No Halo methodology · Version 1.1
A conclusion you can inspect.
No Halo is designed as a fact-based adversary. We look hard for material problems, contradictory evidence, weak claims, and reasons a donation may not accomplish what a donor expects. We do not create an issue merely because we went looking for one. Strong evidence is reported as plainly as adverse evidence.
The automated public report is provisional and is not a human-led forensic investigation. Every material conclusion should be traceable to the evidence used to reach it. Missing evidence is not converted into wrongdoing, and a search that finds no adverse record is not represented as legal or regulatory clearance.
The six donor questions
A report is not complete merely because No Halo collected a Form 990. The research system is instructed to answer six decision questions before presenting its conclusion:
- What does the organization actually do with the money? We look for the operating model, principal programs, grant flows, geography, and who ultimately benefits.
- Is it financially healthy, and is capital being deployed rationally? Financial resilience is separated from the question of how much capital is committed, liquid, restricted, accumulated, or available for new work.
- Does the principal work actually work? We look for program-specific outcome evidence, evaluation design, independence, directness of the outcome, and whether historical evidence still applies to the current program.
- Do the major claims and numbers reconcile? Comparable figures and claims are tested across filings, audited statements, annual reports, grant schedules, official claims, government records, independent evidence, and authoritative guidance when available.
- What material concerns did the research uncover? The system actively looks for audit findings, related-party issues, unusual transactions, regulatory or legal records, fundraising concerns, governance failures, and other decision-relevant evidence. Allegations, proceedings, settlements, dismissals, judgments, and findings are kept distinct.
- What would another donated dollar accomplish? We look for current commitments, expansion plans, reserves, funding constraints, operating capacity, and other evidence relevant to whether more unrestricted funding can be used productively now.
An unanswered question may survive into the report only after the available research pass failed to establish an answer. The report should state what was checked and why the issue remains unresolved.
Reading the assessments
Strong: 80–100. Moderate: 50–79. Weak: 0–49. A dimension score is the percentage of available rubric points earned, subject to that dimension’s minimum evidence requirement. It must be read together with check coverage and evidence confidence. Missing values are not converted to zero.
Evidence confidence is separate. It describes the support for the assessment, not a probability that the charity is trustworthy or fraudulent. A favorable score with limited confidence means exactly that: the available checks were favorable, but the evidence base remains limited.
Financial position and resilience
The current financial rubric is labeled Financial stewardship in the report for continuity, but its automated score should be read more narrowly: it measures reported financial position and resilience. It does not establish spending quality, impact, capital efficiency, or whether another donation is needed.
The rubric applies only to a validated annual Form 990 or 990-EZ ending within 24 months, with at least three of four standard checks calculable. Short periods, private-foundation forms, and returns where noncash gifts represent at least 30% of contributions do not receive the standard score.
| Check | Full points: 25 | Partial points: 12 | No points |
|---|---|---|---|
| Net assets | Above zero | Exactly zero | Below zero |
| Liabilities / assets | 75% or less | Above 75%, through 100% | Above 100% |
| Combined balance across two comparable annual periods | Combined revenue covers combined expenses | Combined deficit no greater than 5% of revenue | Deficit greater than 5% |
| Unrestricted net assets / annual expenses | At least 3 months | At least 1 month, below 3 | Below 1 month |
When only three checks are calculable, the percentage is normalized to those three checks and the report displays that coverage. Large unrestricted reserves can indicate resilience while also creating a separate marginal-funding question. No Halo therefore asks what commitments, expansion plans, liquidity needs, or capacity constraints explain the capital before drawing a donor conclusion.
Governance disclosures
At least six of eight standard checks must be available from a recent validated Form 990. A disclosed “yes” earns the listed points; “no” earns zero. An unanswered or unextracted question is unavailable.
- Written conflict policy: 15 points; monitoring that policy: 15.
- Whistleblower policy, document retention policy, return provided to board, board minutes, and CEO compensation-setting process: 10 points each.
- Reported independent voting board: 20 points at two thirds or more; 10 points above one half; otherwise zero.
This score measures reported safeguards. It does not prove the policies operated effectively. Related organizations, interested-person transactions, appointment rights, audit findings, regulator records, and actual conduct can materially change the broader governance conclusion.
Transparency
The current transparency rubric measures access to evidence within the stated collection scope using four standard checks: a recent validated filing (30 points), at least three distinct financial periods (20), accessible official disclosures (20), and an additional identity-matched financial document (30). At least 80 possible points must be supported before a score is assigned.
A failed website download, inaccessible PDF, unsearched source, or unmatched document never earns a negative transparency judgment. This is an observed evidence-access score, not an exhaustive judgment about openness.
Impact evidence
No Halo does not infer impact from program spending, beneficiary counts, mission statements, or fundraising claims. When suitable program-specific evidence is retrieved and independently verified against the source text, the impact-evidence rubric can now be scored.
| Evidence property | Maximum points | What receives the strongest weight |
|---|---|---|
| Independence | 25 | Evidence produced independently of the charity rather than only self-reported. |
| Evaluation design | 25 | Randomized or strong quasi-experimental designs receive more weight than observational, before-and-after, or self-reported evidence. |
| Outcome directness | 20 | Meaningful beneficiary outcomes receive more weight than intermediate measures or activity counts. |
| Applicability to the current program | 30 | Evidence that clearly applies to the program as it operates today receives more weight than an older study whose current applicability is uncertain. |
At least 70 possible points must be supported before an impact-evidence score is assigned. An excellent historical randomized study can therefore establish meaningful evidence while still receiving a lower current assessment when the program has changed or scaled and current fidelity has not been established. The score grades the evidence for the principal outcomes, not the moral worth of the mission and not every program the organization operates.
Reconciliation before accusation
A difference between two numbers is not automatically a discrepancy, and a discrepancy is not automatically misconduct. No Halo first asks whether the sources cover the same entity, period, definition, accounting basis, geography, population, and scope. When authoritative evidence explains the difference, the report says so. When a plausible explanation exists but the evidence does not establish it, the report says that too. When comparable primary sources genuinely remain in conflict, the disagreement stays visible rather than being silently resolved in favor of one source.
Independent research and current evidence
In addition to IRS filings, official disclosures, and matched Federal Audit Clearinghouse records, the current research system can use live web discovery to look for independent evaluations, authoritative accounting guidance, government and regulator records, court records, credible independent reporting, and current evidence relevant to the six donor questions. Candidate web sources must come from URLs actually consulted during the search and must pass source and identity checks before entering the evidence set.
Independent web research is bounded, not exhaustive. State systems, courts, news archives, and specialized databases differ in coverage and accessibility. A search that does not find an adverse record does not establish that no such record exists. Higher-risk legal, regulatory, and identity-sensitive evidence remains subject to conservative matching and human review when ambiguity could change the conclusion.
Evidence freshness
Tax filings often lag current operations. No Halo therefore separates the latest filed financial period from the date through which the current research pass searched other public evidence. A report may properly use FY2024 or FY2025 filed financials while also incorporating newer annual reports, leadership information, impact evidence, regulatory records, or program developments. Those different clocks are shown separately.
AI analysis and verification
The automated report uses one model to synthesize the collected evidence and a separate verification model to try to break the draft conclusions. Published investigative answers and findings must use source text that can be validated against the retrieved evidence. The verifier is instructed to reject unsupported negative inference, invented reconciliations, allegations presented as findings, compensation judgments without a benchmark, impact inferred from spending, stale studies presented as current proof, and unresolved questions that the collected evidence already answers.
This is still automated research. Model verification is not human review and cannot guarantee factual perfection. Material ambiguity can trigger a human-review state without erasing unrelated conclusions that remain well supported.
Evidence and important limits
- Filings are matched by EIN and amendments are preserved. Identity conflicts involving an official organization tax ID block substantive conclusions until resolved.
- Official-site collection is bounded. Scanned, encrypted, very large, blocked, or inaccessible documents can remain unread.
- Federal audit evidence is matched conservatively. A missing FAC match does not establish the absence of an audit or findings.
- Legal and regulatory evidence distinguishes allegations, proceedings, findings, settlements, dismissals, judgments, and final dispositions where that information is available.
- Same-name matches alone are not sufficient for adverse conclusions.
- Missing evidence is a coverage limitation. It is not evidence of misconduct and is not scored as a failure unless the rubric specifically measures a disclosed negative response.
- Reports are preserved as dated versions. A methodology revision produces a new report specification so older conclusions can be refreshed against the newer research process.
Corrections and evolution
Every report links its evidence and provides a correction path. No Halo will continue calibrating the methodology across causes, operating models, organization sizes, and real investigations. Paying for research does not buy a higher rating, favorable conclusion, or suppression of a material finding.