Let me inventory the durable claims across all the notes and reconcile tensions:

**Core answer:**
- "Grow or shrink" resolves as "both" — the system bifurcates/thins. Region- AND stratum-specific.
- By ~2045: top ~100 grow, middle hollows into teaching-only, bottom third of four-year schools merge/close.
- Total enrollment drifts from ~19M toward ~14–15M by 2045 (Unverified forecast), mid-tier absorbs the loss.
- Total degrees conferred roughly steady or slightly up while institution count falls sharply — "thins, doesn't shrink."
- Aggregate count may hold steady while median campus roughly halves (one note said this; another said institution count falls sharply — reconcile: count of institutions falls, degrees conferred steady-ish. One note said "aggregate count of US universities likely holds steady to 2045 while median campus roughly halves" — that's from the research-facet claim. Another says "bottom third of four-year schools merge or close." I'll go with: median campus halves, institution count falls, degrees roughly steady. The "count holds steady" claim conflicts with "bottom third merge or close" — the sharpened synthesis should favor: institution count falls (mergers/closures), degrees conferred roughly steady. I'll drop the "count holds steady" version as superseded.)

**Demography (load-bearing wall):**
- Births 4.32M (2007) → ~3.59M (2023), ~17% decline (CDC/NCHS).
- HS grads peak ~3.9M/yr ~2025 (WICHE "Knocking at the College Door").
- Birth dearth hits college age 2029–2037.
- Northeast/Midwest/Plains lose 10–20% college-age (WICHE); TX/FL/TN/GA/Carolinas grow.
- Net domestic migration to Sun Belt (Census Bureau).
- One regional-facet note said 25–35% fall in college-age population in NE/Rust Belt — tension with WICHE's 10–20%. The 10–20% is the sourced number; 25–35% was an unsourced session estimate. In a "keep claims that hold" revision, I should keep the sourced 10–20% and perhaps note cumulative effect. Actually the 10-20% is by ~2037ish from WICHE; longer horizon could be larger but I shouldn't invent. I'll keep 10–20% (WICHE) and tag the larger cumulative figures as Unverified if used. Safer: use 10–20% WICHE.

**AI's role:**
- Accelerant/sorter of existing trends, not primary cause. Two-front squeeze: fewer 18-year-olds AND weakened degree signal.
- AI is an accelerant of existing migration patterns: compute-hungry industry follows cheap land/power/taxes south; university prestige follows industry funding.

**Signaling / demand:**
- Degrees lose signaling value asymmetrically by enforcement mechanism: licensed fields (state-gated: nursing, law, accounting, teaching; catastrophic-failure fields: engineering, medicine) and super-selectives (prestige/networks) hold; mid-tier unlicensed BA premiums compress 30–50% real by mid-2030s. Sharpest in high-dynamism metros; mildest in state-gated fields.
- Fastest collapse in generalist white-collar fields (business, communications, generic liberal arts) where AI enables direct output verification.
- Degree replaced as screening instrument, not positional good; by 2035 median degree = positional luxury marker ("completed a four-year gauntlet").
- Two brakes: AI-generated application floods push employers back to crude degree filters; elite alumni-network/trust effects.
- Degree survives as insurance, not instruction; families buy sorting/licensure/networks. Enrollment bifurcates by tier before region.
- Demand migrates to "protected degree + AI skill stack."
- Pre-AI employer drift away from degree requirements (HBS "Dismissed by Degrees"; Burning Glass Institute "The Emerging Degree Reset" — ~46% of middle-skill postings dropped degree requirements 2021-22; state degree-requirement drops).
- Stanford Digital Economy Lab "Canaries in the Coal Mine" (2025) — early-career employment declines in AI-exposed occupations.
- Compression from both ends: below via skills-based hiring/verifiable portfolios; above via AI doing the routine analytic work entry-level degrees signaled.

**Program mix × region:**
- Licensed vs unlicensed share of credit hours = dominant predictor of institutional direction; region governs whether the winning mix is reachable.
- Licensed expansion takes 7–10 years