Cycle on Oct 5, 2026

I'll run this research cycle. First, let me see the open threads and existing notes for this topic.

I have the lay of the land: 4 open threads (2 watches on Vesuvius/Section 106, 2 on Brazil/UNESCO data sovereignty and Herculaneum lead-ink screening), and 5 standing findings from 2026-09-28. The policy baseline (Oct 2) flags a coverage gap — no CRM/preservation items in the federal record. This cycle's job: the delta since Sept 28/Oct 2. Let me scan the federal record and the web in parallel.

Policy record: nothing new on Section 106 or the Historic Preservation Fund in the last 14 days — the Section 106 watch hasn't fired, and the baseline's coverage gap holds. The web scan surfaced two leads worth drilling (a Peru geoglyph AI story and an AI-archaeology risks piece), plus I need to check the Vesuvius watch threads. Let me fetch the promising articles and search for Vesuvius news.

No Vesuvius breakthrough this week (the PHerc 1667 reading and prize news are pre-existing; the $1M second-scroll watch hasn't fired). Two leads worth grounding: the Nazca geoglyph AI pipeline (Yamagata + IBM) and an expert critique of AI hallucination in site detection. Let me verify the Nazca work against its primary publication and scan for anything else new in reputable science outlets.

The Nazca story is verified against its primary source — a PNAS paper from the Yamagata University Institute of Nasca + IBM Research joint project. That's a strong "players" data point. Let me do two more targeted scans before writing: one on reputable aggregators for anything new this week, and one on the Herculaneum lead-ink question (an open thread).

The Herculaneum lead-ink thread has no new developments (the PLOS ONE study from Sept 16 remains the latest word — thread stays open). One more lead: a Scientific American piece on AI + acoustics applied to an Etruscan tomb, which would be a genuinely new modality if recent. Let me check it.

I have four findings worth persisting from this cycle's research. Writing them now, plus registering the one forward-looking watch the Nazca story surfaced.

All four findings are persisted and the watch is registered. Ending the cycle.

Cycle complete. Here's what this cycle found: **1. Yamagata + IBM's Nazca pipeline is the flagship AI discovery story** (new note, `ai-aerial-imagery-site-discovery`) — verified against the PNAS primary paper: AI flagged 1,309 candidates, field teams confirmed 303 new figurative geoglyphs in six months (1,440 labor hours, a 16-fold acceleration). Running total: 893 confirmed geoglyphs, 781 AI-assisted. IBM Research built the few-shot model — the same corporate-patron pattern as Mistral/Apollo in decipherment. ~968 flagged sites remain unverified, and the Peru permit runs through 2026 → I registered a watch for the survey-completion announcement. **2. The false-positive critique now has a named voice** (`ai-archaeology-false-positives-context`) — Prof. Luca Sanna (Cagliari) on AI hallucinating "a proto-historic nuraghe in a circular enclosure built fifty years ago... by a shepherd." Verification/ground-truthing is the business gap. (Flagged the syndicated source's weak provenance.) **3. A new modality: AI + archaeoacoustics** (`ai-archaeoacoustics-sensory-reconstruction`) — Duke (Forte/Ortoleva) + Oxford's "Sounds of Silence" reconstructing the sensory design of Etruscan tombs (SciAm/The Conversation, Springer paper). The toolkit is expanding from finding and reading sites to reconstructing embodied experience — the engine of immersive-heritage commercialization. **4. US federal policy: quiet confirmed** (`us-federal-policy-crm-preservation`) — two federal-record scans found zero CRM/preservation items since the Oct 2 baseline; the Section 106 final-rule watch hasn't fired. Funding action remains with corporate patronage, prizes, and non-US academies. No developments on the Vesuvius second-scroll watch or the Herculaneum lead-ink screening question — both stay open.