Long Yi
← Back to all writing

AI futures evidence — X4: education & reskilling

July 5, 2026

Predictionai-futuresevidencecrosscut

← Futures index · 中文 · Main forecast

Each section: Claim · Why · Evidence · Analogue · Would update if · Conf (H/M/L).

Parent: crosscut x4 education reskilling
Labor gate: node1
Acemoglu: Acemoglu vs Autor AI Labor Economics · Power and Progress Acemoglu Johnson
Utopia: Utopia timeline · my putopia
Date: 2026-07-05
Format: Claim | Why | Evidence | Analogue | Would update if | Conf

Each section documents one probability or timing claim for S_soc X4. Probabilities are subjective elicitation unless noted as derived.

TL;DR — top 5 P’s

PPoint est.Role
CX-EDU-RESPONSE0.60Modal: activity without C4 absorption by 2032
EV-EDU-FED0.22Federal AI-literacy/reskilling package >$10B enacted
Time-to-retrain vs METR0.25Modular pathways beat agent capability doubling
Machine-usefulness tilt0.25Procurement/tax credit favors augmentation over cut-only
Corporate verified placement ≥2M0.35Conditional on EV-EDU-FED; else ~0.15 standalone

Bucket leverage (if CX-EDU-RESPONSE wrong): U4 +3–5pp; friction −3–5pp; whimper −3–4pp. If fail + U2 SWF blocked: whimper +4–6pp.

P = 0.60 — CX-EDU-RESPONSE: reskilling fails to absorb C4 shock by 2032

Claim: ~60% education/reskilling policy does not close the Canaries employment gap or re-employ displaced junior white-collar cohort at ≥80% prior wage within 18 months by 2032 — scattered pilots + corporate PR, not mass absorption.

Why: (1) Time asymmetry — METR 50%-horizon doubles ~90 days; credentialed pathway = 6–24 mo + 12–24 mo bill lag. (2) So-so automation incentive — firms profit from junior cuts without retraining obligation (Block, Cloudflare). (3) Federal fiscal headwind — MASA consolidation proposed, WIOA reauth flat; no >$10B step-change. (4) Corporate programs optimize pipeline/brand, not Challenger-displaced devs.

Evidence:

Analogue: Post-2008 TAA — trade-adjustment training authorized but underfunded relative to shock scale; displacement normalized without pathway closure.

Would update if: Canaries gap closes to <5% rel. for 4+ quarters and federal training spend ≥0.3% GDP/yr sustained and median displaced-worker re-employment wage ≥85% prior within 18 mo.

Conf: M

P = 0.22 — EV-EDU-FED: federal AI-literacy/reskilling package >$10B enacted by 2030

Claim: ~22% (range 0.14–0.32) a single bill or 3-year cumulative supplemental appropriation enacts >$10B for AI-transition training — step-change vs baseline WIOA Title I (~$3.9–5.7B/yr aggregate).

Why: Biden FY25 $8B Career Training Fund was request-only; 119th Congress acceleration default favors deregulation over new domestic spend; Workforce of the Future Act authorizes ~$250M not billions; Node 1 narrow workforce bill P≈0.30 but much smaller dollar envelope. Tail requires C5 peak legislation window + sustained Challenger salience.

Evidence:

Analogue: GI Bill ($15B+ post-WWII, inflation-adj) — historical scale reference, not forecast; COVID-era supplemental UI showed Congress can spend fast but AI training lacks equivalent coalition.

Would update if: Workforce of the Future Act marked up with ≥$3B/yr new AI line; or reconciliation package includes Career Training Fund–class appropriation with R crossover.

Conf: L–M

P = 0.75 — TEGL 03-25: permissive WIOA guidance without new appropriation (modal)

Claim: ~75% DOL TEGL 03-25 (Aug 26 2025) remains the ceiling of federal AI-workforce action through C5 — encourages existing Title I for AI literacy, no mandatory set-aside or supplemental $.

Why: Guidance issued under EO 14277 “Advancing AI Education for American Youth”; reuses governor’s reserve + Title I flexibility; Trump-admin frame = leverage existing statutory authority, not expansion; aligns with Node 1 P(<5%) federal training cap.

Evidence:

Analogue: OSHA guidance letters — clarifies permissible use of existing funds; not a funding event.

Would update if: TEGL superseded by regulation requiring ≥5% Title I set-aside for AI-displaced workers nationally.

Conf: M–H

P = 0.55 — MASA consolidation: ~29% workforce-dev cut remains live threat (FY26–27)

Claim: ~55% the administration’s Make America Skilled Again block-grant proposal (~29% cut vs FY25 WIOA Title I aggregate ~$5.67B → ~$2.97B MASA line) constrains reskilling capacity even if FY26 omnibus rejected consolidation — reprised in WIOA reauth (ASWA 2026 pilot language).

Why: FY26 enacted LHHS bill (signed Feb 3 2026) preserved separate WIOA accounts, but ASWA 2026 reintroduces consolidation pilot; budget justification explicit: $1.533B decrease vs aggregate 2025 enacted; signals fiscal default = shrink, not GI Bill expansion.

Evidence:

Analogue: Block-grant Medicaid debates — even failed consolidation shifts Overton window toward fewer strings, less $.

Would update if: WIOA reauth increases Title I authorization ≥10% real with AI-specific line item and no MASA language.

Conf: M

P = 0.08 — Workforce of the Future Act (S.3319 / H.R.6621) enacted as written

Claim: ~8% the Dec 2025 bicameral bill passes substantially as introduced — studies + $250M authorized grants, not the >$10B EV-EDU-FED threshold.

Why: Referred HELP/Education committees; Dem-sponsored in acceleration Congress; GovTrack-class bills at this scope typically <15% enactment; Anthropic/OpenAI endorsements = optics not whip count; dollar scale two orders of magnitude below absorption need.

Evidence:

Analogue: Sanders S.4214 datacenter moratorium — high-salience Dem marker bill, low enactment (GovTrack ~2%).

Would update if: Committee markup schedules with bipartisan chair sign-on and appropriation ≥$1B/yr in reported text.

Conf: M

P = 0.30 — Narrow bipartisan workforce bill (WIOA expansion / displacement notice, no cap)

Claim: ~30% Congress passes narrow AI-workforce legislation (WIOA reauth slice, displacement notice, training tax credits) — without compute cap — by 2028. Distinct from EV-EDU-FED scale.

Why: Node 1 Stix & Maas short-harm bridge; HELP hearings modal P≈0.75; culture-war vetting coalition leaves economic menu open; Pew 64% expect fewer jobs over 20 years. Dollar envelope likely $1–3B/yr, not $10B+.

Evidence:

  • node1 §P=0.30 — same reference class
  • research ai pause advocacy playbook §1.4 — bridging short ↔ long harm tractable
  • Pew Apr 2025 — job-loss expectation gap
  • AFL-CIO Workers First — retraining in eight principles

Analogue: Trade Adjustment Assistance — recurring authorization, never scaled to shock.

Would update if: 119th adjourns with zero workforce AI bills out of committee despite sustained Challenger AI-cited cuts.

Conf: L–M

P = 0.80 — Amazon Future Ready 2030: high PR, partial reach, wrong cohort mix

Claim: ~80% Amazon Future Ready 2030 ($2.5B global, Oct 2025; $1B Career Choice) does not materially retrain Challenger-displaced junior devs — targets existing employees + cloud/mechatronics pipeline, not Block-class laid-off SWEs.

Why: Announced Oct 23 2025, days before 14k internal cuts; 50M people target is global over 5 yr; Career Choice focuses cybersecurity/logistics/mechatronics — not agent-era SWE stack; prior 700k upskilled were in-house hourly/salaried L4+.

Evidence:

Analogue: Y2K vendor certification programs — trained incumbent IT staff, not displaced categories at scale.

Would update if: Amazon publishes verified placement of ≥50k external dislocated tech workers in 12 mo with wage ≥80% prior.

Conf: M

P = 0.75 — Google AI Opportunity Fund + $1B higher-ed: literacy scale, not displacement absorption

Claim: ~75% Google $75M AI Opportunity Fund (Apr 2024) + $1B/3yr AI for Education Accelerator (Aug 2025) remain foundational literacy + cloud cert plays — not mass re-employment infrastructure for C4-displaced cohort.

Why: AI Essentials ~10h course; 1M Americans target = literacy not credentialed career transition; $1B includes in-kind Gemini/NotebookLM subscriptions (Manyika declined to split cash vs in-kind); 100+ universities signed — supply-side, not dislocated-worker WIOA interface.

Evidence:

Analogue: Grow with Google Career Certificates (11M trained since 2017) — broad digital literacy, uneven placement in recession shocks.

Would update if: Google partners with ≥10 state workforce boards on AI-displaced-worker placement MOUs with public outcome data.

Conf: M

P = 0.70 — EU Skills Guarantee pilot: institutional design ahead, fiscal scale behind

Claim: ~70% EU Skills Guarantee remains pilot-scale through 2028 (€14.5M automotive supply-chain pilot) — reference-class architecture for job-to-job transitions, not US-equivalent fiscal absorption of digital shock.

Why: Launched Nov 27 2025; full scheme deferred to 2028–2034 MFF / European Competitiveness Fund; 6 transnational projects max; automotive vertical first — parallel to physical-AI wave more than Canaries junior devs.

Evidence:

Analogue: ESF transitional funds post-2008 — effective institutional template, always too small vs shock.

Would update if: Skills Guarantee full scheme budget ≥€5B/yr adopted in 2027 MFF negotiation with digital-displacement vertical.

Conf: M

P = 0.65 — Germany dual-VET + Zusatzqualifikation KI: closest absorption analogue

Claim: ~65% Germany Zusatzqualifikation „KI und maschinelles Lernen“ (IHK modular, ~100 UE, stackable to Berufsspezialist) is the best OECD reference class for “deployment without ASI” — but Mittelstand adoption bottleneck limits US-transplant without chambers/apprenticeship infrastructure.

Why: IHK Beschluss Nov 2024; Darmstadt Rechtsvorschrift Feb 21 2025; four modules (Grundbegriffe, Ethik, Daten, ML); participation doubling YoY per X4 parent (BIBB 2025 theme); on-the-job Betrieb integration beats post-layoff WIOA interface; US lacks federal dual-system.

Evidence:

Analogue: German Meister modular upskilling — proved absorption in industrial transitions 1990s–2000s; 10–15 yr institutional lag to US community-college mimic.

Would update if: ≥5 US states enact IHK-equivalent stackable micro-cred with employer co-design mandate and published placement ≥70% at 12 mo.

Conf: M

P = 0.85 — Acemoglu so-so automation: modal GenAI deploy path (NBER WP 32487)

Claim: ~85% frontier GenAI deployment through C6 follows so-so automation — headcount substitution with ≤0.66% TFP over 10y (~0.064%/yr), not machine-usefulness task creation at scale.

Why: Acemoglu task framework: easy-to-learn tasks automated first; hard context-dependent tasks resist; Goldman/McKinsey upside requires reinstatement channel that reskilling alone cannot supply; “reskilling PR without deployment rules” = so-so lane per X4 parent.

Evidence:

Analogue: Self-checkout — adopted for modest savings, few new cashier-class tasks created.

Would update if: BLS TFP print ≥0.15%/yr sustained 2027–2029 with rising employment in AI-exposed SOCs (reinstatement, not pure substitution).

Conf: M–H

P = 0.25 — Machine-usefulness policy tilt (augmentation procurement / R&D conditioning)

Claim: ~25% US enacts direction policy — procurement human-in-loop, R&D tax credit tied to augmentation metrics, antitrust/labor-power wage share — sufficient to shift deploy mix toward machine usefulness before 2030.

Why: Acemoglu mission-economy lane absent in modal US policy (X4 parent table all Absent/Weak); CA Newsom EO N-6-26 is study not procurement rule; EU AI Act focuses risk not task creation; culture-war menu = vetting not usefulness standards.

Evidence:

  • crosscut x4 education reskilling §Acemoglu — R&D conditioning Absent
  • CA EO N-6-26 PDF — WARN review, not augmentation mandate
  • Power and Progress Acemoglu Johnson — mission economy / pro-worker AI direction
  • Workforce of the Future Act — impact study only, no procurement lever

Analogue: Buy-American manufacturing clauses — procurement as industrial policy, rare and contested for AI.

Would update if: Federal or ≥3 large states require human-in-loop for public-service AI contracts with enforceable penalties.

Conf: L–M

P = 0.90 — PNAS 2025: unguarded AI tutoring causes skill atrophy (−17% post-removal)

Claim: ~90% confidence in causal direction: high-school math RCT — GPT Base (ChatGPT-like) → +48% during access, −17% on unassisted exam vs never-had-AI control; GPT Tutor (Socratic guardrails) mitigates loss (+127% during, neutral after).

Why: ~1,000 student field experiment; published PNAS 2025; mechanism = crutch/offloading not learning; directly maps to AI-literacy curricula without anti-offloading pedagogy → failure branch credential inflation.

Evidence:

Analogue: Calculator in exams — performance aid vs skill formation; proctoring/guardrails determine sign.

Would update if: Independent replication on college CS cohort shows no atrophy with unguarded LLM (would weaken X4 failure-branch load).

Conf: H

P = 0.85 — Grade inflation in AI-exposed courses (+13pp A-share, Chirikov 2026)

Claim: ~85% DiD on 500k+ grades (2018–2025): AI-exposed courses (writing/coding) saw +13pp A-grade share post-ChatGPT (~30% rel. to 2022 baseline) — credential signal degradation, not broad learning gain.

Why: Effect concentrated where homework weight high — consistent with AI substitution not learning; GPA +0.12; complements PNAS atrophy mechanism; TEGL “literacy” without assessment reform → inflated human-capital credentials entering N1 labor market.

Evidence:

Analogue: Grade inflation 1960s–80s — compressed signal, employers shift to portfolios/tests.

Would update if: Proctored-skill assessments show no gap between AI-exposed course grades and independent test scores by 2028.

Conf: H

P = 0.70 — Credential inflation: AI micro-credential wage premium compresses within 24–36 mo

Claim: ~70% supply of generic “AI-ready” badges/certs outruns employer demand for human AI-augmented roles (agents absorb tasks) — wage premium follows coding bootcamp saturation 2016–19 path.

Why: Bootcamp grad share rose 45k+/yr by 2019 while junior market tightened; Google/Amazon certs scale to millions; Indeed/LinkedIn skill-tag proliferation; C6 superhuman coder risk = mid-program obsolescence; German ZQ partial hedge via stackable exit points only.

Evidence:

Analogue: MOOC certificates 2014–16 — high completion badges, low hiring signal within 2 yr.

Would update if: BLS OES shows sustained ≥15% wage premium for “AI-adjacent” SOC codes through 2029 as agent horizon >24h.

Conf: M

P = 0.60 — Corporate/community-college placement rates <50% for displaced mid-skill workers at 12 mo

Claim: ~60% aggregate verified placement (job at ≥80% prior wage within 12 mo) for displaced mid-skill workers in AI-transition programs stays <50% — WIOA headline rates mask cohort mismatch.

Why: WIOA Dislocated Worker ERQ4 national ~69–75% any employment Q4 after exit — not AI-displaced, not wage-retention threshold; self-selected participants; corporate programs report trainings completed not Challenger-cohort outcomes; Goodwill/Google cite historical placement, not AI-shock cohort.

Evidence:

Analogue: Coding bootcamp advertised 80%+ in-field vs independent audits showing 40–60% at 12 mo in downturn cohorts.

Would update if: DOL publishes AI-displaced cohort outcome series with ERQ4 ≥60% at ≥80% prior wage (new metric).

Conf: M (headline WIOA H; AI-displaced subset L–M)

P = 0.15 — Time-to-retrain beats agent capability doubling (success conjunct #4)

Claim: ~15% unconditional ( ~30% conditional on EV-EDU-FED + corporate scale) that credentialed pathways (12–18 mo) re-employ workers faster than agent task frontier doubles (~90 d METR post-2024).

Why: Associate/bootcamp 6–18 mo; WIOA dislocated-worker completion median 8–12 mo; policy design lag 12–24 mo; METR extrapolation → 24h+ by 2027 Q1 P≈0.65; C6 superhuman coder obsolesces 2024-stack curricula mid-program — structural 2–4× gap unless displacement slows (not modal).

Evidence:

  • METR TH1.1 — 88.6-day doubling; Opus 4.6 ~12h horizon
  • node1 §METR ≥24h by 2027-Q1 P≈0.65
  • crosscut x4 education reskilling §Time-to-retrain table — effective gap 2–4×
  • COVID vaccine→rollout lag analogue in parent — 12–24 mo policy implementation

Analogue: Y2K — remediation succeeded because deadline fixed and task bounded; open-ended agent doubling has no freeze date.

Would update if: METR doubling stalls >180 days through 2028 and modular 6-mo employer-co-designed certs show ≥70% placement for Canaries cohort.

Conf: M (METR extrapolation); L (beat-automation conjunct)

P = 0.35 — Corporate training ≥2M US workers with verified placement (conditional on EV-EDU-FED)

Claim: ~35% given EV-EDU-FED fires; ~15% unconditional — at least one corporate+federal stack reaches ≥2M US workers with audited 12-mo placement ≥60% at ≥80% prior wage.

Why: Amazon 500k employee target is internal; Google 1M literacy ≠ placement; requires federal reporting mandate + employer co-design (German Betrieb); without EV-EDU-FED, pure voluntary reporting stays aspirational.

Evidence:

Analogue: LEED certification — scaled only after third-party audit mandate; voluntary corporate pledges plateau.

Would update if: Fortune 100 ≥10 adopt common AI-transition placement reporting standard with third-party audit before 2028.

Conf: L–M

P = 0.25 — Success branch conjunction (U4 path): all four legs (do not multiply)

Claim: ~25% success branch mass if conjunctive requirements met: (1) EV-EDU-FED P≈0.22; (2) corporate ≥2M verified P≈0.35|#1; (3) machine-usefulness P≈0.25; (4) time-to-retrain P≈0.30|#1 — rough ≤0.25 after correlation (shared legislation window).

Why: Legs positively correlated post-C5 shock but direction policy and speed policy partially independent; German EU track satisfies (4) partially but not US fiscal (1); treat as scenario weight not product.

Evidence:

  • crosscut x4 education reskilling §Success branch (~25%) — four-leg table
  • my putopia — U4 modest flourishing definition
  • Y2K $300B remediation — successful mitigation looked like overreaction (parent analogue)

Analogue: Climate 2°C pathways — multiple conditions, correlated but not guaranteed bundle.

Would update if: Any two legs confirmed by 2029-H1 → revise success branch to ≥0.40.

Conf: L–M

P(Δ U4) = +3–5pp — if CX-EDU-RESPONSE wrong (reskilling absorbs shock)

Claim: If X4 success branch materializes, U4 (modest flourishing) rises +3–5pp (EV-EDU-FED alone +4pp per parent); partial U2 distribution without full SWF; N1 culture-war coalition intensity −0.05.

Why: U4 is primary beneficiary — reskilling is policy face of machine usefulness + managed inequality; Acemoglu prefers education/retraining over cash-only UBI for distribution; stable employment ↓ race-populist tail ~0.5pp extinction indirect.

Evidence:

Analogue: G.I. Bill → mid-century mobility — policy face of broadly shared growth episode.

Would update if: U4 operational metrics (leisure+meaning surveys) flat despite placement success → revise uplift to ≤+2pp (GDP ok, agency not).

Conf: M (speculative bucket arithmetic)

P(Δ friction) = −3–5pp — if reskilling works (Ghost GDP eases)

Claim: Successful X4 −3–5pp on friction bucket (53% modal); −4–6pp if full success branch — reduces Ghost GDP + culture-war jobs-frame paralysis (c8 society snapshot by ci).

Why: Friction = modal CX-EDU-RESPONSE holding; employment stabilization closes Canaries gap → macro narrative shifts from “AI stealing jobs” to “managed transition”; Pew 56% job-loss concern decays slowly even on success — floor on friction reduction.

Evidence:

Analogue: Post-WWII reconversion — friction fell with visible placement, not GDP alone.

Would update if: Placement success but Challenger AI-cited cuts unchanged YoY → friction reduction ≤−2pp (narrative decoupled).

Conf: M

P(Δ whimper) = +4–6pp — if X4 fails AND U2 SWF blocked (CX-UBI-SPEENHAMLAND)

Claim: If CX-EDU-RESPONSE holds and U2 SWF/meaning policy blocked, P(whimper) +4–6pp — education failure is the political substitute that didn’t land; employed-but-meaningless + deskilled graduates block U7.

Why: Whimper channel = agency loss without extinction; so-so automation + failed reskilling + no cash/meaning backstop compounds C8–C9; Acemoglu education lane closed → cash-only UBI also blocked in his framework = double bind.

Evidence:

Analogue: Rust Belt 1980s — neither retraining nor safety net at scale → chronic agency loss.

Would update if: UBI/SWF pilot ≥1M participants with meaning metrics stable despite X4 fail → whimper increment ≤+2pp.

Conf: L–M

P = 0.15 — Failure branch: reskilling theater + credential inflation + skill atrophy (mass)

Claim: ~15% of worlds land in explicit failure branch — micro-credentials proliferate, employers cite “we offered training” as liability shield, AI tutoring crutch + grade inflation produce deskilled graduates, whimper lock-in even if U4 GDP ok.

Why: Distinct from modal 60% “activity without absorption” — failure branch requires active harm (atrophy, signal collapse) not mere insufficiency; PNAS + Chirikov establish mechanisms; corporate training without guardrails dominant in TEGL/US K-12 modal.

Evidence:

Analogue: For-profit college boom 2000s — credentials ↑, outcomes ↓, regulatory theater.

Would update if: ≥50% US states mandate anti-offloading pedagogy standards in AI literacy curricula by 2028 → downgrade to ≤8%.

Conf: M

P = 0.05 — X4 direct extinction leverage (±0–1pp)

Claim: X4 resolution moves P(extinction by 2050) ±0–1pp at most — education/reskilling is S_soc friction/utopia lever, not misalignment or CBRN tail.

Why: Mechanism is human-capital and political economy; race narrative +0.5pp indirect only if reskilling failure → populist acceleration; no capability path.

Evidence:

Analogue: TAA — large social policy, zero extinction relevance.

Would update if: (None expected — scope boundary.)

Conf: H

P = 0.55 — EU Union of Skills / AI Skills Academy ahead of US on literacy architecture

Claim: ~55% through C6 the EU AI Skills Academy + micro-credential framework + 2030 Digital Education Roadmap deliver superior literacy architecture vs US TEGL — but 6.2–7M AI workers needed by 2027 (COM/2025/90) remains unfilled at fiscal scale.

Why: COM/2025/90 Union of Skills; AI Skills Academy sectoral training; Mar 2026 Council recommendation on upskilling/VET/STEM — fragmented MS implementation; US lacks equivalent federal skills guarantee scale despite TEGL permissiveness.

Evidence:

Analogue: GDPR → US sectoral patchwork — EU architecture first, US scale later (if ever).

Would update if: US federal AI literacy framework (Feb 2026 AI Literacy Framework) gains mandatory WIOA alignment with EU-equivalent micro-cred stack by 2028.

Conf: M

Falsifiers (master)

ObservationImplication
Canaries gap <5% rel. for 4+ quartersCX-EDU-RESPONSE wrong
Federal >$10B/yr training sustained 3+ yearsEV-EDU-FED fired; CX-EDU-RESPONSE 0.60 → ~0.35
Median displaced re-employment wage ≥85% prior within 18 moAbsorption success
WIOA AI-displaced ERQ4 ≥60% at wage thresholdUpgrade placement P’s
METR doubling >180 days + modular cert placement ≥70%Time-to-retrain leg plausible
Proctored skills match inflated AI-exposed gradesCredential inflation overstated
Whimper stable despite X4 fail + UBI pilot ≥1MSWF decouples whimper coupling

Source index

ResourceURL / path
X4 parent(internal note)
Node 1 labor(internal note)
Acemoglu WP 32487https://www.nber.org/papers/w32487
METR TH1.1https://metr.org/blog/2026-1-29-time-horizon-1-1/
TEGL 03-25https://www.dol.gov/agencies/eta/advisories/tegl-03-25
S.3319https://www.congress.gov/bill/119th-congress/senate-bill/3319/text
MASA / FY26https://www.nawb.org/3752-2/
Amazon Future Readyhttps://www.aboutamazon.com/impact/amazon-future-ready
Google $1B higher-edhttps://www.reuters.com/world/us/google-commits-1-billion-ai-training-us-universities-2025-08-06/
EU Skills Guaranteehttps://employment-social-affairs.ec.europa.eu/news/commission-launches-skills-guarantee-support-workers-transition-learning-new-skills-strategic-2025-11-27_en
Germany ZQ KIhttps://www.ihk.de/darmstadt/produktmarken/aus-und-weiterbildung-channel/ausbildung-channel/rechtsvorschrift-zusatzqulifikation-ki-6494754
PNAS tutoringhttps://www.pnas.org/doi/10.1073/pnas.2422633122
Grade inflationhttps://cshe.berkeley.edu/publications/artificial-intelligence-and-grade-inflation-cshe-higher-education-working-paper-series
WIOA performancehttps://www.dol.gov/agencies/eta/performance/wioa-performance
Crux registry(internal note)
U4 / utopia(internal note)

Update log

DateChange
2026-07-05Initial X4 evidence rationale — 25 P sections + TL;DR