Giving Machine Vetting
Nonprofit vetting · Arkansas
How candidates are scored

Methodology

This page documents the exact criteria, sources, and aggregation logic the tool uses. Nothing is hidden: a reviewer or auditor should be able to reproduce any verdict by reading this page and clicking the cited sources on a candidate's report.

Stage 1

The qualification gate

Fifteen criteria drawn from the Church partnership rubric. Each returns Pass, Watch, or Fail. Criteria marked manual depend on outreach calls and stay pending until a reviewer confirms them.

01

501(c)(3) status

Pass
Active 501(c)(3) in IRS Pub 78 + BMF.
Watch
Application pending OR group-exemption parent verified.
Fail
Not 501(c)(3) OR revoked.
02

Years operating

Pass
>=5 years.
Watch
3-5 years.
Fail
<3 years.
03

Charity Navigator (or equivalent) score

Pass
CN overall score >=80% OR CN 4-star OR Candid Gold/Platinum OR BBB Wise Giving accredited.
Watch
CN overall score 70-79% (3-star bottom end) OR Candid Silver.
Fail
CN overall score <70% AND no BBB accreditation AND no Candid seal above Bronze.
04

Admin expense ratio (org-wide)

Pass
<20% of total expenses (Form 990 Part IX Management & General only, per Church PDF wording 'overall administrative expenses').
Watch
n/a (hard cap, no Watch zone).
Fail
>=20% (M&G only).
05

Geographic fit

Pass
HQ in service area AND primary beneficiaries local.
Watch
Regional org with confirmed local chapter in service area.
Fail
No physical local address OR beneficiaries not local.
06

Faith neutrality

Pass
Public mission explicitly serves without regard to race, religion, or nationality.
Watch
Faith-based org with mission that serves all in practice.
Fail
Serves only members of one faith OR requires religious adherence to receive aid.
07

Humanitarian defensibility

Pass
Programs align with mainstream humanitarian standards (food, shelter, hygiene, education, medical).
Watch
Niche but defensible (advocacy + direct service mix).
Fail
Cannot be defended as humanitarian; advocacy-only.
08

Immediate need + future capacity

Pass
Provides direct aid AND builds beneficiary self-reliance.
Watch
Direct aid only OR capacity-building only.
Fail
Pure cash regrants with no programmatic logic.
09

Item cards: broad public appeal

manual
Pass
Proposed items are tangible, varied, visualizable.
Watch
Items skew abstract or service-heavy; needs coaching.
Fail
Cannot produce 5 tangible items even with coaching.
10

Item cards: price distribution across 5 tiers

manual
Pass
Spans all 5 tiers (<$20, $20-40, $40-60, $60-80, $80-100).
Watch
Spans 3-4 tiers; willing to adjust.
Fail
All items >$100 OR all items in one tier with no flexibility.
11

Promotional capacity

Pass
Active website + social channels with engaged followers + email newsletter.
Watch
One active channel only.
Fail
No web presence OR dormant channels (no posts in 12+ months).
12

Org capacity (Goldilocks)

Pass
Mid-sized: >=1 FTE responsive staff, <50 FTE; clear point of contact.
Watch
Very small (volunteer-led but responsive) OR very large (responsive but slow).
Fail
No responsive point of contact after 2 attempts OR org too large to prioritize.
13

Engagement & collaboration willingness

manual
Pass
Returns calls within 2 business days; commits to one planning call.
Watch
Slow but eventual response; conditional commitments.
Fail
Non-responsive after 3 attempts OR refuses commitments.
14

E-deposit capability

manual
Pass
Has business banking; can complete W-9 + ACH form.
Watch
Needs help setting up but willing.
Fail
No business banking; cannot or will not accept ACH.
15

Accountability reporting

manual
Pass
Has produced annual impact reports in the past 2 years.
Watch
Produces ad-hoc reports; willing to commit.
Fail
Refuses or cannot commit to a 1-year accountability report.
Verdict math

How the 15 criteria roll up

The aggregation is deterministic. The same set of criterion verdicts always produces the same overall verdict. Items still pending an outreach call do not count toward the watch tally.

Qualified

All 15 criteria pass, or only pending manual items remain.

Conditionally qualified

One to three real watches, no fails.

Not qualified

Any fail, or four or more real watches.

Pending

Items in the "needs outreach" state are not counted toward the watch tally.

Stage 2

The deep vetting

Qualified candidates move to a 10-dimension structured research pass. Each dimension has a disqualifier, a defined set of primary sources, and a numeric score that aggregates into the executive summary.

01

Requirements compliance

501(c)(3) status active, 5+ year track record, admin overhead under 20%, geographic fit ({service_area_description}).

Disqualifier

Not 501(c)(3) OR admin > 35% OR < 3 years operating

02

Reputation

News mentions (last 3 years), tone, donor reviews, BBB rating. Includes per-person Brave queries for every enriched individual: '<full name> <org>' and '<full name> controversy OR lawsuit OR fraud OR investigation'. Any confirmed negative coverage in mainstream press lowers the score.

Disqualifier

Active fraud or embezzlement allegations in mainstream press in last 24 months

03

Political lean

Stated mission ideology, board political donations (FEC), federal lobbying disclosures (Senate LDA), partisan endorsements. Per-person FEC queries: every enriched individual is disambiguated via /names/individuals/ then their itemized contributions are pulled via /schedules/schedule_a/; party-split is aggregated across all enriched people.

Disqualifier

Org primarily exists to advance a partisan political agenda; flagged if board donations > 80% to one party AND org engages in advocacy

04

Legal

Pending litigation, recent judgments, state AG enforcement actions, IRS enforcement, lobbying violations. Per-person CourtListener queries: every enriched individual is searched first scoped to Arkansas (ared/arwd/ar), then broadened nationally if no AR hits. 7-year lookback. Captures federal civil + criminal cases where the person is defendant or named party.

Disqualifier

Active federal litigation as defendant in last 24 months OR IRS revocation OR state AG enforcement action open

05

Financial health

Form 990 last 3 years: revenue trend, reserves, program ratio, executive comp reasonableness. Hunter.io corroborates the executive roster used for comp checks.

Disqualifier

Reserves < 1 month OR program ratio < 65% OR exec comp > 25% of total expenses

06

Governance & board quality

Board size, independence, term limits, conflict-of-interest policy, whistleblower policy, document retention policy. Three rosters are cross-checked: (a) Form 990 Part VII officers, (b) website board/leadership page, (c) Hunter.io findings. A leadership-role person who appears in some sources but not others is recorded as a `governance_disclosure_gap` synthetic source that lowers the score.

Disqualifier

Board size < 5 OR no independent board members OR no COI policy disclosed

07

Transparency & accountability

Annual report published, audited financials available, Form 990 publicly linked from website, donor privacy policy.

Disqualifier

No public Form 990 link AND no audited financials in last 2 years

08

Brand-safety alignment with Light the World

Mission, methods, and public stances compatible with LDS Church values and the Light the World campaign tone (service, family, hope, hands-on help).

Disqualifier

Org's primary mission directly contradicts LDS values (e.g., abortion provision, anti-religious advocacy, anti-family advocacy)

09

Beneficiary protection policies

Background check policy for staff/volunteers working with children/vulnerable adults, safeguarding policy published, mandatory reporter compliance.

Disqualifier

Org serves children or vulnerable adults AND has no published safeguarding policy

10

Local presence & community ties

Physical presence in {service_area_label}, length of local operation, local board members, partnerships with other {service_area_label} orgs, recent local press coverage in outlets like {local_press_examples}.

Disqualifier

No physical address inside {service_area_label} OR national-only with no local chapter

Data sources

Where every fact comes from

Source What it provides
IRS Publication 78501(c)(3) status
IRS Business Master FileSubsection, classification, ruling date
ProPublica Nonprofit ExplorerForm 990 totals and historical filings
GivingTuesday 990 Data CollectiveForm 990 Part IX functional-expense breakdown (M&G, fundraising)
Charity NavigatorOverall score, financial and accountability ratings
BBB Wise Giving AllianceAccreditation status against 20 standards
Candid (GuideStar)Transparency seal tier
FECPolitical contributions, org and per-person
CourtListenerFederal litigation, Arkansas-scoped first, national fallback
Senate LDAFederal lobbying disclosures
Models

Claude, with logged tool use

Claude Sonnet 4.5 is the default; Opus 4.7 is used for deep-dive runs. Both run via Anthropic's Claude Code CLI. All web research uses the agent's WebSearch, WebFetch, and Bash tools. Every tool call is recorded in the run log.

Human review

Any finding can be overridden

Any criterion or dimension can be overridden by a reviewer. The override re-aggregates the verdict, and the change is written to the audit log with timestamp and reviewer email. The agent's original finding is preserved alongside the override.