# Attalus bibliography review apparatus

These main editorial reviews accompany the fifteen metadata records. Internal
JSON filenames refer to the retained review packet, not additional downloads.
The public calibration table retains source grounding and locators while
omitting private response paths. Raw source bodies are excluded. All current
atlas citation labels and historical judgments remain unchanged.

# Attalus metadata — Fact-Check and Revise

Main review, 9 September 2026; Fact-Check and Revise v2.8.

## Intake and findings

The venue remains Craig's own research site and accompanying downloads.
Model-assisted metadata composition is authorized. The subject is the fifteen
metadata candidates, not a rewrite of the monograph or historical case prose.
Both independent reviewer requests concern separate work and remain pending.

**Correct.** Fourteen inspected page bodies identify fifteen access records.
The Cicero components supply their speech/book titles and named translation
credits: Watts/1928, Freese/1930, Hodge/1927, Winstedt/1919 and Lord/1937.
Warmington/1940 is the supplied translation credit for the Public Works
inscription collection. These are host-reported credits, not inspected printed
imprints. The page ranges are retained separately from the original case locators.

**Overstatement avoided.** Orosius is explicitly an adaptation of Raymond's
translation. Memnon's page identifies Jacoby's Greek text and opens with Photius'
introduction; its chapters are not the original book numbers. Granius preserves
the distinction between Criniti's sentence numbers and Flemisch's page numbers,
and the host's warning about fragmentary text. The two Granius access records
share one downloaded body but have distinct URL/anchor scopes. They are not
merged. The two Roscius pages are complementary parts, not duplicate page bodies.

**Unverifiable fields.** Four records lack explicit personal translator credits:
the Samos inscription, Memnon and the two Granius accesses. Orosius does not name
its adapter. Those five fields remain null. Andrew Smith's HTML author metadata
on Memnon/Granius is retained in its actual role; it is insufficient to fill a
translator field. The two name indexes' copyright credits likewise do not prove
personal authorship. Unlabeled footer dates stay literal and unassigned. No
first-publication day or modification date is silently inferred from them.
The five formal contributor exceptions are not an exhaustive count of every
omitted imprint or date field.

**Source quality and attribution.** The host's guide explains the limitations
of online translations, including uncertain readings and the absence of
scholarly review for its new translations. Its translation index says that
earlier translations were re-edited. These are attributed host disclosures,
not this review's blanket reliability judgment. The notation key identifies
editorial additions, textual gaps and uncertain meanings. These notices were
consulted, not inferred from the presence of translated text. No source-funding
or independence claim is introduced.

**Structure and register.** Dentatus and Nequinum are modern name indexes with
links to events and translated sources. Their inherited inventory labels call
them Primary source; the typed view corrects that access-type description while
preserving the original label as provenance. The actual atlas labels already
describe them as ancient references. Historical support must rest on the cited
texts, not the fact that an index exists. The other accurate purpose notes are
preserved rather than mislabeled as defective bibliography titles. Descriptive
metadata does not adopt Cicero's advocacy or the host's evaluative descriptions.

The Samos page dates the inscription soon after 321 BCE; the existing restoration
case displays 322/1 BCE. This creates no new chronology hold. Original actor,
cohort, motive and outcome judgments remain unchanged; this is not a historical
adjudication of Samos, Gracchan land division or the Sullan cases.

## Grounding and revision

SOURCE_OBSERVATIONS.json binds the headings, credits, metadata and footer
observations to fourteen exact HTML bodies and decoded text. Four host guidance
pages were also consulted through the web reader: the home page, translation
key, usage guide and translation index. The underlying Greek/Latin editions
and printed title pages were not inspected. Only the metadata and relevant
source-use notices were reviewed; downloading a body is not full historical
reading of it. No source claim relies on model memory.

The initial retrieval stopped before network access because an optional parser
was absent. The replacement used the standard library. Four derived text files
were then decoded again using the saved pages' declared character sets; raw
HTML stayed identical and no source request was repeated. The record is in
TEXT_REEXTRACTION.json. No public archive URL is verified. The earlier archive
timeout was not replayed; this packet's exact bytes provide private preservation,
not a claimed Wayback or archive.today snapshot. Public access is verified for
the recorded access date only.

The revised candidate file retains the proposed source-qualified metadata; no
additional wording change is required by this round. Each record carries its
source URL, original identity, access date and limitations. The review notes,
field exceptions and calibration table must accompany later publication.

## Required self-audit

| Required check | Disposition | Evidence or reason |
|---|---|---|
| Grounding | Pass | Fourteen fetched source bodies and four host guidance pages support the recorded metadata; no remembered edition is supplied. |
| Required vs. offered | Pass | Modern index classification is corrected; faithful historical purpose notes remain intact. |
| Claims | Pass | Explicit names and years are retained clearly; translator/adapter gaps remain null rather than guessed. |
| Deletions | Pass | All fifteen access records and 72 original occurrences remain. |
| Criticisms | Pass | No external reviewer finding or replacement text was adopted. |
| Internal consistency | Pass | Text authors, translators, adapters, web metadata and copyright roles remain distinct. |
| Register discipline | Pass | The metadata does not adopt an ancient speech's advocacy or a host's praise as its own assessment. |
| Moral framework register | Pass | No normative historical or moral claim is introduced. |
| Political-spectrum contamination | Pass | No political placement label is introduced. |
| Attribution posture | Pass | Translation years and source limitations remain host-reported; no print-scan verification is implied. |
| Quotations | Pass | Titles and short credit strings follow the inspected page text; paraphrases remain unquoted. |
| Causal reasoning | Pass | No new motive, cause, mechanism or adaptation-to-environment claim is inferred from bibliographic metadata. |
| Categorical vs. dimensional | Pass | Work/access typing does not create a binary historical refugee or colonization judgment. |
| Denotation/connotation and convenient layer | Pass | A source index is not promoted to an ancient text; a copyright or meta-author credit is not promoted to a translator. |
| Proper-name variants | Pass | Supplied initials and the Samos title/identifier remain explicit; names are not expanded from memory. |
| Engagement posture | Pass | Historical arguments and their existing qualifications remain intact. |
| Confidence calibration | Finding retained | Five personal contributor fields remain open; unlabeled footer dates and uninspected imprints remain qualified. |
| Fetched-source disclosures | Pass with limits | Host translation/notation warnings were read; no funding, independence or scholarly-review certification is invented. |
| Citation resolution | Pass at metadata scope | Fifteen stable IDs map to the inspected pages and existing twenty citation IDs; passage acceptance is unchanged. |
| Adopted reviewer text | Pass | No reviewer-provided replacement prose is present. |
| Editorial synthesis (SYNTH) | Pass | The shared-body and complementary-part relationships are bounded metadata judgments, not new historical conclusions. |


# Attalus metadata — LLM Artifact Tells

Main review, 9 September 2026; LLM Artifact Tells v2.7. Input:
REVISED_IDENTITY_CANDIDATES.json and the first-round findings. Craig's own
site and downloads permit this model-assisted metadata work. This pass does
not claim human authorship or seek to conceal the method.

## Findings and tier coverage

| Scope | Finding | Tier | Disposition |
|---|---|---|---|
| Candidate fields and first-round prose | Actual marker, URL and encoding searches found no leaked citation tokens, tracking markers, replacement characters or writing-block residue. | 1 | No repair required; TELL_CHECKS.json records the searches. |
| Contributor and date claims | Web-author, copyright, translator and adapter roles remain separate. Missing personal credits and unlabeled footer dates are not filled with plausible names or dates. | 2 | No fabricated identity or unsupported historical synthesis found. |
| Metadata structure | Repeated scope/limit fields are part of the data schema and perform distinct provenance work. | 3 | Keep them; uniform structure is not a prose defect here. |
| Vocabulary and qualification | Terms such as adapted, fragmentary and index identify observed source forms. | 4 | Retain meaningful distinctions; no vocabulary-only finding. |
| Punctuation and formatting | Initials, ranges, CIL superscript notation and field names have bibliographic functions. | 5 | No cosmetic rewrite required. |

All five tiers were considered in order. The original case prose and source
captures are outside this decontamination pass. Captured web-reader traces are
private evidence, not public metadata prose. No DOI or ISBN was introduced;
all fifteen access dates parse as real dates. The exact old labels remain as
provenance and are not falsely claimed to have been decontaminated.

The revised metadata remains unchanged by this pass. Its access-type correction
does not imply a new verdict about the source's historical reliability. No
guideline-violating synthesis or significance claim needs repair. Private
capture paths belong only to review evidence; any public apparatus must omit
them while retaining source URLs and grounding.

## Required self-audit

| Required check | Disposition | Evidence or reason |
|---|---|---|
| Checklist vs. heuristic | Pass | Required mechanical searches were run; meaningful metadata terms were assessed in context. |
| False positives | Pass | Schema repetition and explicit role distinctions are retained because they carry information. |
| Hard artifacts | Pass | TELL_CHECKS.json records the required marker searches, encoding check and fifteen valid access dates; no new DOI/ISBN is present. |
| Tier coverage | Pass | Tiers 1–5 are reported above; the pass did not begin or end with vocabulary. |
| Decay weighting | Pass | Objective residue checks carry more weight than stylistic familiarity. |
| Guideline vs. style sort | Pass | No source-integrity defect is recast as a stylistic preference; no significance inflation was found. |
| Introduced tells | Pass | This round makes no candidate rewrite. |
| Writer's voice | Pass | Craig's monograph and case prose are unchanged. |
| Register homogenization | Pass | Bibliographic data structure is preserved; no historical prose was flattened. |
| Quality tradeoffs | Pass | Supplied names, dates, ranges and explicit gaps remain readable and precise. |
| Editorial synthesis (SYNTH) | Pass | No connecting historical claim is added; the next calibration round checks the final metadata. |
| Venue check | Pass | The authorized venue is Craig's own site/downloads, with model assistance retained in provenance. |
| Provenance honesty | Pass | A clean metadata pass does not change authorship or establish full source/printed-edition verification. |


# Attalus metadata — Calibration Review

Main review, 9 September 2026; Calibration Review v1.2. The final candidate
fields were audited against the inspected page bodies and fetched host guidance.
No independent reviewer was launched.

## Full audit table and coverage

[CALIBRATION.json](CALIBRATION.json) contains the complete twenty-one-row table:
claim locations, load-bearing terms, source URLs, exact grounding spans and
locators, licensed scope, materiality, verdicts and proposed wording.

Fourteen page groups cover all fifteen metadata records; the two Granius access
IDs share one group because they use the same retrieved body. Two further
groups cover the host's translation and notation disclosures. Five separate
rows cover the unfilled personal contributor fields. The counts are **sixteen
Matched and five Uncheckable**, with zero Overclaimed, Underclaimed or
Unsupported-but-sound. All rows concern detail-level metadata or source-use
disclosures. Every Matched row has exact source grounding; copied spans were
checked against the recorded source text or HTML.

The audit covers named texts and components, supplied contributor/rights roles,
host-reported translation years, literal footer strings, modern-index typing,
and the stated interpretation limits. An unassigned footer date remains
unassigned: this is not a finding that its original-publication date is known.
The five counted exceptions cover personal contributor roles; omitted imprint
fields and unclear date roles remain disclosed without pretending this is a
complete census of all possible bibliographic omissions.

Operational access timestamps, source-body hashes, case joins and preserved
occurrence counts are checked separately against local evidence. Full ancient
translation accuracy, printed-edition identity, source independence and all
historical assertions in the original labels/cases are outside this audit.
The Samos date observation preserves the existing 322/1 case display and does
not re-adjudicate its chronology. No public archival capture is certified.

## Concentration and reverse check

The five Uncheckable fields concern the Samos translator, the Memnon translator,
both Granius translator fields, and the Orosius adapter. None is filled from a
host credit, HTML author field or model recollection. The source remains
reachable; the unresolved issue is the missing role-specific credit. No
Unsupported-but-sound classification or associated re-cite claim is used.

The reverse check actively tested whether explicit credits, dates and index
descriptions had been unnecessarily weakened. None had. The candidates state
the supplied names and years directly while preserving their reported roles.
No candidate rewrite is required by calibration.

## Required review self-audit

| Required check | Disposition | Evidence or reason |
|---|---|---|
| Direction of changes | Pass | Sixteen groups are Matched and five personal fields stay Uncheckable; no blanket softening occurred. |
| Lane discipline | Pass | The review assesses source-to-metadata alignment, not prose style or historical classification. |
| Deletion vs. flagging | Pass | All fifteen access records and five unresolved fields remain present. |
| Mandatory re-cite flags | Pass | No Unsupported-but-sound verdict was used; explicit nulls and FIELD_EXCEPTIONS.json carry the actual contributor gaps. |
| Verbatim grounding | Pass | Every Matched group has short exact source spans and locators, with substring checks against bound captures. |
| Coverage disclosure | Pass | All metadata groups and host disclosures are listed; historical, physical-edition and operational exclusions are explicit. |