1. Bifurcation into ~150 fortress research institutions (physical/organizational moat: instruments, clinical trial networks, proprietary datasets, federal-lab partnerships, grant pipelines) vs ~3,000 teaching-first schools whose product AI makes near-free — shrink/merge/close. Research gravity follows money and population: SEC footprint, Texas, Research Triangle gain R&D/talent share; Northeast teaching privates and Midwest small-regionals shrink fastest. System splits along research/teaching line, survivor map is regional.

2. Scarce academic asset shifts from producing knowledge to verifying it — credibility infrastructure: peer review with provenance, replication services, standards-setting, trusted credentialing. Universities leaning into verification survive; those competing on output volume lose to industry labs on speed and AI on cost.

3. 20-year research split: small tier of "verification institutions" monetizes trust; long tail of research universities abandons discovery, reverts to teaching colleges. Credential value migrates from who produces findings to who certifies them. Regional hardening: legacy R1s with endowments and federal-lab ties become verification monopolies; SEC schools flush with money buy star talent but not reputation, overpay for names while trust rents accrue to Harvard/Stanford/MIT-class brands. Consolidation, not uniform growth: research headcount per campus shrinks 30–50% as AI absorbs drafting/analysis labor, but a few hundred institutions capture premium pricing on verification.

Prior notes on this session: demography as load-bearing wall, program-mix × region (licensed pivot), cost structure (savings competed away), bifurcation by 2045 (top ~100 grow, middle hollows, bottom third merges/closes), revenue shock before headcount.

This note should focus on the research function facet. The most evolved durable claim here: the bifurcation between fortress research institutions and teaching-first schools, plus the verification pivot — the scarce asset shifts from producing knowledge to verifying it. And the regional map: research gravity follows money/population to SEC/Texas/Research Triangle, but trust/verification rents stay with legacy elite brands (Harvard/Stanford/MIT). SEC can buy talent, not reputation.

Load-bearing numbers: ~150 fortress research institutions vs ~3,000 teaching-first schools; research headcount per campus shrinks 30–50%. These are projections from the session — speculative, so maybe flag "Unverified" or present as modeled. The ~3,000 figure — US has roughly 2,600–4,000 degree-granting institutions depending on counting (IPEDS lists ~3,900 degree-granting postsecondary; ~2,600 four-year). The claim says ~3,000 teaching-first schools. I can keep it as session-modeled figure. IPEDS is a real source for institution counts — degree-granting institutions: roughly 2,600 four-year. So "~3,000 teaching-first schools" vs "~150 fortress" — I could attribute the count of institutions to IPEDS as the denominator. Safer: "Unverified:" for the 150/3000 split and 30–50% headcount.

Interesting tension within the claims: research gravity (R&D money, talent) moves to SEC/Texas/Triangle, but verification/trust rents stay with legacy brands. That's a genuinely sharp synthesis — two different geographies: production geography follows money/population; certification geography follows accumulated trust. Worth making the central line.

Also connects to prior notes: middle tier loses research identity first, becomes teaching-only (from earlier session). Cost structure note said savings competed away. This note is the research-facet one.

"# By ~2045 the US research function splits into two geographies: R&D production follows money and population to the SEC footprint, Texas, and the Research Triangle, while verification rents — provenance peer review, replication, trusted credentialing — concentrate in legacy elite brands; teaching-only schools outside both moats shrink or close (assessment 2026)"

Maybe slightly shorter but keep the substance. The example titles are long one-liners.

Structure: title, 1-3 short paragraphs, end with "What would change this: …" and a source line or Unverified.