Let me parse this carefully.

The final instruction: "Revise the notes as a set: keep every claim that still holds, drop repetition, sharpen wording. Output one Markdown note that is the current best statement of what is established. Start with a # title, then short paragraphs, then: What would change this: … Keep sources or Unverified: tags. Do not mention tools, sessions, or that you are revising."

So this is the grand synthesis/consolidation note. I need to produce ONE Markdown note that is the current best consolidated statement of everything established about "Do universities grow or shrink due to AI?" over the next 20 years, with regional dependence.

Format:
- # title (the claim in one line)
- Short paragraphs
- End with "What would change this: …"
- Keep sources or Unverified: tags
- No tool/session/revision mentions

Let me compile the durable claims inventory from all the notes:

**Core answer:**
- "Grow or shrink" resolves as "both" — the system thins and bifurcates; region- and stratum-specific.
- Total enrollment drifts ~19M → ~14–15M by 2045 (Unverified).
- Degrees conferred roughly steady while institution count falls — "thins, doesn't shrink."
- Median campus roughly halves in enrollment (maybe too strong; keep as inventory claim).
- AI is an accelerant/sorter of existing trends, not the primary cause. Two-front squeeze: fewer 18-year-olds AND weakened degree signal.

**Demography (load-bearing, sourced):**
- 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 cohorts (WICHE); TX/FL/TN/GA/Carolinas grow.
- Net domestic migration to Sun Belt (Census Bureau).

**Signaling:**
- Screening vs insurance decomposition: screening dies first (AI-verified work samples cheaper than proxy-filtering); insurance (blame-shifting/liability cover) is sticky residue, persists in regulated/litigation-exposed sectors through 2040s. Post-2035 degree demand tracks a sector's legal/liability regime before it tracks region.
- Licensed/statute-gated fields hold (nursing, law, accounting, teaching; catastrophic-failure: engineering, medicine); super-selectives hold.
- Mid-tier unlicensed BA premiums compress 30–50% real by mid-2030s (Unverified).
- Fastest collapse in generalist white-collar fields (business, communications, generic liberal arts).
- By ~2035 median degree = positional luxury marker; degree reprices from entry ticket to exit credential (licensure, grad gatekeeping, late-career management).
- "AI doesn't kill the degree; it kills the undifferentiated degree."
- Credential-value decay hits by major before by institution — compositional churn; campus fates track program mix.
- Pre-AI drift: Burning Glass Institute "The Emerging Degree Reset" (~46% of middle-skill postings dropped degree requirements 2021–22); HBS "Dismissed by Degrees".
- Stanford Digital Economy Lab "Canaries in the Coal Mine" (2025) — early-career employment declines in AI-exposed occupations.
- Total degree-seeking enrollment falls 15–25% below demographic baseline by 2045 (Unverified).
- Mid-market enrollment contracts 25–40% by ~2035 (Unverified).
- Micro-credentials are a middle-market illusion — complement at top, fail to stabilize bottom.
- Two brakes: AI application floods push employers back to crude degree filters; elite alumni networks.
- Elite degrees gain value BECAUSE verification makes mid-tier signals redundant.
- Degree survives as insurance (sorting, licensure, networks), not instruction.
- Demand migrates to "protected degree + AI skill stack."
- Agentic hiring stacks / work-sample simulators; "bachelor's required" listings fall 30–50% in AI-exposed occupations by late 2030s (Unverified).
- Demand-side claim: residual degree value set by statutory gatekeeping, not brand signal. Enrollment holds/grows where license legally requires degree + physical attestation; decays fastest where degree is pure wage-signal.
- Supply-side: binding constraint flips from demand to physical attestation capacity (proctored assessment, supervised labs, clinical placements, defended capstones). Tiered online-plus-residency models (Georgia Tech OMSCS, ASU) absorb demand and grow; residential-only smalls stay capacity-bound and lose share.
- Hiring decomposes into sectors: degree requirements vanish first in at-will, project-based, reputation-portable