Using AI Productivity Telemetry for Layoff Selection: Unprorated Leave Denominators, State ADS Rules, and Disparate-Impact Audits
Feeding raw Git commits, Jira velocity, or AI-assistant token logs into a RIF spreadsheet feels objective—until unprorated medical leave skews the denominator, Illinois demands discharge AI notices, and California enforces four-year ADS audit retention.
- Jurisdiction
- Federal FMLA/ADA/UGESP, CA CRD ADS Regs (2 CCR § 11008) & IL 775 ILCS 5/2-102(L)
- Employer threshold
- 5+ employees in CA (CRD ADS rules); 1+ employee in IL (IHRA AI amendment); 15+/50+ Federal
- Decision trigger
- RIF ranking sheet ingesting Git commits, ticket counts, badge scans, or AI-tool token telemetry
- Review cadence
- Event-triggered on Does v. Meta rulings, CA SB 951 (Jan 1, 2027), and CA SB 947 (Jul 1, 2027)
Direct operational answer
Normalize every layoff-selection metric to active workdays under the hr.cafe telemetry-RIF gate before ranking Git commits, ticket counts, or AI token logs. Federal 29 CFR § 825.220(c) bars penalizing protected leave through unprorated cumulative totals, Illinois 775 ILCS 5/2-102(L) mandates employee notice when AI informs discharge, and California 2 CCR §§ 11008–11013 requires four-year retention of all automated-decision scoring formulas, inputs, and selection outputs.
How "objective" engineering telemetry encodes illegal leave penalties
Cumulative six-month or twelve-month counts of Git commits, Jira story points, closed support tickets, and AI-coding token prompts look neutral on a slide deck. Without active-workday denominators, those formulas mathematically target employees who took protected medical or parental leave.
Every technical restructuring inside AI and changing work eventually produces the same spreadsheet. An engineering or operations executive exports twelve months of merged pull requests, resolved Zendesk tickets, badge swipes, and internal AI-assistant prompt volume into a weighted z-score column. Anyone scoring in the bottom 15% of their job family lands on the preliminary Reduction in Force (RIF) tab.
Inspect the SQL query or XLOOKUP behind that bottom-15% cutoff and a textbook statutory defect jumps off the screen. Most telemetry exports sum raw volume across 260 annual workdays without joining HRIS leave tables. Consider two Senior Engineers who each merge 4.0 pull requests per active week:
- Engineer A (zero leave): Works 50 active weeks and logs 200 merged pull requests across the 12-month window.
- Engineer B (12 weeks of FMLA or PWFA recovery leave): Works 38 active weeks at the identical 4.0 weekly rate and logs 152 pull requests—a 24% drop in raw annual volume caused 100% by protected leave.
Under federal 29 CFR § 825.220(c), employers cannot count FMLA leave as a negative factor in discipline or layoff selection. Parallel rules protect intermittent ADA leave and PWFA recovery. Feeding unprorated 12-month volume into a stack-rank sheet turns every protected leave day into a zero-output penalty.
Role seniority creates a second distortion tied to age. Staff and Principal engineers aged 45 and older often spend 60% of their week on architecture reviews, incident postmortems, and AI code audits rather than raw commits or token prompts. Ranking a unit on raw commit or AI-token volume pushes older technical leads into the bottom decile, colliding directly with your OWBPA Exhibit A age-table disclosures.
What Does 1–26 v. Meta Platforms signaled in July 2026
In Does 1 through 26 v. Meta Platforms, Inc. (N.D. Cal., July 2026), 26 terminated workers challenged algorithmic RIF selection criteria for penalizing protected leave and older cohorts. Judge William H. Orrick found plaintiffs raised "serious questions going to the merits" of discrimination.
Algorithmic layoff litigation is active right now. On July 13, 2026, twenty-six affected workers filed Does 1 through 26 v. Meta Platforms, Inc., Case No. 3:26-cv-07122-WHO (N.D. Cal.), alleging that an automated layoff pipeline penalized employees who took protected leave alongside older cohorts.
On July 17, 2026, Judge William H. Orrick denied a general temporary restraining order because most plaintiffs held compensable monetary claims subject to arbitration. Crucially, Judge Orrick found the plaintiffs raised "serious questions going to the merits" of discrimination and ordered supplemental briefing on irreparable harm for visa-dependent workers.
For practical people operations teams, Does 1–26 proves a blunt truth. Wrapping unprorated inputs inside a scoring script never launders bias; it preserves exact mathematical proof for plaintiff experts.
State ADS mandates already in force: California CRD and Illinois IHRA
As of October 2026, California enforces broad Automated-Decision System (ADS) anti-discrimination and four-year recordkeeping rules, while Illinois mandates affirmative employee notice whenever AI informs discipline or discharge.
Teams often note that NYC Local Law 144 covers hiring and promotion rather than terminations, then falsely assume no state AI rule touches layoff spreadsheets. Two state mandates apply today:
- California CRD ADS Regulations (effective October 1, 2025): Under Cal. Code Regs. tit. 2, § 11008 et seq., an ADS covers any computational or weight-based scoring process that facilitates HR termination or layoff decisions. Critically, 2 CCR § 11013(c) doubles retention from two years to four full years for all ADS inputs, formulas, overrides, and outputs.
- Illinois IHRA AI Amendment (effective January 1, 2026): Under 775 ILCS 5/2-102(L) (P.A. 103-0804), employers must give mandatory employee notice when using AI in recruitment, promotion, discipline, or discharge, and cannot produce discriminatory effects or use zip-code proxies.
The enacted California 2027 statutory horizon: SB 951 and SB 947
Multi-quarter workforce plans extending into 2027 must also build compliance gates for two enacted California statutes signed September 30, 2026: SB 951 (effective January 1, 2027) and SB 947 (operative July 1, 2027).
Separate active 2026 mandates from two newly signed California laws taking effect in 2027:
- California SB 951 — AI Cal-WARN 90-Day Notice (Ch. 860, effective January 1, 2027): Amends Cal. Lab. Code § 1401 to mandate 90-day advance notice and explicit disclosure when mass layoffs stem from AI or automation.
- California SB 947 — "No Robo Bosses Act" (Ch. 859, operative July 1, 2027): Enacts Cal. Lab. Code §§ 1550–1557, barring termination or discipline based solely on an ADS output without human review of corroborating records.
Telemetry-RIF & Automated-Decision Compliance Matrix
| Authority & Effective Status | Trigger in Layoff / Stack-Rank Pipeline | Primary Failure Mode | Mandatory Operational Control |
|---|---|---|---|
| Federal FMLA / ADA / PWFA (29 CFR § 825.220(c)) — Active | 12-month commit, ticket, or AI-prompt totals used in RIF ranking. | Zero-output leave days depress cumulative score and select leave-takers. | Divide all output metrics strictly by net active workdays (excluding all protected leave). |
| Federal UGESP Disparate Impact (29 CFR § 1607.4(D)) — Active | Weighted telemetry score selects protected age, sex, race, or disability cohorts disproportionately. | Retention rate for 40+ or female staff falls below 80% (four-fifths) of reference group or |Z| ≥ 1.96. | Run privileged pre-RIF 4/5ths + Fisher's exact test; strip unvalidated proxy metrics. |
| California CRD ADS Rules (2 CCR §§ 11008, 11013(c)) — Active (Oct 1, 2025) | Spreadsheet formula, analytics model, or AI score facilitates CA layoff selection. | Overwriting draft scoring tabs or purging raw telemetry within 2 years. | Archive immutable 4-year audit package: raw inputs, formula weights, overrides, and outputs. |
| Illinois IHRA AI Rule (775 ILCS 5/2-102(L)) — Active (Jan 1, 2026) | AI tool evaluates productivity or recommends IL employees for discharge. | Using AI in discharge ranking without prior written AI-use notice to IL staff or using zip-code proxies. | Issue written 775 ILCS 5/2-102(L) AI notice prior to evaluation; strip geographic zip fields. |
| California SB 951 & SB 947 — Enacted Sept 30, 2026 (Eff. Jan 1 & Jul 1, 2027) | 2027 AI-driven mass layoffs (SB 951) or ADS-assisted terminations (SB 947). | Using 60-day WARN calendar for 2027 AI displacement or rubber-stamping raw ADS scores. | Build 90-day Cal-WARN lead time from Jan 1, 2027; require human corroboration gate by Jul 1, 2027. |
Forensic normalization and privileged pre-RIF statistical testing
Never pass raw engineering or support telemetry straight into a termination list. Normalize denominators to net active workdays, validate job-family parity, run both four-fifths and exact small-sample tests under legal privilege, and archive immutable California 4-year snapshots.
Before any telemetry-assisted selection sheet reaches executive sign-off, route the pipeline through four technical controls under attorney-client privilege:
- Active-day denominator join: Join HRIS leave, disability, PWFA accommodation, jury duty, and bereavement tables against the telemetry window first. Compute
Velocity = Output / Net_Active_Workdays, and exclude ramp periods following return from extended medical leave. - Level-band stratification: Never pool Junior (
L3) and Principal (L6) engineers on a single commit-volume curve. Stratify within identical level and role archetypes, or drop volume telemetry altogether where senior scope centers on code review and architecture. - Dual statistical adverse-impact screen: Under UGESP 29 CFR § 1607.4(D), compare retention rates across sex, race/ethnicity, age (40+, 50+, 60+), disability, and recent leave-takers against the highest-retained group's rate (flagging any ratio below 80%, or 0.80). Because moving one worker in a 14-person team flips the 80% ratio, pair the four-fifths check with Fisher's exact test for small cohorts and a two-standard-deviation binomial z-test (
|Z| ≥ 1.96) for larger pools. - Immutable 4-year ADS preservation & Illinois notice check: Confirm written Illinois 775 ILCS 5/2-102(L) AI notices were delivered before any AI scoring touched Illinois staff, and lock a read-only four-year archive in compliance storage per California 2 CCR § 11013(c).
PRIVILEGED & CONFIDENTIAL — PRE-RIF TELEMETRY & ADS AUDIT GATE
Prepared at Direction of Legal Counsel | Archive Retention: Minimum 4 Years (2 CCR § 11013(c))
Decisional Unit: [Org Unit] Evaluation Window: [Start] to [End] Headcount: [N = ___]
1. DENOMINATOR NORMALIZATION (29 CFR § 825.220(c) / ADA / PWFA):
[ ] Joined HRIS FMLA, state PFML, ADA intermittent, and PWFA accommodation tables?
[ ] Formula verified as [Metric / Net_Active_Workdays] (0 protected-leave days in denominator)?
[ ] Excluded 30-day post-leave re-onboarding ramp windows from velocity penalty?
2. PROXY & JURISDICTIONAL AI GATES (775 ILCS 5/2-102(L) / 2 CCR § 11008):
[ ] Removed zip code, home location, and non-job-validated AI-token adoption counts?
[ ] Confirmed prior written AI-use notice delivered to all covered Illinois employees?
[ ] Checked 2027 horizon triggers: CA SB 951 (90-day AI Cal-WARN eff. 2027-01-01) & SB 947 (eff. 2027-07-01)?
3. STATISTICAL ADVERSE-IMPACT RESULTS (29 CFR § 1607.4(D)):
Cohort Tested | Retained / Pool | Retention % | 4/5ths Ratio (≥0.80) | Fisher p / Z-Score
--------------------|-----------------|-------------|----------------------|-------------------
Age 40+ vs. <40 | [__ / __] | [___%] | [0.__] | [p = 0.__ / Z = __]
Protected Leave 12m | [__ / __] | [___%] | [0.__] | [p = 0.__ / Z = __]
SIGN-OFF: [ ] Cleared [ ] HALTED — Recalibrate criteria & require independent human review.Join HRIS protected-leave logs into every productivity export before calculating rankings, issue mandatory Illinois AI notices, and lock all scoring inputs and versions in a 4-year California CRD archive.
Never treat raw Git commits, closed tickets, or AI-prompt counts as standalone layoff rankings; corroborate every proposed selection against documented role deliverables and level expectations.
Conduct privileged 4/5ths and Fisher's exact disparate-impact audits across age and leave-taking cohorts before finalizing RIF lists, and build 90-day lead times for 2027 California AI displacement notices under SB 951.
Primary statutes, regulations, and enforcement dockets
- California Civil Rights Council — Automated-Decision System (ADS) Regulations, 2 CCR §§ 11008–11013 (Effective Oct. 1, 2025) — Defines ADS broadly across computational/formulaic HR tools, confirms anti-discrimination liability, and mandates 4-year ADS recordkeeping under 2 CCR § 11013(c).
- Illinois Public Act 103-0804 (775 ILCS 5/2-102(L)) — Illinois Human Rights Act AI Notice & Non-Discrimination Mandate (Effective Jan. 1, 2026) — Requires employee notice when AI is used in recruitment, promotion, discipline, or discharge, and prohibits discriminatory impact or zip-code proxies.
- 29 CFR § 825.220(c) (FMLA Negative-Factor Prohibition) & 29 CFR § 1607.4(D) (UGESP Four-Fifths Adverse Impact Rule) — Federal rules barring unprorated leave penalties in employment selections and establishing the 80% selection-rate benchmark for disparate-impact screening.
- Does 1 through 26 v. Meta Platforms, Inc., Case No. 3:26-cv-07122-WHO (N.D. Cal., Order Denying General TRO & Finding Serious Merits Questions, July 17, 2026) — Northern District of California challenge to algorithmic layoff selection alleging leave and age bias, where Judge Orrick found serious questions going to the merits.
- California SB 951 (Ch. 860, eff. Jan. 1, 2027) & SB 947 (Ch. 859, operative July 1, 2027) — California Legislative Information — Enacted September 30, 2026: SB 951 expands Cal-WARN to 90 days for AI/automation mass layoffs; SB 947 bars sole reliance on ADS for termination without human corroboration.
Practitioner questions on the edge cases
Does a standard Excel or Google Sheets weighted formula count as an Automated-Decision System (ADS) in California?
Yes. Under 2 CCR § 11008, California defines an Automated-Decision System to encompass any computational process derived from statistical modeling, data analytics, or algorithmic scoring that facilitates an employment decision. An internal weighted ranking sheet ingesting commit counts and performance scores triggers the 4-year retention rule in 2 CCR § 11013(c) just like third-party AI software.
Why does NYC Local Law 144 not cover layoff selection spreadsheets when Illinois 775 ILCS 5/2-102(L) does?
NYC Local Law 144 (Admin. Code § 20-871) applies specifically to Automated Employment Decision Tools used to screen candidates for employment or employees for promotion within New York City. Illinois 775 ILCS 5/2-102(L), effective January 1, 2026, explicitly lists recruitment, hiring, promotion, training, discipline, and discharge—bringing AI-assisted terminations squarely inside Illinois's notice and anti-bias mandate.
Can we rank engineers on internal AI coding-assistant adoption rates (tokens or accepted completions) when choosing who to retain?
Using raw AI token volume as a termination metric carries severe validity and disparate-impact risk. Token consumption often measures verbosity or junior boilerplate generation rather than verified code quality, penalizes employees on protected leave unless prorated to active days, and frequently skews against senior architectural roles held by workers over 40 unless job-relatedness is empirically validated.
Should we delete early iterations of a layoff ranking spreadsheet if our disparate-impact audit flags an age or leave skew?
Never destroy discovery-relevant records or violate California's 4-year ADS retention rule under 2 CCR § 11013(c). Instead, run pre-decisional statistical modeling at the direction of legal counsel under attorney-client privilege, document the lawful remediation (such as fixing unprorated leave denominators before any decision is made), and preserve the privileged audit trail properly.
Get the Thursday edition
One email a week with new practitioner stories and primary-sourced HR playbooks. Free, and one click to unsubscribe.