How the Pharometric Index is computed, and the conditions under which it is wrong.
A measurement is only as useful as it is trustworthy, and trust in a measurement comes from being able to inspect how it was made and to check whether it holds. This page describes how the Pharometric Index is computed, what it does and does not claim, and — most importantly — the specific conditions under which the index should be considered wrong. We publish the falsification conditions first-class, alongside the method, because a measure that cannot be proven wrong cannot be trusted to be right.
We state uncertainty explicitly throughout. Where we do not yet know something, we say so. Where a choice is a judgment call, we name it as one.
01
The first Pharometric Index measures contraction in entry-level hiring opportunity — the rate at which early-career workers are being hired into an occupation, in a place, relative to that same occupation-and-place's own pre-2020 baseline, adjusted for the broader hiring cycle.
It measures contraction. It does not, by itself, measure cause. A falling index value means fewer young people are being hired into that occupation in that place than the baseline and cycle would predict. It does not, on its own, establish why. Attribution to AI is a separate, explicitly weaker layer described in Section 5.
Every value is expressed on a common 0–1 scale: 1.0 means hiring at or above the occupation-place baseline; 0.0 means the worst contraction observed in the reference period. This inversion to a shared scale is what allows different metrics, sources, and domains to be compared and combined.
Values are computed per occupation family × geography × time period. Geography is reported at the county level (the native resolution of the primary source) and mapped to a hexagonal grid for spatial analysis, with an explicit fidelity penalty applied to any sub-county disaggregation (see Section 4) — we do not manufacture precision the source data does not support.
02
The index is computed from public and licensed data. We publish only derived values; we do not redistribute raw source data.
Quarterly Workforce Indicators (QWI), U.S. Census Bureau — hires by county, industry, and worker age band, from administrative records. Public domain. This is the backbone; it is administrative (not survey-sampled), which makes it hard to game but introduces a publication lag of roughly two to three quarters.
03
Multiple sources covering the same quantity are combined using a generalized least squares (GLS) estimator with an explicit source-covariance matrix. The covariance matrix encodes a fact most data products ignore: sources are not independent. When two sources republish the same underlying survey, they are two correlated witnesses, not two independent confirmations, and naive averaging would over-weight their agreement. The covariance structure corrects for this.
The estimator and the covariance assumptions are part of the open Pharometric standard. The calibrated parameters — the specific source weights and decay rates, tuned against observed outcomes — are not public; they are the part that improves with each cycle.
04
Every published value carries a fidelity score: an explicit, calibrated statement of how much the value can be trusted, given how many independent sources constrain it, how recent and granular they are, and how much they agree after correlation correction.
Fidelity decays when a place or metric goes unmeasured. A region not measured for several cycles has its confidence reduced — so neglected areas are visibly flagged as low-confidence rather than silently presented as if current. Absence of data is shown as low fidelity, never as zero and never as "no problem." A place we cannot see is reported as a place we cannot see.
Sub-county (hex-level) values inherit a fidelity penalty proportional to the disaggregation. We do not present interpolated precision as if it were measured.
05
The index measures contraction. Attributing that contraction to AI is a separate claim, held to a higher standard, and reported with greater uncertainty.
We cross contraction against task-based AI-exposure scores derived from public O*NET data. Where high contraction coincides with high AI-exposure, we label the cell an "AI-attributable candidate" — candidate, not conclusion. We do not publish a causal percentage. We do not claim that source independence proves causation: two independent sources agreeing that youth hiring fell in AI-exposed occupations raises confidence that the pattern is real; it does not, by itself, rule out a common cause such as interest rates, offshoring, or a sector-specific downturn.
The identification work — separating AI-driven contraction from cyclical and structural alternatives — is carried by the cycle adjustment (Section 6) and the falsification test (Section 7), not by the agreement of sources.
06
Hiring is cyclical. To avoid reading an ordinary downturn as displacement, each occupation-place value is adjusted against broader hiring conditions, including the hiring trend for experienced (mid-career) workers in the same occupation and place. The intent is that the index reflects youth-specific contraction — a gap that opens between early-career and experienced hiring — rather than a general decline that affects everyone. The exact adjustment is specified in the technical appendix (forthcoming) and is itself subject to the validation in Section 7.
07
How to prove the index wrong.
We publish these first-class. If any of the following holds, the index is not measuring what it claims, and we will say so publicly:
08
09
Published values are timestamped and immutable. Errors are corrected in the next release with a public changelog — we do not silently edit past values. Methodology changes receive a version number and a one-period parallel run against the prior version so the effect of the change is visible.
v0.1 (draft) — Initial public methodology. Open verification items flagged: QWI age-band / entry-level proxy (Section 2); cycle-adjustment specification (Section 6); technical appendix forthcoming. No live index values published yet; this document precedes first publication.
Pharometric is intended as a standard for expressing human conditions measured under uncertainty. This methodology, and the schema it describes, are published openly so that any value carrying the Pharometric name can be independently inspected and verified. Calibration parameters and the live data feed are maintained by Clattershop LLC as the reference instance.