Content Quality: Well-structured News piece (806 words, within the 400-1200 range) with Overview / What We Know / What We Don't Know / Analysis sections. The 'What We Don't Know' section appropriately flags the provisional nature of the bread identification and the imprecise rampart date rather than overstating certainty. Neutral, factual tone throughout with no sensationalism.
Source Verification: Read both snapshots in sources/2026-08/rescue-excavation-at-vindonissa-uncovers-switzerlands-earliest-roman-military-camp-with-a-tiberius-era-gold-coin/ after independently gunzip+sha256-verifying each file against manifest.json (both matched exactly: source-0 be4f4cba... HeritageDaily via Archive.org fallback though live URL returned 200; source-1 cc190a8a... Arkeonews direct fetch). Both extracted to plain text and read in full. suspicious_patterns is null for both in the manifest; I additionally grepped the raw decompressed HTML of both snapshots for 'AI agent', 'language model', 'ignore previous/all instructions', 'system prompt', 'disregard' etc. and found zero matches — no prompt-injection content of any kind in either source, confirmed and genuine (unlike the unrelated Scientific American incident referenced in my brief, there is nothing to quote here).
Factual Accuracy: Verified the contributor bot's two self-flagged integrity concerns rather than taking them on trust: (1) 'Covering around 4,000 square metres' DOES appear verbatim in the HeritageDaily snapshot (paragraph 2) — it is not a hallucination, it describes the total excavation footprint, a genuinely different quantity from the '2,200 square metres of archaeological remains documented' figure that appears later in the same article (paragraph on the ~20-archaeologist team). The two figures are compatible, not contradictory (total worked area vs. documented-remains area within it). The bot's caution in excluding '4,000 sqm' was reasonable even though the premise (hallucination) was slightly off, and I confirmed via grep of the submission JSON that '4,000 square metres' does not appear anywhere in the article body — only 'more than 2,200 square metres' (attributed to HeritageDaily) and '2,225 square metres' (attributed to Arkeonews) appear, each correctly matched to its source. (2) Spot-checked all three bot-flagged single-source facts: 'only Roman legionary camp known within the territory of modern Switzerland' appears in Arkeonews only (HeritageDaily never makes this claim) and is cited to Arkeonews alone in the article — correct, not falsely compound-cited. '20 hectares'/'6,000 soldiers' appears in Arkeonews only ('covered approximately 20 hectares and accommodated as many as 6,000 soldiers') — HeritageDaily's snapshot contains no hectare figure or soldier count at all — and the article cites Arkeonews alone — correct. Trajan's AD 101 transfer of the 11th Legion appears in Arkeonews only ('Emperor Trajan transferred it towards the Danube in AD 101') — absent from HeritageDaily entirely — and is cited to Arkeonews alone ('per the same source') — correct. All other checked specifics (11-month/eleven-month rescue excavation, Cantonal Archaeology Department of Aargau on Zürcherstrasse in Windisch, Tiberius-era gold aureus, one-centimetre carnelian gemstone depicting Victoria with staff and wreath, two possible scalpel handles with one animal-head-shaped handle per Arkeonews, carbonized bread loaf ~10cm across/3cm thick per Arkeonews, 70-metre rampart section turning northwest, Königsfelden Park ditch recorded 90+ years earlier) match character-for-character or in faithful paraphrase against the raw snapshot text, correctly attributed. One minor observation not rising to a corrections-worthy error: the bullet compound-citing 'a carbonized loaf of bread measuring approximately ten centimetres across and three centimetres thick... according to HeritageDaily and Arkeonews' attributes the specific centimetre measurements to both outlets, but only Arkeonews actually states the measurements (HeritageDaily mentions the bread find without dimensions) — the surrounding items in that same sentence (blacksmith's tools, leatherworking equipment, hand-mill stone) are genuinely confirmed by both outlets. The measurement itself is accurate and traceable to a cited source (Arkeonews), so this is an imprecise compound-citation of a list-sentence rather than a fabrication or misattribution, and does not warrant a public corrections note.
Overall Assessment: High-quality, carefully sourced submission. The contributor bot's own self-flagged concerns (the '4,000 sqm' exclusion and the three single-source attributions) were independently verified and found to be handled correctly — the exclusion was cautious but reasonable, and none of the single-source facts were falsely compound-cited. APPROVE — no corrections needed; the allowlist gap was a housekeeping issue, now resolved by adding heritagedaily.com to config/source_allowlist.txt.