Under what conditions could algorithmic tax-audit leads improve selection while allowing officials and affected people to detect errors and contest targeting on illegitimate grounds?

Contribution to the handbook

Priority · Assessed 2026-10-30

This is a concrete executive design and testing question with consequential effects on people targeted for audits. It compares algorithmic leads with an appropriate baseline and asks whether officials and affected people can detect and contest errors. The tax-audit benefit is proposed, not demonstrated, and leads are not automated enforcement decisions (draft-proposal-3-cafba6b29de3; s-danaher-threat-of-algocracy-7eb9f04cb824f91b; s-danaher-threat-of-algocracy-73c1645a3756bf8d).

Does this challenge the goal?

In-scope gap

The question seeks comparison, legitimate targeting criteria and effective review within the existing executive and accountability scope; it does not contest the goal or desired deliverable.

The appropriate grounds, review arrangements and acceptance thresholds remain open.

Applying laws and acting →

Authority, accountability and resilience →

Judging situations and resolving disputes →

Tests, stop conditions and unresolved work →

Compare tax-audit lead selection, accuracy, and errors with an appropriate non-algorithmic or existing selection process; do not infer a tax-specific gain from prediction results in other fields.

What we can say so far

Danaher proposes that algorithms could identify audit leads by finding links across financial datasets that individual staff may miss, particularly where staff resources are limited. The supplied evidence does not establish whether this would improve tax-audit selection or reduce errors compared with an existing selection process. Reported prediction advantages in other fields cannot establish a tax-specific gain.

What remains open

• A comparison with an appropriate existing or non-algorithmic tax-audit selection process.
• Tax-specific evidence on selection accuracy and the types and rates of errors.

Specify which grounds for selecting tax-audit targets would be legitimate, which would be arbitrary or discriminatory, and how those choices should be authorized.

What we can say so far

A candidate ground for a tax-audit lead is a genuinely higher likelihood of the relevant tax noncompliance, rather than arbitrary or discriminatory targeting. That general distinction does not establish which tax-specific indicators are legitimate, or whether predictive accuracy alone would justify using them. The supplied perspective also calls for independent oversight and safeguards against capture, but does not specify who should authorize selection criteria or how.

What remains open

• Which specific tax-audit indicators, including potentially discriminatory proxies, should be permitted or prohibited, and under what conditions.
• How predictive relevance should be weighed against discrimination, fairness and participation.
• Which authority should approve the criteria and by what process, including how changes should be authorized.

Specify how the lead’s data, design, and targeting patterns could be checked for errors and discriminatory effects, including bias introduced during system design.

What we can say so far

A starting point is to examine how the audit-selection task is translated into instructions and code, since bias can enter during design. Targeting should be assessed against legitimate grounds rather than assumed fair merely because it is automated. The supplied material does not specify how to check the lead’s data for errors, test targeting patterns for discriminatory effects, or conduct a detailed design audit.

What remains open

• A procedure for detecting errors in the data used to produce leads.
• A detailed method for auditing design choices, instructions and code for bias.
• A method and criteria for assessing targeting patterns and discriminatory effects.

Assess whether officials have the information and practical freedom to reject a lead, and whether affected people can examine relevant evidence and reasons and contest targeting before an impartial reviewer; distinguish formal review rights from effective error detection and correction.

What we can say so far

Effective review would require officials to have usable information about why a tax-audit lead was generated and practical freedom to reject it, not merely formal authority to do so. Affected people would need notice, access to relevant evidence and reasons, and an opportunity to contest the targeting before an impartial reviewer with a route to correction. Crawford and Schultz propose notice, examination and impartial appeal as rights; the supplied material does not show that those rights would work effectively in practice. Danaher argues that people should be able to examine and question the factors used, without necessarily understanding every technical detail. Opaque methods or dependencies across systems may nevertheless obstruct review. The evidence does not establish whether officials or affected people can actually detect and correct errors or illegitimate targeting in a tax-audit system.

What remains open

• No evidence establishes what information or practical freedom officials actually have to reject tax-audit leads.
• No evidence demonstrates that affected people can examine the relevant evidence and reasons, obtain impartial review, or secure correction of errors or illegitimate targeting in practice.
• The practical threshold for sufficient understanding and the effectiveness of proposed review safeguards remain unspecified.