PA Instruments (measurement + decision building blocks)

Type: algorithm Origin repo(s): people-analyst (the People Analytics Toolbox), with instruments living across anycomp, forecasting, calculus, preference-modeler, and performance-validity Extraction readiness: live across the toolbox; packaging as a named set is in progress Depends on: the per-spoke contracts and the shared metric envelope (every instrument speaks the same value + provenance + enrichment shape) Last reviewed: 2026-05-26

What it is

PA Instruments are composable measurement and decision building blocks — think ingredients — that combine into finished products — think meals. Each instrument does one thing well and exposes it over the same typed contract, reachable from a browser over HTTP and from an AI agent over MCP. You assemble them rather than rebuild measurement machinery for every new question.

There are nine today, drawn from across the toolbox.

The nine instruments

• Best/worst preference capture (MaxDiff) — respondents pick best and worst from small sets, forcing trade-offs and yielding stronger rankings than rating scales.
• Present-vs-future preference capture — separates what people value now from what they expect to value later.
• Rater alignment — measures how much a set of raters agree, beyond chance.
• Directional alignment (up / down / lateral) — measures how aligned a person is with the people above, below, and beside them.
• Distribution fitting — fits an appropriate distribution to observed data so downstream tools can reason about uncertainty, not just point estimates.
• Stated-vs-observed importance reconciliation — reconciles what people say matters with what their behavior reveals matters.
• Interval calibration scoring — scores how well someone's stated confidence intervals match reality (are their 80% ranges right 80% of the time?).
• Bayesian multi-source combination — combines evidence from several noisy sources into one better-calibrated estimate.
• Value-of-information-ranked recommendations — ranks which measurements to take first by how much each one is worth to the decision at hand.

Products built from them

The point of building blocks is what you assemble. Two example products:

• Leadership Index — assesses leaders by the quality of their predictions and how aligned they are with the people above, below, and beside them. Composed largely from interval calibration scoring and directional alignment.
• Analytics-Plan Generator — takes an organization plus its priorities, identifies the models that matter, and returns a value-of-information-ranked analytics plan: what to measure, in what order, and why each step earns its place.

Why it is shaped this way

• One contract per instrument. Every instrument carries its own typed contract and version, so a product that composes five of them can rely on each one's shape without coupling to its internals.
• Two transports, one algorithm. The same instrument answers a browser request and an AI-agent call from the same core, so the two never drift.
• Ingredients, not a monolith. A new product is an assembly of existing instruments wherever possible, plus a thin layer of product-specific glue — not a fresh measurement engine.

Related capabilities

• Compensation scenario modeling — a decision-first product that composes several of these instruments.
• Monte Carlo simulation engine — the stochastic engine several instruments draw on.
• Statistical analysis engine — the trust-the-numbers layer underneath the measurement instruments.