Workforce Glossary — every term as a citable card.

AQO is defined as cost / (eval × outcome), where eval is the score on a sealed task bank and outcome is the verified production result. Lower AQO is better. Two systems with identical eval scores but different costs and outcomes will produce different AQO values — the score collapses the three dimensions into one comparable number.

AQO is the foundational quantity from which the WLI is constructed: a transaction publishes a verified AQO, AQOs aggregate into a volume-weighted median, and that median becomes the reference rate.

The WLI is constructed per category as a volume-weighted median over verified transactions, with a bootstrap confidence interval and a Bayesian shrinkage prior to handle low-volume categories. Methodology is versioned (currently v1.0) and published under CC-BY-4.0.

Closest analogs by function: SOFR (transaction-anchored rate), Kelley Blue Book (durable reference price), Bloomberg (citable market data).

IOSCO Principles for Financial Benchmarks (PD415) set out 19 principles covering benchmark governance, methodology, transparency, and oversight. After LIBOR, regulators effectively require IOSCO compliance for any rate cited in financial contracts.

WorkForce publishes a Statement of Compliance against PD415 at /wli/iosco-compliance; every methodology revision is re-attested.

Each published median is accompanied by a 95% bootstrap confidence interval computed from 10,000 resamples of the underlying transaction set. The CI is wider when sample size is small or dispersion is high; medians whose CI exceeds the publication threshold are held rather than released.

A simple median treats one $1 trade the same as one $1M trade. A volume-weighted median weights observations by transacted volume, so the published rate reflects where capital actually clears. This is the construction adopted by SOFR (2018) for the same reason: it is robust to manipulation by low-volume outliers.

WLI v1.0 uses κ = 30. A category with fewer than 30 verified transactions is shrunk toward the prior; categories with more transactions dominate their own estimate. This prevents a single early transaction from defining an entire category rate, and is one of the mechanisms by which the index is robust to manipulation.

Pass 1 — recompute from raw transactions. Pass 2 — perturb inputs (drop top/bottom 5%, recompute) and verify the median moves within the CI. Pass 3 — hand-validate a sample of transactions against the source_url for that price point. A median that fails any pass is held.

Pass 0 sits before the 3-pass pressure test. It verifies each transaction has a verifiable counterparty, a non-zero settled amount, and a source matching its declared tier. Pass 0 failure rejects the input from the dataset entirely — it never reaches median construction.

The distinction matters because survey-based rates (LIBOR being the cautionary tale) can be manipulated or simply guessed. A transaction-anchored rate cannot exist without a corresponding settled exchange of money for work — every basis point traces to a Stripe-confirmed payment.

Methodology revisions follow semver: a patch version fixes typos and clarifications, a minor version adds non-breaking inputs (e.g. a new category), a major version changes how the median is constructed. Every version is published immutably and remains addressable at its versioned URL.

Tier A is the only tier that contributes directly to published medians under IOSCO compliance. Every Tier A row has a Stripe payment intent, a verified counterparty, and a logged eval/outcome pair. The marketplace is, by design, a Tier A factory.

Tier B data comes from contractually attested third parties (resellers, marketplaces, integrators) under a data agreement. It supports Tier A — for example, by validating directionality of category movement — but does not itself anchor a published median.

Tier C is what you can scrape from a vendor’s published pricing page. Useful as context (sticker price), but not what buyers actually pay after volume discounts, custom contracts, and rebates. Used in /compare surfaces and analyst-facing context, never in WLI medians.

Analyst estimates, “industry sources say” figures, and panel-derived guesses are Tier D. They appear in commentary and editorial only, never as a contributor to a quantitative output. This is the explicit anti-LIBOR commitment.

An agent has a single role (e.g. “AI SDR”, “support resolver”), a sealed eval score on the corresponding task bank, and a price. It can be embedded in workflows or teams. Marketplace fee on agent transactions: 15%.

Where an agent is a worker, a skill is a verb: “draft brief”, “qualify lead”, “reconcile invoice”. Skills are composable and can be reused across many agents. Marketplace fee on skill transactions: 15%.

A workflow encodes the steps, branches, and approvals required to take a job from input to verified output. Workflows are evaluated end-to-end on a sealed task bank. Marketplace fee on workflow transactions: 10%.

Teams package multiple agents under a shared brief, with internal coordination rules and a team-level AQO. Marketplace fee on team transactions: 5% — the lowest tier, reflecting the operational complexity of the seller and the high cart-value of the transaction.

Each marketplace category has a sealed bank of 50 reference tasks at a specific version (e.g. cs-resolution/v1.2). Banks are sealed before publication — sellers cannot see them. This is the foundation of comparable eval scores across vendors.

A listing’s AQO score is its computed AQO at the time of last eval, displayed alongside its eval percentile and median outcome rate. Buyers can sort and filter the marketplace by AQO — the lowest AQO is the most cost-effective verified solution for a category.

A commodifiable output is a discrete deliverable that can be priced per-unit and judged against an objective rubric (a resolved ticket, a qualified lead, a translated document). Non-commodifiable work (strategy, taste, judgement calls) sits outside the index by construction — and that is the boundary the index honors.

The shift from per-seat SaaS pricing to per-unit AI labor is what makes a labor rate computable. When the buyer pays for a resolved ticket rather than a software license, the price-per-output is observable and a market rate can be aggregated.

SOFR is the reference rate for dollar funding. The WLI aims to be the reference rate for AI labor categories. Once a reference rate exists and is institutionally trusted, contracts can be written against it — and the rate becomes durable infrastructure.

A median is held when its bootstrap CI is too wide, its sample is too small, or it failed a pressure-test pass. Held figures remain internal and are not citable. The discipline of holding a number rather than publishing it weakly is one of the central commitments of the methodology.

When a procurement document or a financial model references WLI cs-resolution v1.0, the citation contract guarantees that v1.0’s definition, inputs, and computed values remain immutable and addressable. Subsequent methodology versions do not invalidate the citation — they coexist.

CC-BY-4.0 allows any party to use, cite, redistribute, and build on the methodology so long as attribution is preserved. This is the explicit commitment that the methodology is a public good — a precondition for institutional adoption and for the methodology to be reviewable by regulators, journalists, and standards bodies.

Enterprise procurement reviews typically require an explicit residency commitment. WorkForce stores all transaction and customer data in US-East-1 (Virginia) by default; EU residency (Frankfurt) is available to enterprise customers under a signed addendum.

A DPA codifies processor obligations under GDPR (and analogous regimes): purpose limitation, sub-processor disclosure, breach notification, return/deletion on termination. WorkForce’s standard DPA is published at /legal/dpa and is the artifact required for enterprise onboarding.

SOC 2 is the AICPA control framework covering security, availability, processing integrity, confidentiality, and privacy. Type 1 is a snapshot; Type 2 covers a 3-12 month observation window and is what most enterprise procurement teams require. WorkForce’s SOC 2 trajectory is published in the trust center.

GDPR-compatible means lawful basis is documented, data-subject rights (access, rectification, erasure, portability) are honored, cross-border transfers use approved mechanisms (SCCs or adequacy), and a DPA is in place with every processor. EU residency is available; the default DPA references SCCs for transatlantic transfers.