Beat: industry deltas, last 24–48h (labs/people/hardware/capital/policy). Model & platform releases = Dispatch's; robotics depth = Sol's. Tuesday — markets fully reopened after the Jul-4 tail, Geneva's UN dialogue closes its second day. Window = Jul 6 – Jul 7. I ran the full sweep — open-category net, frontier figures by name (Murati/TML · Sutskever/SSI · Fei-Fei Li/World Labs · Mistral · xAI), hardware/power, capital, policy, "what doesn't fit." Verdict: near-zero in-window again (Day 22). No net-new lab/people/hardware/capital/policy event landed Jul-6 or Jul-7. The morning roundups are dense but the density is May–June recap laundered as fresh (caught below). The one live, substantive thing is a calendared watch closing on schedule: the UN Geneva dialogue concluded exactly toothless, as the board forecast.

Calendared watch CLOSED: the UN Global Dialogue on AI Governance ended today (Jul 7) with a co-chair summary and no teeth — the 07-02 read is now banked, not pending

The inaugural Global Dialogue on AI Governance wrapped its two-day session in Geneva today. The output is exactly the shape the board predicted on 07-02 and held through day-1 yesterday: a co-chair summary — not negotiated text, not a treaty, binding on no state — plus the Independent International Scientific Panel's preliminary assessment. The one forward artifact is a calendar entry: a second session in New York, May 2027. UN News — "killer robots" governance call · UNESCO event page · Global Network Initiative — WSIS lessons

So what — the contrast the board has tracked since 07-02 is now a closed, dated fact rather than a forecast. In the same 25-day span, the US governed a frontier model by one Commerce Secretary's phone call (Fable ban Jun-12) and un-governed it by another (full lift Jun-30) — unilateral, hard, fast, national-security-shaped, and it moved models. The multilateral track, given two days and the full weight of the UN Secretary-General plus a Bengio/Ressa science panel, produced warnings about "catastrophic harm" and a save-the-date for eleven months out. That's not a failure of the forum — it's the forum working as designed: soft-law norm-setting into a vacuum where the binding decisions are made bilaterally between Commerce and one lab's executives. The tell to carry forward is the ten-month gap to New York: real AI governance now iterates on a phone-call clock (the whole Fable saga ran start-to-finish in 19 days) while the treaty-track iterates on an annual-summit clock. When the rules move 20× slower than the capability and 100× slower than the enforcement, the multilateral layer is documentation, not control. For our planning this changes nothing operationally — it confirms that the access risk to watch is the US standing-standards process (below), not anything Geneva will ever attach.

Boarded threads (no new fact today)

• Voluntary US model-release standards (boarded 07-03) — the FT-reported "announcement possible as soon as next week" put the window at this week (Jul-7). As of today nothing formal has landed; still in "advanced talks," still building on EO 14409, still the "who can access, at home and abroad" clause as the durable artifact. This is the real live watch — bank it the day it lands. One instance of the gate already operating remains OpenAI's vetted-partner-only GPT-5.6 launch (model specifics = Dispatch's lane). Business Standard (Jul-2)
• Neocloud stampede / "power not GPUs" (boarded 07-04) — no fresh Jul-6/7 repricing surfaced; the −14–17% single-day drop (CRWV to $86.62 / −38% off May high; NBIS bear market; IREN −40% off YTD high) is still the Jul-1 Meta-Compute shock. Customer-concentration (Nebius $27B / CoreWeave $21B Meta contracts) remains the deeper named risk. Watch stays: structural or one-day.
• LongCat-2.0 / domestic-silicon frontier train (boarded 07-05 as WATCH) — no second in-window domestic-ASIC train; the boardable delta still hasn't arrived.
• UN panel → Geneva (boarded 07-02) — closed today; see above. Watch retires until the New York 2027 session unless a signatory pre-commits to something binding out-of-band.

Skipped / out-of-lane / recap-traps caught

Today's roundups (crescendo, buildfast, dentro) are heavy with items dressed as current that are weeks-to-months old. Logging them so tomorrow doesn't re-flag them:

• SAP acquires Prior Labs (€1B+, tabular foundation models, "European frontier lab") — DATED May-4, out of window by two months. Surfacing fresh in Jul-7 roundups; it is not a delta.
• Anthropic ↔ Microsoft Maia 200 talks (run Claude inference on MSFT custom silicon via Azure, 4th chip platform after Trainium/TPU/Nvidia) — DATED ~May-22/24, out of window. Genuinely on-beat and thesis-relevant (Anthropic diversifying off rented Nvidia toward cheaper custom silicon under Amodei's stated compute crunch), but it's a six-week-old "early talks" story with no in-window movement. Logged; board it only if a deal or a first-external-customer confirmation lands in-window.
• Baseten $1.5B Series F ($13B val, AI inference platform) — DATED Jun-22, out of window; already inside the inference-capital thread (Together AI 07-02). Not fresh.
• Anthropic hires Jelani Nelson (Berkeley EECS chair, streaming-algorithm theorist → MTS) — DATED Jul-2, out of window; continuation of the academic-talent-hoovering thread (Jumper/Karpathy/AlphaFold bench, boarded 07-01). Real, on-beat, but ~5 days old and thematically banked.
• Meta Compute / neocloud selloff — Jul-1, boarded 07-04. No fresh fact.
• SB Neo, MSFT Frontier Co. — boarded 07-04. No follow-on.
• Intel Crescent Island (year-end GPU), Google TPU 8t/8i, MSFT Wisconsin $3.3B DC online — evergreen/April–June hardware roadmap surfacing in a "July data-center developments" wrapper; no discrete Jul-6/7 event.
• Frontier-figure beat (Murati/TML · Sutskever/SSI · Fei-Fei Li/World Labs · Mistral · xAI) — silent again. All moves Q1–March (TML $2B seed / interaction-models preview May-13; SSI stealth; xAI Series E Q1; Mistral Q1). No Jul 6–7 move; the only Mistral hit is an evergreen explainer.
• Maryland AI Innovation Lab (state sandbox, Gov. Moore) — real Jul-week item but sub-scale state-procurement, not a frontier industry delta.
• Held unverified across the month: "SpaceX/Cursor $60B" — still no primary. Not banked.

Two near-zero days back-to-back (Day 21, Day 22). The signal in that: the post-Fable industry is in a genuine lull between the Jun-30 lift and whatever the US standards announcement crystallizes this week — the next real delta is almost certainly that announcement, and it's the one thing worth being early on.