# Catalogue falsa, identification and chronology audit · v8

17 October 2026. This release screens **8,483 imported catalogue records** and records **60 targeted review decisions on 55 records**. It applies **13 exclusions from ordinary milestone analysis**, including the existing Donnington exclusion, **38 caution classifications**, four retained controls and **26 chronology overlays**. It preserves all imported dates, texts and coordinates. These are bounded documentary judgments, not complete authentication of the catalogue.

## Coverage and evidence standards

All records were screened for broad source date intervals, invalid/unresolved components, questioned or different EDH monument types and starred CIL concordances. The screen identifies 401 records with one or more review flags: 238 broad source ranges, 11 unresolved dating cases, 39 questioned partner types, 31 different partner support types and 112 starred CIL concordances. Counts overlap. A broad range means at least 100 inclusive historical years in a source component, skipping year zero; it is a review priority, not evidence of an error.

The British pass inventories all 103 Britannia rows: 101 have explicit RIB citations, resolving to 107 successfully retrieved RIB entry pages; the two other rows were separately checked. The linked RIB pages yielded no additional Falsum title beyond Donnington. RIB's exclusion of the Vindolanda column supplies a separate misidentification decision. Retrieval identities, source locators and qualification controls are in [the RIB audit](catalogue-rib-audit-v8.md) and its JSON.

The CIL pass directly examined selected pages in the original CIL II (1869) and III (1873), selected modern CIL II² entries , and primary publications. It inventories 94 CIL XVII starred candidates without adjudicating all 94. The [CIL identity audit](catalogue-cil-identity-v8.md) records twelve decisions and exact local page identities. Starred numbering alone never causes exclusion: Sacquenay demonstrates why an older adverse attribution cannot replace later scholarship.

[The dating review](catalogue-dating-review-v8.md) records 27 targeted rows. Its 23 date overlays, Sacquenay's independently reviewed date and two conditional transmission overlays make the release's 26. A second pass independently checked the RIB proposals against retrieved corpus pages. The four additional roadwork/bridge decisions have exact sources in `catalogue-roadworks-review-v8.json`.

## Exclusions and their different meanings

| Catalogue record | Imported locality | Review classification |
| --- | --- | --- |
| EDCS-08701359 | La Carlota / Astigi | La Carlota: public roadwork pillar |
| EDCS-09700636 | Chesterholm / Vindolanda | Vindolanda: published as a milestone, reclassified as uninscribed column/pilaster |
| EDCS-44600074 | Donnington | Donnington: RIB 2392* Falsum; milestone identification unproved |
| EDCS-54900234 | Solin / Salona | Salona: road-building panels, not a mileage marker |
| EDCS-54900235 | Solin / Salona | Salona: second road-building inscription, not a milestone |
| EDCS-28600193 | Solin / Salona | Hippius bridge: building/honorific inscription |
| EDCS-29900424 | Alconchel de la Estrella | CIL II 464*: fabricated transmitted text |
| EDCS-05501843 | Chaves / Aquae Flaviae | Chaves: Padrão dos Povos and its copies |
| EDCS-29900452 | Mazarron / Almazarron / Municipium Ficariense / Carthago Nova | CIL II 467*: certainly interpolated transmission |
| EDCS-09100349 | Sagunt / Sagunto / Saguntum | Saguntum: reused column with honorific inscription |
| EDCS-67100113 | Vilches, iuxta / Baesucci | CIL II 466*: rejected corpus tradition |
| EDCS-29900423 | Villarejo de la Penuela | CIL II 465*: fabricated transmitted text |
| EDCS-26600085 | Sarikaya | Sarıkaya: rock-cut road restoration record |

CIL II p.45* supplies distinct editorial judgments: nos.464*–465* are rejected fabricated texts; no.467* is described as interpolated; no.466* is retained here as a corpus-rejected tradition, without claiming a modern autopsy or that an identified entire stone was manufactured. The original public-domain page can be inspected from the atlas's record audit. No supposed fabrication date or original roadside coordinate is inferred.

Other exclusions concern monument function rather than forgery. [RIB's Milestones introduction](https://romaninscriptionsofbritain.org/milestones), “Other Monuments Excluded from RIB,” identifies AE 1971,224 at Vindolanda as an uninscribed column or pilaster. Saguntum is a reused honorific support in the consulted CIL entry. La Carlota is a roadwork pillar. Chaves requires a distinction between the ancient column recovered in 1980, later bridge copies and differing interpretations of the named communities; its text date cannot automatically date the bridge.

The Salona entries record genuine construction panels, as discussed in [Glavaš 2016](https://hrcak.srce.hr/169743), pp.7–8. The Hippius record is a bridge-building/honorific inscription in [Šačić Beća 2020](https://godisnjak.anubih.ba/index.php/godisnjak/en/article/download/4/4/10), pp.77–78, no.3. Sarıkaya remains the separate R1 rock-cut restoration record, supported by CIL III346, p.61, and the [published bilingual transcription](https://epigraphy.packhum.org/text/277805). These objects remain evidence for public works; exclusion from the milestone layer is not a judgment that they are false inscriptions. Their raw catalogue rows remain accessible.

## Dates: imported assertions and review overlays

`start`, `end`, `rawDating`, `sourceDates` and the parsed source `phases` have not been rewritten. A separate `reviewedChronology` supplies the filter interpretation, with a policy, sources and qualifications. Where no review exists, the earlier parsed-component rules remain in force. Discovery, publication, custody and replica dates never become inscription phases.

- RIB entries giving no date no longer gain a positive year match from a wide numeric export range. For example, RIB2243's raw71–330 range remains visible, while its reviewed assertion is undated. These records require uncertainty opt-in.
- RIB2224 preserves two commentary phases and the conflicting Date-field reading. RIB2290–2292 separates an undated erased component from two dated reinscriptions. Gaps and unknowns remain explicit.
- Smith's Shield's fourth-century assignment applies to the secondary inscription; its primary inscription remains undated. It does not date the original base, its installation or the road.
- RIB2236's305–353 range remains an object-level assignment. RIB3522's222 year is a probable attribution, and its qualification is shown beside the reviewed date. Century conversions provide a filter interval, not a measured erection date.
- Donnas's mileage text remains undated; the Augustan hypothesis stays an alternative. CIL X8009's single-year catalogue value is replaced in the analytical overlay by a qualified Philip attribution,244–249, while the raw244 is preserved.
- [Reddé's Sacquenay discussion](https://una-editions.fr/23-autour-de-mirebeau/), paragraph before note20, explicitly dates it to January44. The reviewed filter uses44 and displays the raw43 conflict. Its starred CIL concordance does not make it a falsum.
- Cesiomaggiore and the Troas manuscript retain their numerical claimed/transmitted dates under `conditional-transmission`. Uncertainty opt-in may display them when their retained bounds are compatible; they never count as affirmative dated overlaps and cannot enter road comparisons. This is different from claiming that every date in the text is absent.
- Ritopek preserves the distinction between a supported lower terminus and a tentative upper suggestion; its overlay is one-sided, explicitly scoped to the consulted metadata rather than falsely claiming full collation of the underlying article. The reversed364–357 source remains invalid, without a speculative repair.

The resulting strict1–2CE query has47 eligible bounded catalogue overlaps and none in Britannia. The previous48 source-component count remains reproducible from unchanged imported phases; Donnas now requires the reviewed undated policy. Neither count authenticates the remaining broad intervals.

## Location and partner-identity controls

RIB2313 supersedes “probably near Cramond” with Ingliston. Maxwell, PSAS113(1983),379–390, discusses the Newbridge context but leaves the exact original spot unresolved. The raw Cramond coordinate is preserved and withheld; no replacement coordinate is fabricated. This is one additional catalogue-review locality block alongside the six existing curated coordinate quarantines.

Beek's linked EDH entry describes a tile from a different French context. This is an incompatible partner enrichment, not evidence that the Beek milestone is a tile or a falsum. The raw EDH fields remain in provenance, while ordinary findspot/repository presentation withholds them. Question marks, modern museum copies and mistaken imperial titulature are not treated as authenticity verdicts. Possible duplicate records remain separate until object identity is established.

## Atlas behavior and reproducibility

The **Review catalogue exclusions & cautions** control opens all targeted decisions directly in the inspector, including excluded records outside the active query. Each decision shows its reason, sources and limitations. Excluded rows are hidden in normal milestone results and road comparisons; date-compatible diagnostic display requires the separate rejected-record option. The ordinary date rules still apply. Hollow red symbols identify rejected diagnostics, hollow grey symbols uncertainty admissions, and amber outlines/fills cautions or screening flags. Cluster descriptions and member lists expose these differences.

The catalogue review filter can focus on targeted source reviews or cautions/metadata flags. It composes with the existing dates, spatial tiers and layer controls. An explicit inspector selection does not add an excluded point or stale halo to the map. Conditional text and withheld locations cannot enter distance comparisons. The current number of 47 curated dossiers and573 curated claim-ledger entries is unchanged: these new catalogue reviews are a separate layer of assertions, not fabricated new physical objects.

`tools/catalogue_audit.py` attaches reviews without rewriting source assertions. `data/catalogue-review.json` records the merged ledger, input identities, counts and scope; `data/milestones.json` retains each raw assertion alongside its review and metadata flags. The underlying source files remain unmodified with their SHA256 values in the import manifest. Public-domain CIL pages are bundled for inspection; modern CIL pages and publication PDFs are not redistributed. Tests and release receipts are in `qa/` in the portable package. Full global CIL collation, current custody checks, field verification and authentication of unchecked records remain outside this release's scope.

## Verification receipt

Final local checks passed:126 semantic/source-preservation checks,32 publication-chronology regressions,69 base chronology regressions,46 v8 catalogue browser checks (five actual map clicks),42 retained pointer/filter checks (ten actual map clicks), and118 existing core/primary-context/site-position/renewed-evidence browser checks. No browser runtime error was recorded. Desktop/mobile audit and reviewed-date panels were visually inspected. `qa/repair-release-v8.json` binds the new browser receipt to the tested code/data hashes; ZIP CRC and payload checks are recorded separately at packaging. External basemap availability and a physical object autopsy are not established by these local browser checks.