Cycle on Sep 28, 2026

I'll run a research cycle on this topic. Starting by checking open threads and existing notes, then hunting for the delta since the last cycle (Sep 21–24).

I have 5 standing findings and 4 threads (including a watch on the Section 106 final rule and the Vesuvius second-scroll prize). The delta window is roughly Sep 21–28. Let me pull the HN story on the Saqqara tomb, check the policy delta on Section 106, and scan for fresh science news.

Two strong leads already: **Apollo** — an Ancient Greek LLM from the Austrian Academy of Sciences + Mistral AI unveiled Sep 23 — and the Saqqara HN thread, which is a vivid data point for the AI-slop-policing theme. The policy query returned noise, so I'll check the ACHP Section 106 rule status directly. Let me drill into the primary source on Apollo and scan other fronts.

The Apollo article is a rich primary-adjacent source (reproducing the ÖAW release): first foundation LLM for Ancient Greek, €400K funded by Mistral AI + SAIL Reply, 80% accuracy on masked passages, free and open-access. Now let me check whether the ACHP Section 106 final rule has landed (that would fire a watch), and scan for other science-news deltas.

Key deltas emerging: (1) Apollo LLM launch; (2) the ACHP Section 106 final rule has **not** yet published (watch stays open — latest ACHP doc is a Sep 18 categorical-exclusion notice); (3) a C&EN piece on handheld XRF lead-screening of Herculaneum scrolls, which directly advances my open question thread. Let me read the C&EN article and check the Nature LiDAR piece.

Both articles are substantive. The C&EN piece validates the XRF triage workflow (advancing my open question thread), and the Nature piece reveals a new player pattern: Yamaha Motor selling forestry-derived LiDAR services into Maya archaeology. Let me do a final sweep for funding news and any recent HN archaeology-tech stories I might be missing.

I have a clear picture of the delta now. The Section 106 final rule has **not** published (watch stays open — verified against the ACHP's Federal Register page directly), the Vesuvius prize watch hasn't fired, and no news on the Brazil/UNESCO regulatory watch. I have four solid evidence-backed updates. Writing them now, reusing existing slugs.

The call went out empty — retrying with the actual note payloads. All four updates are independent, so I'll write them together.

All four notes are written. Thread checks: the XRF lead-screening question stays open (the method is now validated, but no systematic screening announcement or lead-positive scroll count yet); the Section 106 watch was verified against the Federal Register today and correctly remains unfired; the Vesuvius prize and Brazil/UNESCO watches likewise have no news. Nothing new needs a thread — the Latin-model roadmap is tracked inside the decipherment note.

Cycle complete. Here's what this week's research turned up: **The big story: Apollo — the first foundation LLM for an ancient language (Sep 23)** The Austrian Academy of Sciences unveiled **Apollo**, built with **Mistral AI and SAIL Reply** plus David Smith (Northeastern — the Vesuvius orbit). Trained on just 600M words of Ancient Greek, it hits ~80% accuracy restoring masked papyrus passages, and in a blind study experts rated its restorations *at least as good as human scholars* in 77% of cases — better in 16–20%. It demoed on a Roman birth certificate, a carbonized Vesuvius scroll, and a Black Sea inscription yielding new Roman-law evidence. Free, open-access, ÖAW-owned, with a Latin model (~12B words) next. Meanwhile C&EN validated the **handheld XRF lead-screening triage** that decides which Herculaneum scrolls can enter the CT pipeline at all. **Funding pattern shift: corporate patronage is the new decisive funder.** A frontier AI lab bankrolled a national academy's flagship tool for **€400K** — rounding error for Mistral, transformative for the humanities. Yamaha Motor, separately, is selling its forestry-derived UAV LiDAR service into Maya archaeology at Copán (April 2026 preliminary survey; 2,500-ha valley survey planned). Heritage is becoming a customer segment for industrial sensing and a legitimacy play for AI labs — that's the lucrative-opportunity thread: integration/validation layers between labs and institutions, sensing-service resale, and compliance tooling. **US regulatory check:** the ACHP's final Section 106 rule (36 CFR 800) has **still not published** as of today — verified directly against the Federal Register; the watch stays open. **Culture note:** HN's discussion of a real, ministry-announced Saqqara tomb discovery derailed entirely into AI-detector policing of the article's prose ("AI writing smell is as bad as bad grammar"). Public archaeology comms is now credibility-filtered by detectors — outreach needs human-verified writing as trust infrastructure. Four standing findings updated in place; no new threads or watch events needed.