How to read auditor flag tiers

Detector tier glossary

Every flag on this site carries a tier label — A through E1. The letters classify what kind of evidence the detector surfaces, not how severe the implied wrongdoing is. A Tier A flag and a Tier D flag describe different layers of the same possible problem, not different severities of fault.

Within a tier, severity ranking happens via the per-detector score (Benford chi-square, HHI, money amounts, etc.). Across tiers, the right composite is "stack of evidence at increasing depth": Tier A says outlier, Tier B says relationship-with, Tier C says timing-aligned, Tier D says all-three-stacked.

Tier A — Statistical anomaly

Tier A Statistical anomaly

An entity stands out as a numerical outlier inside a single dataset.

Detectors in this tier

Benford's Law deviation. First-digit distribution of contribution amounts deviates from the calibrated NV baseline (Tam Cho & Gaines 2007).
Donor-base concentration (HHI). Single-donor share of a recipient's contribution base above the U.S. DOJ antitrust 'highly concentrated' threshold of 0.25.
Isolation Forest outlier. Per-entity feature vector lies far from the bulk of its peers (Liu, Ting, Zhou 2008).

What it means

An entity is statistically unusual relative to its peers. By itself this is not evidence of wrongdoing — outliers exist in every distribution. Tier A is the starting point: an entity worth looking at more closely.

Tier B — Relationship anomaly

Tier B Relationship anomaly

Two or more entities share a structural relationship that creates dependence or hidden alignment.

Detectors in this tier

Shared-officer / shared-agent clustering. An officer or registered agent serves N+ politically-active orgs (Olson 1965 coalition-shell signal).
PAC operators. One person registered as the contact-of-record for many different PACs.
Address co-occurrence. Multiple distinct entities at the same physical address — apparent independence is a paperwork artifact.
Lobbyist-as-PERS-retiree. Currently-registered lobbyist is also drawing a NV PERS pension.
IRS 990 cross-org officer. One person serves as officer/director on N+ federal-level 501(c) entities.

What it means

An entity sits in a relationship structure that creates a coordination problem. The structure itself is sometimes legal (a 501c3 and its 501c4 sister org sharing officers, for example, is a normal pattern) — Tier B surfaces the relationship for human assessment of whether it raises a conflict.

Tier C — Temporal correlation ("shadow work")

Tier C Temporal correlation ("shadow work")

Events that align in time are doing more than the public record makes obvious.

Detectors in this tier

Bill-timing correlation. A client newly registers as a lobbyist client within N days of a specific bill drop.
Contribution-spike change-point. PELT change-point detection on weekly contribution flow per recipient (Killick, Fearnhead, Eckley 2012).
Shell-entity formation timing. A nonprofit formed within N days of a major bill, with a politically-active first activity.
Cross-state coordination. NV-targeted PAC registered to a national parent org (RGA, DGA, EMILYs List, etc.).
Arabella / fiscal-sponsor pass-through. Out-of-state 501(c)(3)/(c)(4) hub grants money into a NV recipient.

What it means

Two signals that look unrelated in isolation are actually coupled. A contribution timed against a vote, a PAC formed against a bill, a hub grant arriving as state legislation moves — the timing IS the signal. Tier C calls these out without claiming intent.

Tier D — Chained pattern

Tier D Chained pattern

Multiple distinct relationships line up in a way that produces a documented sequence of events.

Detectors in this tier

Quid-pro-quo signature (D1). Money OUT to lobbyist + L represents C + C money IN to R + R sponsored/voted-Yea on bill B benefitting C. Each link innocent in isolation; the bundled correlation is the signal.
Recusal failure (D2). Legislator voted on a matter despite a known business or contribution relationship to the affected party (NRS 281A.420).
Adverse interest (D3). Same lobbyist represents both a regulator and the regulated industry.
Cleanup-of-evidence (D4). Entity changes registered agent / formation status / officer roster within N days of being publicly flagged.
Contribution-to-access ratio (D5). Contribution amount disconnected from documented testimony presence — access through non-money channels.
Triangulation-on-action (D6). Campaign contribution + lobbyist registration + bill drop + agency rule change all within N days.

What it means

The strongest leads. Tier D entries connect multiple data points into a documented sequence-of-events. They still require human verification before any allegation, but they are the most-actionable starting points for journalist or litigation work.

Tier E1 — Disclosed-vs-actual gap

Tier E1 Disclosed-vs-actual gap

What the entity reported on a public filing differs from what other public records show.

Detectors in this tier

Lobbyist compensation gap. NRS 218H disclosure of compensation differs from cf_contribution flow visible in the same window.
Annotation-vs-NELIS verification. A site annotation claiming a bill position doesn't match the NELIS exhibit record.
IRS 990 / NV SOS / LDA consistency. Officer rosters or revenue figures differ across federal vs state filings for the same entity.

What it means

A legally-required disclosure differs from another legally-required disclosure or from observed activity. This can indicate (a) genuine error, (b) reporting-period misalignment, (c) intentional understatement. Each finding is a research lead — read the underlying filings to assess.

Why these tiers, not severity letters?

The traditional severity model (high / medium / low) doesn't translate well to corruption research. A "high severity" Benford anomaly in isolation is meaningless. A "low severity" recusal-failure that turns out to be the smoking gun in a vote-buying chain is consequential. The tier model classifies by detector type so readers can compose evidence across tiers — most strongly when an entity appears in 3+ tiers simultaneously.

The composite "most-flagged" leaderboards (/most-flagged/, /politicians/most-devious/) implement that composition: an entity flagged at Tier A AND Tier B AND Tier D outranks an entity flagged at Tier D alone, even if the single Tier D severity score is higher.

Evidence-strength sub-tier (Direct / Indirect / Inference)

Within a tier, individual findings are also tagged with an evidence-strength sub-tier (Paul Mitchell research methodology):

Direct — verifiable against a single authoritative public filing.
Indirect — supported by cross-referenced public records but no single filing carries the full claim.
Inference — pattern strongly suggests the relationship; corroboration would require additional records.

Verification policy

Per the standing site policy: every flag is a research lead, not an adjudicated claim. Each row links to the per-entity page where the evidence chain + a "submit correction" form is one click away.