-layout-results.json`. The private deployment receipt is `deployment.json` in the portable release. No source PDF subject to modern publication rights is bundled.