Policy Brief: The Algorithmic Tribunal Gap

Cluster: axiom — agency — tribunal — monopoly — automation

Source analyses: 095 (governance axioms as commitment-systems; epistemological seigniorage), 105 (the NLRA as syntactic mint; containment by denomination), 120PB (cross-boundary recognition claims; jurisdictional voids), 135 (automation’s dual registration; the neutrality permit; temporal arbitrage), 147 (automation censors the relational through denominational incapacity), 156 (agency requires substrate — rights, temporal security, associative infrastructure — which novelty-denominated restructuring erodes), 1253 (institutions are memories; deregulation as amnesia)

Classification: U.S. labor governance | State-level actionable, federal-level preparatory | Urgent

Problem Statement

Algorithmic management systems now set the governance axiom of work for tens of millions of U.S. workers: they define what counts as a task, what constitutes performance, when a shift begins and ends, why a worker is terminated. These systems encode the efficiency axiom (095) — commensurability, observability, optimization — into operational infrastructure that workers cannot inspect, contest, or bargain over.

Simultaneously, the tribunal minted to adjudicate labor disputes — the National Labor Relations Board — is being structurally dismantled. The current administration has fired board members, cut regional office staff, and reduced enforcement activity. Even at full capacity, the NLRB was minted (105) for the employer-employee dyad of 1935: it processes claims denominated in “unfair labor practices,” “bargaining units,” and “mandatory subjects of bargaining.” Algorithmic scheduling, automated termination, surveillance-based productivity scoring, and opaque performance metrics are not well-formed claims in this grammar. The tribunal cannot hear them — not because they are illegitimate but because no institution was authorized to process them (120PB’s jurisdictional void).

The conjunction is the problem. The axiom is automated (the employer’s governance grammar now runs at machine speed, opaque by design). The tribunal is being destroyed (the institutional memory of how to process labor claims is degrading through 1253’s amnesia sequence — from archival memory to amnesia-as-opportunity). And the monopoly is grammatical: the employer who controls the algorithm controls the denomination in which work is defined, measured, and settled. Workers face a governance grammar they cannot read, before a tribunal that no longer functions, against an employer whose monopoly is over the syntax itself.

Decision needed: How to restore adjudicatory capacity for algorithmic workplace governance before the temporal arbitrage (135) closes — before algorithmic management becomes infrastructure too entrenched to restructure.

Primary decision-makers: State legislators in labor-friendly jurisdictions (California, New York, Illinois, Washington, Minnesota, Colorado). Secondary: Congressional committee minority staff preparing model legislation for changed federal conditions.

Background

The axiom encoded

Algorithmic management systems operationalize a specific governance axiom: the worker is an input whose output is optimizable. This axiom was always implicit in Taylorist management, but human managers introduced friction — discretion, relationship, the gift-dimension of supervision (147) — that softened the axiom’s denominational demand. The algorithm removes the friction. It processes the worker through precisely the four requirements 095 identified for the utility axiom:

Individuable: The algorithm tracks individual workers, not crews or teams. Collective performance is decomposed into individual metrics. Solidarity is syntactically invisible.
Commensurable: All work is reduced to a common metric (packages per hour, ride acceptance rate, task completion time). What cannot be metrified does not exist in the system’s grammar.
Comparable: Workers are ranked against each other in real time. The ranking determines dispatch priority, shift allocation, and continued employment.
Observable: Only behavior the algorithm can detect counts as work. The care a nurse takes with a patient, the judgment a driver exercises in traffic, the mentorship a warehouse worker provides to a new hire — all tangent-points (147) captured as data and stripped of relational content.

The axiom’s seigniorage (095) — the gap between what the worker does and what the algorithm sees — is invisible by construction. The algorithm defines the observation apparatus. What it cannot process is, by definition, not performance.

The tribunal hollowed

105 established that the NLRA minted the syntax of labor relations: bargaining units, mandatory vs. permissive subjects, certification elections, unfair labor practice categories. Each mint-operation contained labor demands by denominating them in the sovereign’s currency. But the mint at least existed. Workers could file charges. The NLRB could investigate. The grammar, however constrained, was operative.

Three concurrent erosion channels are now destroying this:

Political dismantlement. Board vacancies, staff cuts, regional office closures. The NLRB’s case processing time was already measured in years; it is now approaching functional non-existence for contested cases. This is 1253’s Stage 4: the institution’s success in preventing the worst pre-Wagner-Act labor abuses has destroyed the evidence for its own necessity.
Grammatical obsolescence. The NLRB was minted for the factory and the office. The platform worker, the algorithmically managed warehouse worker, the gig-classified delivery driver all present claims that cross the tribunal’s jurisdictional boundaries (120PB). Is algorithmic scheduling a “working condition” (mandatory subject) or a “management prerogative” (permissive subject)? Is automated termination a “discharge” requiring just cause, or a “deactivation” outside the employment relationship? The grammar has no stable answer because the grammar was minted before the question existed.
Monopoly over the evidentiary substrate. The employer controls the algorithm. The algorithm generates the data. The data constitutes the evidence. A worker challenging an automated termination must prove the algorithm’s decision was retaliatory or discriminatory — but the algorithm’s logic is proprietary, its training data is inaccessible, and its decision process is opaque even to the engineers who maintain it. The employer holds a monopoly not merely over the workplace but over the grammar in which claims about the workplace can be formulated.

The agency erosion

156 established that agency requires substrate: enforceable rights, temporal security, associative infrastructure, and institutional environments whose behavior is predictable enough to orient action. Algorithmic management erodes each element:

Rights are formally intact but operationally empty (105’s seigniorage widening). The “right to organize” means little when the algorithm can adjust schedules, reduce dispatch priority, or restructure task allocation in response to organizing activity — all without any action legible as an “unfair labor practice” in the NLRB’s grammar.
Temporal security is destroyed by just-in-time scheduling, algorithmically determined shift lengths, and income volatility that makes planning — the precondition of collective action — impossible.
Associative infrastructure is eroded by atomization. Algorithmically dispatched workers may never meet their coworkers. The “bargaining unit” presupposes a community of workers who share a workplace; algorithmic management can structure work so that no such community forms.
Institutional predictability is zero. The algorithm can change the rules of work — the pay structure, the performance thresholds, the dispatch logic — unilaterally and without notice. Workers orient their labor toward metrics that shift beneath them.

Constraints on action

Budget: State labor agencies are modestly funded. Any new adjudicatory capacity must be buildable within existing institutional infrastructure or require minimal new appropriation.
Time: 135’s temporal arbitrage is closing. Algorithmic management is becoming infrastructure — path-dependent, constituency-creating, naturalized. Each year of inaction makes restructuring harder. The window for governance intervention that precedes infrastructure lock-in is measured in years, not decades.
Politics: Federal action is blocked. State action is viable in 10-15 jurisdictions. Tech industry lobbying is intense and well-funded. The “neutrality permit” (135) — the framing of algorithmic management as a technical efficiency question rather than a governance question — is the primary discursive obstacle.
Legal: Federal preemption under the NLRA is a significant constraint. The NLRA preempts state regulation of conduct “arguably subject to” Section 7 (organizing rights) or Section 8 (unfair labor practices). States cannot simply create parallel labor boards. But states retain authority over areas the NLRA does not reach: wage and hour law, occupational safety, data privacy, consumer protection, and — critically — areas where the NLRB has declined jurisdiction or the employment relationship is contested.
Information asymmetry: The employer knows everything the algorithm does. The worker knows nothing. The regulator knows less than nothing — it cannot even formulate the right questions without access to the system’s logic.

Options

Option A: Algorithmic Transparency and Contestation Statutes (the “Audit and Appeal” approach)

Mechanism: State legislation requiring employers using automated decision systems in the workplace to (1) disclose the existence and general logic of algorithmic management tools to workers, (2) provide individual explanations for automated decisions affecting employment (termination, discipline, scheduling, pay), (3) establish internal appeal processes with human review, and (4) submit to periodic independent audits of algorithmic systems for disparate impact and labor-law compliance.

Precedent: New York City’s Local Law 144 (2023, automated employment decision tools); Illinois AI Video Interview Act (2020); EU AI Act’s high-risk classification for employment AI (2024). Colorado’s SB 24-205 (algorithmic discrimination prevention, effective 2026).

Implementability: High. Builds on existing legislative templates. Can be enacted through state consumer protection or labor committee channels. Does not directly regulate the employment relationship (avoiding NLRA preemption). Compliance costs are moderate — the audit industry already exists for AI systems. Fits within existing state regulatory budgets by placing audit costs on employers.

Expected impact: Moderate. Breaks the information monopoly — workers and regulators can see what the algorithm does. Creates an evidentiary substrate for future claims (workers who know why they were terminated can challenge the reason). But transparency alone does not create adjudicatory capacity. Knowing the algorithm is unjust and having an institution that can do something about it are different problems. The “audit and appeal” approach addresses the monopoly dimension but not the tribunal gap.

Risk: The “transparency theater” problem (100’s simulated resonance). Disclosure requirements can become compliance rituals that satisfy the letter without breaking the information monopoly. The algorithm’s “explanation” may be a post-hoc rationalization generated for compliance rather than a genuine account of the decision logic. Audits may certify systems as compliant with narrow technical criteria while missing structural effects.

Option B: State Labor Board Jurisdiction over Algorithmic Management (the “Mint Extension” approach)

Mechanism: State legislation declaring that algorithmic management decisions (automated scheduling, performance scoring, dispatch allocation, automated discipline, and automated termination) constitute “working conditions” subject to state labor-law jurisdiction. Empower state labor agencies to investigate complaints, issue findings, and order remedies (including algorithm modification orders) for algorithmic management practices that violate state labor standards.

Precedent: California’s AB 701 (2021, warehouse production quotas — requires employers to disclose quotas and prohibits quotas that prevent compliance with health and safety standards). Washington’s SB 5115 (2021, algorithmic accountability for state agencies). The warehouse-quota model is directly extensible: it treats the algorithm’s demands as a “working condition” subject to regulatory scrutiny.

Implementability: Medium. Requires expanding state agency jurisdiction, which means new staff, new expertise (algorithmic auditing), and new procedural infrastructure. Federal preemption risk is moderate — the approach frames algorithmic management as a health-and-safety or wage-and-hour issue (areas of traditional state authority) rather than a collective-bargaining issue (NLRA preempted). Legal challenges are likely but defensible. Budget impact: $5-15M annually per state for dedicated enforcement staff and technical capacity. Achievable through fee-on-employer models.

Expected impact: High. This is the option that creates actual adjudicatory capacity — a tribunal that can hear claims about algorithmic governance. It mints a new denomination: “algorithmic working condition” becomes a processable claim. It does not solve the atomization problem (workers still need to file individual complaints rather than bargain collectively), but it creates the institutional infrastructure on which future collective action can build. The state labor board becomes the tribunal the NLRB was never minted to be.

Risk: The conservation constraint (120PB): state labor boards have finite capacity. Adding algorithmic-management jurisdiction without expanding resources produces seigniorage inflation — the new right’s face value exceeds its institutional backing. The political economy of enforcement also matters: agencies that depend on annual appropriation are vulnerable to the same dismantlement cycle that is destroying the NLRB. The new tribunal must be structurally insulated from the administration cycle.

Option C: Sectoral Bargaining for Algorithmically Managed Industries (the “Grammar Bypass” approach)

Mechanism: State legislation establishing sectoral bargaining frameworks for industries with high algorithmic management penetration (warehousing, ride-hail, delivery, food service). Rather than organizing bargaining-unit-by-bargaining-unit (105’s containment-by-denomination), sectoral bargaining sets minimum standards for an entire industry through tripartite negotiation (worker representatives, industry representatives, state mediator). Algorithmic management practices — scheduling algorithms, performance metrics, automated discipline thresholds — become mandatory subjects of sectoral negotiation.

Precedent: California’s FAST Act (AB 257, 2022 — fast food sector council, partially enjoined then modified by referendum). New York’s Fastfood Wage Board. European sectoral bargaining systems (Germany’s Tarifvertrag, France’s conventions collectives). Australia’s modern awards system.

Implementability: Low-Medium. Sectoral bargaining is structurally unfamiliar in U.S. labor law. The FAST Act’s political trajectory — enacted, challenged by industry referendum, modified — illustrates the intensity of opposition. Requires substantial political capital. Federal preemption risk is lower than Option B because sectoral bargaining does not replicate NLRA collective-bargaining procedures — it establishes minimum standards through a state-created body, which is more analogous to minimum-wage-setting (clearly within state authority) than to collective bargaining (NLRA preempted). Budget impact: moderate ($10-25M for sectoral council infrastructure).

Expected impact: Very high — if enacted. This is the only option that addresses the atomization problem. By setting industry-wide standards, sectoral bargaining bypasses the bargaining-unit mint (105) entirely. Workers do not need to organize employer by employer. The algorithm’s governance grammar is contested at the industry level, where individual employer monopoly over the syntax is broken by cross-employer standard-setting. Algorithmic management practices that are currently unilateral employer decisions become subjects of tripartite negotiation. This is the structural intervention: it changes who defines the denomination in which work is measured.

Risk: Political feasibility is the binding constraint. Industry opposition will be overwhelming — the fast-food industry spent $50M+ to challenge California’s FAST Act. The “neutrality permit” (135) will be deployed aggressively: algorithmic management will be framed as technical optimization that sectoral bargaining would “stifle.” If enacted in weakened form (advisory-only councils, narrow scope), the face value of the intervention will exceed its material backing — seigniorage inflation again.

Option D: Worker Data Portability and Cooperative Infrastructure (the “Information Sovereignty” approach)

Mechanism: State legislation requiring that (1) workers own the data their labor generates within algorithmic management systems, (2) employers provide portable, machine-readable exports of worker performance data, scheduling history, and algorithmic evaluations upon request, and (3) state agencies support the formation of worker data cooperatives that aggregate individual data into collective bargaining intelligence.

Precedent: EU GDPR’s data portability right (Article 20). California’s CCPA/CPRA data access provisions. Colorado’s data privacy act. Driver data portability proposals in ride-hail regulation.

Implementability: High. Data portability legislation is well-understood, politically moderate (framed as consumer rights, not labor regulation), and triggers minimal preemption risk. The cooperative infrastructure component requires modest state investment ($2-5M in grants and technical assistance). Can be enacted through privacy or consumer-protection committees without engaging labor-law preemption at all.

Expected impact: Moderate, but strategically enabling. Data portability alone does not create adjudicatory capacity or bargaining power. But it breaks the employer’s monopoly over the evidentiary substrate. Workers with their own data can (a) detect algorithmic discrimination or retaliation patterns that are invisible at the individual level, (b) support organizing efforts with quantitative evidence, (c) file complaints with state agencies that have an evidentiary basis, and (d) build the informational infrastructure for future collective action. The data cooperative is the structural innovation: it converts atomized individual data into collective intelligence — the informational equivalent of the union’s aggregation of individual bargaining power.

Risk: Narrow scope. Data without adjudicatory capacity is information without power. The employer can acknowledge the data, dispute its interpretation, and continue the same algorithmic practices. The option must be combined with at least Option A or B to have material effect. Also: employers will argue trade-secret protection for algorithmic logic (distinguishing worker data from the employer’s system), and this argument has legal weight that will constrain what “portability” delivers.

Trade-offs

Criterion	A: Audit & Appeal	B: Mint Extension	C: Sectoral Bargaining	D: Data Sovereignty
Implementability	High	Medium	Low-Medium	High
Impact on tribunal gap	Low (no adjudicatory capacity)	High (creates tribunal)	Very high (bypasses unit-level containment)	Low (enables but doesn’t create)
Impact on information monopoly	High	Medium	Medium	Very high
Impact on atomization	None	Low (individual complaints)	Very high (industry-level)	Medium (cooperative aggregation)
Preemption risk	Low	Moderate	Low-Moderate	Very low
Political cost	Low-Moderate	Moderate	Very high	Low
Speed to effect	1-2 years	2-4 years	3-6 years	1-2 years
Conservation risk	Low	High (capacity strain)	Moderate	Low

The fundamental trade-off is between implementability and structural depth. Options A and D are enactable now, in multiple states, with moderate political capital. They break the information monopoly but do not create the tribunal or the collective capacity that structural change requires. Options B and C create adjudicatory capacity and collective bargaining power but require more political capital, more institutional investment, and more time — during which the temporal arbitrage (135) continues to close.

Recommendation

Sequenced implementation: A+D now, B within 18 months, C as medium-term objective.

Phase 1 (immediate, 2026-2027): Enact Options A and D as a legislative package in 3-5 lead states. Frame as “worker data rights” and “algorithmic accountability” — language that fits existing consumer-protection and data-privacy legislative channels, avoids labor-law preemption triggers, and denies the opposition the “anti-innovation” framing that the neutrality permit enables. This phase breaks the information monopoly and creates the evidentiary substrate for everything that follows. The data cooperative provision is the strategic seed: it builds the organizational infrastructure that Options B and C will need.

Phase 2 (2027-2028): Enact Option B in states where Phase 1 is operational. The transparency data from Phase 1 provides the political case: once workers and regulators can see what algorithmic management does, the case for adjudicatory capacity becomes concrete rather than abstract. Frame as extending existing labor-standards enforcement (wage-and-hour, health-and-safety) to algorithmic working conditions — a regulatory update, not a regulatory expansion. Use the warehouse-quota model (California’s AB 701) as legislative template.

Phase 3 (2028-2030): Advance Option C in jurisdictions where Phase 2 has demonstrated the feasibility of algorithmic management regulation. Sectoral bargaining becomes politically viable only after the argument “the state can regulate algorithmic management without stifling innovation” has been demonstrated empirically by Phase 2. Use the Phase 1 data cooperatives as the worker-side infrastructure for sectoral negotiation.

Why this sequence: The temporal arbitrage (135) favors rapid action, but the political economy favors incremental escalation. Attempting Option C first (the structurally deepest intervention) risks the FAST Act trajectory: enactment, industry counter-mobilization, weakening through referendum or litigation. The sequenced approach builds institutional capacity, evidentiary infrastructure, and political precedent at each phase — so that each subsequent phase arrives with material backing rather than face value alone. The sequence is designed to prevent seigniorage inflation (120PB): each new right is backed by institutional infrastructure built in the prior phase.

The risk of the sequenced approach: Time. Each phase takes 1-3 years. The full sequence is 4-6 years. Algorithmic management will deepen its infrastructure lock-in during that period. This is the structural dilemma: the intervention that arrives in time (A+D) is insufficient; the intervention that is sufficient (C) cannot arrive in time. The sequence is a bet that breaking the information monopoly first (making the axiom visible as axiom) creates the political conditions for structural intervention — that 095’s observation circuit (“axiom presents itself as conclusion”) can be interrupted by forcing the axiom’s premises into public view.

The alternative — waiting for federal conditions to change — is the amnesia bet (1253): hoping the institution will be rebuilt before the memory of why it was needed has fully degraded. That bet has a negative expected value. The NLRB’s institutional memory is in Stage 3 (archival) trending toward Stage 4 (amnesia-as-opportunity). State-level action is not a substitute for federal labor law reform. It is the mechanism by which the memory is kept alive — the institutional equivalent of the union’s epistemic function: maintaining the between-layers perception as permanent orientation, preventing the naturalization of a governance grammar in which workers have no standing to contest the axiom that governs them.