01
Beat 01
Slice the world by work, not by industry.
Automation maps to work, not to logos. The same UoP for legal drafting ships across financial services, insurance, and government — the alignment math is the same. Industry is downstream of work.
01 · Methodology
The methodology, at the depth you'd take to your own team.
You're here because you want to know whether the score in the Workbench is something you could justify to your own technical leadership, your CFO, or your board. This page is built to answer that — by showing the architecture, the composite logic, the confidence model, the source catalog, and the flywheel that closes the loop.
It is deliberately one level less detailed than the implementation. Exact weights, prompts, thresholds, and per-vendor allocation logic live behind the line — under change control, under quarterly review, visible to design partners under NDA. What you see below is enough to evaluate the rigor — not enough to lift the methodology straight out.
02 · First principles
Four convictions everything else derives from.
The substrate, the scoring, and the primitive all follow from these.
02
Beat 02
Enterprises are complex adaptive systems, not task lists.
Work is coordination, not tasks. Trying to match and automate tasks assumes they are static and still relevant in the new system of work. They are not. Tasks are downstream of what humans align on doing.
03
Beat 03
Human-centered — productivity captured, not deflected.
We stay neutral on the outcome — not a zero-displacement enforcer. But by surfacing how people can be redeployed around new AI capability, we give senior leaders a clear alternative to layoffs: capture the productivity gain internally through reorganization and reskilling. Every UoP carries a named redeployment plan; the decision stays with the enterprise.
04
Beat 04
Large organizations sit below their latent potential. Emergence is a recipe.
Most large organizations run well below what their people, AI, capital, and tooling could deliver together. Human productivity depends on motivation, and motivation depends on conditions — the right manager, the right incentives, the right environment, the right AI tools to augment the work. Get the recipe right and emergence happens. Get it wrong and the same resources deliver a fraction of what they could. That's why the primitive is Unit of Potential, not Unit of Plan. The plan is the floor; emergence is the ceiling.
03 · The substrate
Every score traces to the Global Labor Graph.
The Global Labor Graph is the substrate. Every score in the Workbench, every UoP candidate, every Vendor Allocation cell is computed dynamically from it — not hand-tuned, not made up.
The GLG is a joined multi-source dataset. No single vendor's product. It carries depth (TAG's proprietary placement substrate), breadth (Lightcast + public stats from BLS, ILOSTAT, OECD), and frontier signal (capability research from OpenAI, METR, Scale, Anthropic). The composition is the point: any one source can be wrong; the joined substrate is auditable.
1.28B
job postings · 16-yr depth
594M
professional profiles
104M
annual placements (TAG)
249
countries · cross-country wages
Source catalog · provenance + refresh per source
| Source | Role in the substrate | Coverage | Refresh |
|---|---|---|---|
| Lightcast Delta Share | Substrate · postings, profiles, skills, firmographics | 1.28B postings (16-yr) · 594M profiles · 28,795 skills · 7.5M+ companies | Daily |
| TAG (proprietary depth) | Substrate · workforce telemetry, placements, role transitions | 100K enterprise clients · 104M placements/yr · 300M candidate interactions/yr | Continuous |
| O*NET | Substrate · DWAs, KSA ontology, occupation reference | 19,265 DWAs · 1,016 occupations | Per release |
| BLS · OEWS · ILOSTAT · OECD · Eurostat | Substrate · cross-country workforce + wages | 249 countries · 396 metros · CPS + OEWS p10–p90 wages | Per release |
| Eloundou et al. 2023 (OpenAI) | Frontier capability · exposure structure | 100% of O*NET occupations · α/β/γ exposure | Static (research) |
| GDPval (OpenAI 2025) | Frontier capability · task capability benchmark | 1,320 tasks · 44 occupations × 9 GDP sectors | Quarterly |
| METR HCAST Time Horizon | Frontier capability · agentic horizon | Doubling cadence · 7mo → ~3mo (2024-) | Quarterly |
| Remote Labor Index (Scale AI + CAIS) | Frontier capability · end-to-end project deliverability | 240 real Upwork projects · max 2.5% frontier automation | Annual |
| Acemoglu · Frey-Osborne · Dingel-Neiman | Macro · automation caps, bottleneck residuals, teleworkability | Macro economic priors · cross-checks | Static (research) |
| WEF Future of Jobs | Workforce readiness · CHRO-surveyed signal | 1,000+ CHROs surveyed · reskilling-need trajectories | Annual |
| Fusebox AgentFuze telemetry | Live telemetry · per-UoP outcome + workforce actuals | Per-UoP real-time signal | Continuous |
| Anthropic Economic Index | Frontier capability calibration · task-level usage data | Task-level usage signal | Quarterly |
The substrate refreshes continuously. Every score that ships is anchored to a snapshot. Coverage and refresh cadence per source are listed above — and visible per-cell in the Workbench.
04 · From substrate to score
We compute dynamically from the substrate — not from hand-tuned fixtures.
The pipeline cascades: substrate signal → archetype classification → enterprise-specific readiness → gated composite → scored cell. Every Workbench score, every UoP candidate is the output of this cascade, not a stored constant.
Methodology · topology
how a score gets computed, at the system level
Tier 01
The Global Labor Graph
multi-source · provenance per signal
S
Breadth
Lightcast · public stats · O*NET
Depth
TAG placement substrate
Frontier signal
OpenAI · METR · Scale · Anthropic research
Live telemetry
Fusebox AgentFuze · per-UoP signal
provenance + refresh cadence per signal
Tier 02
The pipeline · cascade
dynamic compute · binding-constraint composite
C
Archetype classification
what work, at what density
Enterprise readiness + gating
outside-in macro · inside-out coalition
Composite + ceilings
binding constraint surfaces · caps visible · no silent downweights
scored + tier-tagged · ceiling-flagged
Tier 03
What ships to the Workbench
every cell is auditable to its substrate inputs
O
Scored cell
Alignment + Absorption · per archetype
Confidence tier
Estimated · Modeled · Configured · Measured
Reason chip
when a ceiling caps the cell
UoP candidate
scored UoP proposed to the operator
01
Compute dynamically, not statically.
Substrate signal in, real-time compute out. When the substrate refreshes — a new posting, a new placement, a new capability benchmark — the next score reflects it. No hand-tuned constants in the cells you see.
02
Binding constraint, not weighted average.
A UoP is a chain. The weakest link breaks it. The composite raises the binding constraint — the single dimension that gates the deployment — rather than averaging away chain-breakers. Weighted sums hide the failure modes the weakest signal already reveals.
03
Ceilings are explicit, not silent.
When the substrate doesn't carry enough signal for a confident score, the cell shows a ceiling cap with the reason — no sponsor, external data only, governance dimension low, archetype-volume below floor. Operators close the gap to lift the score. We do not silently downweight.
04
Multi-source provenance, per signal.
Every signal feeding the composite traces to its source in the catalog above. The blend isn't opaque; the contribution is auditable. When sources disagree, the cell reflects the discord — usually as a lower confidence tier.
The specifics — exact weights, sub-dimension counts, threshold values — are versioned and evolve as the substrate deepens. Today's pipeline matches what's defensible at today's data depth. Every change ships behind audit (see §06 Governance).
05 · What we measure
Three measurements. Two scores and a telemetry layer.
Absorption Score
Can the organization structurally take this on?
Measures the organization's capacity to absorb AI in a given work archetype: culture, processes, workforce readiness, metrics, incentives, norms. Slow-to-change, system-level. The lever when the score is low: reorganize the system of work.
Alignment Score
Are the conditions right for humans and AI to co-create value on this specific UoP?
Measures whether Value Clarity, Agency, and Substance are in place on a single deployment. Stakeholder-level, faster to move than Absorption. Both scores need to be high: a high-Absorption org with low Alignment deploys the tool and watches it sit unused; a high-Alignment team in a low-Absorption org builds a brilliant pocket that fails at scale.
Blended Telemetry
Did the deployment actually work — for the business, the agent, and the human?
Agentic telemetry alone — what the AI did, what it cost, where it succeeded — does not prove the deployment created business value. We blend three streams into one neutral measurement layer baked into every UoP. No single vendor controls the score.
01 · Agentic
What the AI actually did, per workflow step. Cost, success and failure, latency, escalation rate.
02 · Business
Did the productivity gain reach where the CFO can see it? Top-line growth, bottom-line margin, throughput, cycle time, customer outcomes the business actually feels.
03 · Human · Worker Voice
Across the frontline and middle management, was the job actually augmented? Are the people inside the deployment engaged?
Status — Work in progress. We’re instrumenting it with agent providers and enterprise customers.
06 · Confidence
Every score wears its tier.
Four tiers, each with a visible badge wherever a score appears. The tier tells you not just how much signal we have, but what kind — external priors only, or modeled with org context, or stakeholder-calibrated, or live-measured.
A CFO acting on an Estimated number believing it's Measured is the credibility-killing event. The ladder prevents that.
Estimated
Modeled
Configured
Measured
Estimatedcap ≤ 60
External priors only — Lightcast firmographics, posting density, public benchmarks, frontier-capability research. Raw substrate signal, no further enrichment yet.
Modeledmethodology-derived
Multi-source signals composed through our scoring algorithm and work-archetype ontology. Methodology-derived, not raw — but no enterprise-specific engagement yet.
Configuredenterprise-calibrated
Enterprise context layered in — sponsor identified, governance posture mapped, workflow dimension scoped. Stakeholder alignment cleared. Deployment plan committed.
Measuredlive telemetry
Live telemetry from agentic + business + human (Worker Voice) streams. Realized outcome measured against the deployment plan. Independent of the deploying entity. Audit-grade.
3,315 / 3,316
scored cells today are Estimated
1
is Measured — TAG × Harvey · Legal Drafting
The progression up the ladder is the product. Every cycle moves at least one cell up. The substrate refreshes from Measured outcomes — that's the flywheel below.
07 · The flywheel
The lifecycle compounds back into the substrate.
The methodology isn't a one-way pipeline. The substrate computes scores onto Workbench cells. Users act on those scores to generate UoPs. UoPs travel through their lifecycle. Outcomes, blockers, Worker Voice, and user behavior all feed back into the substrate. The next cycle is sharper.
This is the structural reason the platform compounds — and the reason no single vendor can credibly produce this measurement on their own.
Stage 01
Global Labor Graph
Multi-source substrate · refreshes continuously
Stage 02
Workbench scoring
Substrate computes onto every cell · with confidence tier
Stage 03
UoP generation pipeline
Scored UoPs proposed to enterprises
Stage 04
UoP lifecycle
Map · Generated · Configured · Deployed · Gated · Compounding
Stage 05
Deployment Intelligence + Realized Potential
Outcomes, blockers, Worker Voice, user behavior
Refreshes the substrate · next cycle starts smarter
Every deployment teaches the substrate. Every Gated UoP teaches the substrate. Every Worker Voice signal teaches the substrate. The moat isn't data volume — it's credibility that compounds.
08 · Governance
Change-controlled. Quarterly review. Audit-trailed.
Every weight, every threshold, every ceiling rule is owned by the methodology board. Changes ship on a quarterly cadence. The audit log captures every revision and its rationale.
- Global Labor Graph as substrate
- Dynamic compute (not fixtures)
- Binding-constraint composite
- 4-tier confidence ladder
- Explicit ceilings + reason chips
- Multi-source provenance per signal
- Peer-score structure (Absorption + Alignment)
- Agent Capability gate (per-archetype currency)
- Workforce-readiness signals (TAG depth integration)
- Vendor allocation dimension count
- Substrate refresh cadence per source
- Cross-cohort weight normalization
- Density-penalty calibration
- Workforce-volume floor for capping
- Live-telemetry write-back to substrate
When weights revise: deployed UoPs see their composite shift ±2–5 points; tier classifications hold within Ready / Gated. We track this explicitly so revisions are calibrated, not disruptive.
Full audit log available on request to methodology@rpotential.ai. Methodology canon is maintained by the methodology-architect agent under board review.