System Documentation

How the system works.

The complete research pipeline — from source scanning to publication. Every stage is documented, every decision rule is explicit, every metric is real.

01Source Universe→
02Extraction→
03Scoring→
04Red-Teaming→
05Ranking→
06Publication→
07Update Cycles
01

Source Universe

What we scan and why

Continuously monitored

The system monitors a curated universe of sources across academic research, field operations, and philanthropic funding flows.

  • Academic papers from PubMed, Google Scholar, and SSRN — filtered for health economics, development, and cost-effectiveness
  • NGO field reports from WHO, UNICEF, MSF, GiveWell, and regional health authorities
  • Philanthropic funding databases tracking foundation grants, government aid flows, and bilateral commitments

Status: Updated as new papers and reports are published. 32 primary sources cited in the Q1 2026 Impact Index.

02

Extraction

How candidates are identified

Structured candidates

Extraction agents process the source universe and identify candidate interventions — structured, comparable units of philanthropic opportunity.

  • Each candidate is structured: problem domain, intervention mechanism, evidence quality, estimated cost, scale of impact, and implementation context
  • Natural language processing identifies interventions that are both effective AND underfunded — the intersection that defines leverage
  • Deduplication and cross-referencing ensure no intervention is double-counted or conflated with adjacent programs

Status: 10 opportunities selected and published in the inaugural cycle (March 2026)

03

Scoring

The INT Framework

3-dimensional assessment

Every candidate is scored on three dimensions using the INT (Importance × Neglectedness × Tractability) framework — the standard analytical lens in effective altruism research.

  • Importance: measured in DALYs averted per dollar, lives saved, or welfare-adjusted equivalents. Sources must be peer-reviewed or from established field data.
  • Neglectedness: ratio of current funding to estimated funding need. Cross-referenced against philanthropic databases, government commitments, and NGO spending reports.
  • Tractability: evidence of cost-effectiveness at scale, implementation feasibility, political viability, and organizational capacity in the field.

Status: Each dimension scored on a standardized 0–100 scale. Composite INT score computed as weighted geometric mean.

04

Red-Teaming

Adversarial challenge

Adversarial review

A dedicated adversarial agent challenges the evidence base, cost estimates, and tractability assumptions for every scored candidate. This is not a formality — it is the mechanism that prevents overconfidence.

  • Evidence test: Are the core studies peer-reviewed? Are effect sizes replicated across contexts? Are there significant methodological concerns?
  • Neglectedness test: Is the funding gap real, or is it filled by unreported government spending or private philanthropy?
  • Tractability test: Are there implementation bottlenecks — political barriers, supply chain constraints, organizational capacity limits — that the scoring missed?

Status: Candidates are rejected for insufficient evidence, overestimated neglectedness, or low tractability. Each rejection is logged.

05

Ranking

Final ordering

Top 10 selected

Surviving candidates are ordered by their composite INT score, adjusted for uncertainty. The ranking reflects expected impact per marginal dollar under honest uncertainty bounds.

  • Uncertainty-Adjusted Expected Value (UAEV): scores are discounted by the width of their confidence intervals. High-uncertainty candidates are penalized.
  • Sensitivity analysis: rankings are tested against plausible variations in key parameters. Stable rankings are flagged as high-confidence.
  • The final top 10 represents the best available opportunities for high-leverage giving under current evidence.

Status: 10 opportunities ranked in Q1 2026 inaugural edition

06

Publication

Open access release

Full methodology disclosed

Findings are published as the Impact Index — a free, open-access ranking of high-leverage giving opportunities. All methodology, scoring rationale, and uncertainty estimates are disclosed.

  • Impact Index: quarterly-updated ranking with summary analysis for each opportunity
  • Deep Dive reports: detailed research for the most impactful candidates, with full cost-effectiveness models
  • All research published under CC BY 4.0 — freely available for any use

Status: Q1 2026 Impact Index published with 10 ranked opportunities

07

Update Cycles

Continuous improvement

Triggered by data + schedule

The system does not publish once and freeze. It operates in cycles — re-scoring existing candidates, incorporating new evidence, and refining its own methodology.

  • Data-triggered updates: new WHO reports, major study publications, or funding announcements trigger immediate re-scoring of affected candidates
  • Scheduled cycles: quarterly comprehensive re-analysis of the full candidate pipeline, including new source scans
  • System improvement: after each cycle, the pipeline itself is evaluated — scoring heuristics, extraction accuracy, and red-teaming thoroughness are optimized

Status: Inaugural cycle completed March 2026. Next scheduled update: Q2 2026.

Allocation Rules

How resources are allocated.

The system allocates resources according to explicit, parametric rules. These rules are not editorial — they are programmatic constraints that govern every allocation decision.

Rule 1

Research receives ≥ 40% of all resources

Scanning, scoring, deep analysis, and candidate evaluation always receive the largest share. Research is the core function.

Rule 2

Communication receives ≤ 25% when publication is ready

Outreach, website deployment, and amplification are funded only when there are research findings to communicate.

Rule 3

System improvement receives ≤ 20% after each cycle

Pipeline optimization, scoring refinement, and infrastructure improvements are funded from operating surplus.

Rule 4

Reserve of ≥ 15% maintained at all times

A reserve is held for deep investigations, unexpected opportunities, and ensuring institutional continuity.

These rules are parametric and can be adjusted as the system evolves. All changes are logged in the lab log.

The system is transparent by design. Explore more of the institution.

Open Architecture →Public Ledger →Lab Log →