🖥️ Leverage analysis

Binding constraint is regulation (0.4) --- FDA/EMA trial pathways and LMIC regulatory harmonization, not compute or grid. Capex (0.55) is the second bind: neglected-disease economics don't clear private hurdle rates, so the pipeline needs either AMCs, PDPs, or public-balance-sheet underwriting to bridge from candidate to approved therapeutic. Suffering numerator looks right: disease 0.85 and mortality 0.7 are the load-bearing layers and this intervention touches them directly via AMR and neglected infectious disease. Mental health 0.15 is generous --- psychiatric drug discovery has been a graveyard for decades and AI hasn't changed that yet. Poverty 0.3 is downstream-only (healthy people work) and slightly overcredited.

Binding constraint is grid (0.4), full stop --- interconnection queues and transmission are the physical ceiling regardless of how much capex or enterprise demand exists. Regulation (0.7) and public (0.7) are softer. The suffering_reduction_scores are the shakiest in the whole ranking: compute is an enabler, not a terminal suffering-reducer. Crediting it 0.4 disease and 0.4 mortality assumes the compute actually gets pointed at drug discovery and clinical deployment rather than ad targeting and coding copilots, which is precisely the reallocation problem the manifesto exists to solve. The composite is double-counting the suffering layer that intv_drug_discovery already books.

Binding constraint is regulation (0.45) intertwined with liability --- a chatbot that misses suicidal ideation is a headline and a lawsuit, and no health system wants to own that risk without explicit regulatory shelter. Public (0.5) is the second bind: therapist-guild resistance and patient trust both degrade uptake. Enterprise (0.55) maps to payer and public-health-system procurement cycles, which are glacial. Suffering numerator is directionally right --- mental_health 0.7 is the correct primary layer and the ~70% untreated global burden is real --- but mortality 0.3 probably overstates the suicide-prevention evidence base, which is thin. This is the intervention where the camp coalition is wider than the ranking implies.

Binding constraint is regulation (0.3) --- NEPA, state PUCs, interconnection queues, transmission siting. This is the single lowest friction score in the entire ranking and it is correctly placed: permitting is the chokepoint for every physical-layer AI scale-up in the US. Suffering_reduction_scores are honest about grid being an enabler not a terminal intervention --- all scores ≤0.3 --- but the same enabler-double-counting critique from intv_compute applies in reverse: grid scores lower than compute on suffering despite being upstream of it, which is inconsistent.

Essentially no external friction --- this is the highest mean_robustness in the ranking (0.82) and correctly so; alignment research has enterprise buy-in, public sympathy, regulatory tailwind, and zero grid load. The binding constraint is internal: talent depth and research-agenda tractability, neither of which the friction model captures. Suffering_reduction_scores are low (max 0.2) and that is actually a category error --- alignment is insurance against tail catastrophe, not a per-DALY intervention, and scoring it on the same suffering axes as drug discovery systematically underrates it. Either it belongs in a different numerator or the composite should treat it as a multiplier on every other intervention's expected value.

Binding constraint is public (0.35) --- consumer acceptance of cultivated meat is the chokepoint, compounded by active state-level bans (FL, AL) that the friction_regulation=0.5 understates. Capex (0.5) is real: bioreactor scale-up economics still haven't closed parity with commodity meat. Suffering numerator is the most honest in the ranking: animal 0.85 is the load-bearing term and it is the one layer no other intervention touches. This is the only intervention in the ranking that scores the animal-suffering numerator at civilizational scale, and the composite ranks it sixth. That is a coalition-breadth artifact, not a suffering-magnitude result.

No hard friction --- high robustness (0.76) reflects broad political support. Binding constraint is effectiveness: the empirical track record of retraining programs (TAA, WIOA) is poor, and labor-market absorption is the real bottleneck, which the friction model doesn't represent. Suffering_reduction_scores are appropriately low because training doesn't touch disease, mortality, or animal layers; the 0.3 poverty credit is defensible if you believe retraining moves displaced workers back to comparable wages, which the evidence does not strongly support. Correctly ranked last.

DoD obligated AI-related contract spending rose substantially 2022-2025, driven by JWCC, Project Maven, and CDAO-managed pilots; precise totals are hampered by inconsistent AI tagging on contract line items.

No other pure-play US defense-AI software vendor has matched Palantir's contract backlog or combatant-command integration depth; cloud-provider primes (AWS, Microsoft, Google, Oracle via JWCC) supply infrastructure, not mission-software integration.

Credible 2030 forecasts for US datacenter share of electricity consumption diverge by more than 2x --- from ~4.6% (IEA/EPRI conservative) to ~9% (Goldman Sachs, EPRI high scenario) --- reflecting genuine uncertainty, not measurement error.

Frontier-lab and big-tech employees have episodically resisted DoD contracts (Google Maven 2018, Microsoft IVAS 2019, Microsoft/OpenAI IDF deployments 2024), producing temporary pauses but no sustained shift in vendor willingness.