Briefs / Brief №023 / Audit Packet
← Return to brief
Autonoma / Intelligence Brief №023 · Public Audit Packet · September 2026
Read the brief →

Audit Packet: The System of Record Has No Object for the Act

This public Audit Packet documents the evidence basis, claim boundaries, counterarguments, editorial judgments, and falsification tests behind Brief No. 023.

This audit packet supports Brief №023: The System of Record Has No Object for the Act. Read the brief first for the full argument.

Autonoma briefs are designed to be inspectable. This packet shows what the brief claims, how each claim was tested, what it does not claim, and where caveats remain — without exposing raw internal logs, prompts, operator notes, source-routing mechanics, hashes, local paths, secrets, or unpublished candidate claims.

← Open Brief №023 — The System of Record Has No Object for the Act

§ 01

Brief Summary and Audit Verdict

Brief 023 asks whether, when an agent screens a candidate, approves leave, routes an HR case, or flags a performance concern, the HR system of record has a record type that can hold that act on the person it touched. It argues that the record keeps the outcome and drops the actor, that employment law already requires those records to be kept, and that the agent registries vendors now sell answer a different question.

Audit verdict: The Brief is supported at the mechanism and legal-duty layer. Four independent domains carry the argument: a practitioner essay for the mechanism, a state civil-rights agency and the federal regulation for the retention duty, and a vendor announcement for product behavior only. The Brief doesn't claim that a named employer lost a legal challenge because of this gap, and it doesn't estimate how many HR systems lack an actor field. The reading of the two retention rules together, the registry-versus-record distinction, and the 12-month forecast are labeled as Autonoma synthesis.

§ 02

Claim Register

#Claim in the BriefStatusDomainBound
1When an agent screens a candidate, approves time off, or routes an HR case, the system of record typically captures the outcome but not the actorSupported as practitioner mechanismhrexecutive.comMittana, HR Executive, 27 April 2026; one author's account, not a survey
2Most enterprise HRIS platforms were not built to store, manage, or audit workflows that involve non-human workers; agents arrive as IT service accountsSupported as practitioner judgmenthrexecutive.comSame essay; "most" is the author's judgment
3An act record would show which agent acted, on which record, with what inputs and outcome, and whether a human reviewed it, and HR should produce it from the HR system rather than a vendor dashboard or server logSupported as practitioner recommendationhrexecutive.comSame essay; a recommendation, not a standard
4Employers must preserve personnel and employment records tied to hiring, promotion, termination, pay, and selection for training for one year, and keep relevant records until a discrimination charge is resolvedSupportedecfr.gov29 CFR 1602.14, current text as last amended 28 May 2024
5California's automated-decision regulations, approved 27 June 2025 and effective 1 October 2025, require employers to keep employment records, including automated-decision data, for at least four years, and define "agent"Supportedcalcivilrights.ca.govCivil Rights Department announcement, 30 June 2025
6Workday's Agent System of Record is generally available, manages agents "alongside employees," and records and tracks agent interactionsSupported as product behaviorblog.workday.comVendor announcement, 18 February 2026; not evidence of customer configuration
7A registry for the agent is a different object from a record of the act on the person's file; retention duties reach both the file and the vendor log, but the schema covers only the fileAutonoma synthesis—Reading across 1, 2, 3, and 4
8A named employer's record failed a legal challenge because the actor was missingWithheld—Not in the sources
9Over the next 12 months, suites will ship agent registries faster than act records, and buyers will treat the registry as the audit trailAutonoma forecast—Labeled in the Brief
§ 03

Source Ledger

The Workday announcement's figure on how many organizations are expanding agent use, and its customer count, were read and kept out of the Brief because they are vendor figures that do not bear on the question.

§ 04

Evidence Boundaries

  • A practitioner essay establishes a mechanism. It does not count how many platforms have the gap.
  • A retention rule establishes that a record must be kept. It does not require a particular field or database design.
  • A vendor announcement establishes what a product is built to do. It does not show what any customer's records contain.
  • The six-months-later scene in the Analysis is an illustration built from the essay's description. It is not a documented case.
  • Briefs 007, 012, 021, and 022 remain separate. This Brief does not argue that the agent belongs on the org chart (007), that it acts on stale data (012), who assigns and writes a learning record (021), or whether an agent's approval joins its tool call inside the agent's own trail (022). It asks only whether the employment record the act changed can name the actor.
  • This packet contains no legal advice and no finding about any specific employer.
§ 05

Dissent and Limiting Case

The live objection is that logs already solve the problem: a vendor registry records agent interactions, so a legal team can match the log to the employee record and reconstruct the act. The Brief accepts that logs are real evidence, that neither retention rule says where records must live, and that the Workday product addresses the visibility gap. It rejects the conclusion. Two records that agree on a time are not one record of an act, and reconstructing the join per request from a system HR doesn't control leaves HR accountable for a decision it can't see in its own record. The objection also assumes that vendor logs are kept as long as the employment record, which is a contract term, not a given.

The limiting case is the evidence base. The mechanism rests on one practitioner's essay, and its "most" is a judgment rather than a measurement. A vendor document or regulatory finding showing that major HR platforms already write the acting agent onto the person's record would narrow the Brief. None was found.

§ 06

Falsification

This Brief is wrong, or must be rewritten, if:

  • major HR platforms are shown to record the acting agent on the candidate, leave, case, or performance record by default, in which case the gap becomes a configuration problem rather than a schema problem; or
  • the California regulations are shown not to cover automated-decision data in their retention requirement, or 29 CFR 1602.14 is amended to exclude records produced by automated systems; or
  • the Workday product is documented as writing the agent's identity into each affected employee record, in which case the registry-versus-record distinction narrows for that product.

The Brief tightens if a named challenge to an employment decision turns on an HR record that could not show an agent produced the outcome.

§ 07

Forecast Label

The sentence "Over the next 12 months, HR suites will ship agent registries and agent analytics faster than they add an actor to the employee and candidate records those agents change" is Autonoma Intelligence synthesis. So is the claim that buyers will accept the registry as the audit trail. Neither is a quotation-level fact from 1, 2, 3, or 4.

§ Verified

Editor-verified sources

Source [1] (hrexecutive.com) was verified by the SIGNIT pipeline. Sources [2], [3], and [4] are listed below.

These sources were checked by the editor against saved page copies; they were not verified by the SIGNIT pipeline.

§ Method

Methodology

This packet audits Brief 023 against the four public sources listed above, each read at its public web address. Every claim in the register is held to what its source states, the vendor source is limited to product behavior, and synthesis and forecast are labeled. This draft was prepared with an AI assistant and reviewed and approved by the editor.