Explained · Adaptive Narrowing

Ask only what the last answer made worth asking.

Performix starts with a short CAMS instrument — the minimum load-bearing survey — then chooses each next analytic step from what your team's answers have already revealed. It only goes deeper when Value-of-Information says the added insight is worth the extra time.

This is the system-of-learning move: not a fixed 40-question survey and not an open-ended chat — a precision instrument that narrows probabilistically and gates expensive tracks behind an explicit cost-benefit check.

What Performix is best at

Revealing the issues you can't see.

A team's real constraint could be any of a thousand things. The diagnostic that could find it for certain would run to thousands of questions — and no one would finish it.

Standard engagement surveys ask the same fixed set, over and over. They confirm what you already suspected. They don't surface what's hidden.

Performix's survey algorithm narrows in. It learns from each answer and asks only the few questions that locate your team's actual binding constraint — fewer than 20, in five to ten minutes. Knowing what to ask is the hard part. That's what Performix does.

01 · The narrowing flow

CAMS minimum → dimension → subconstruct → VOI-gated depth.

See the proof

Layer 1 is the short starting survey: four CAMS dimensions, enough to name the binding constraint and estimate the value of moving it. Probabilistic narrowing de-prioritizes dimensions that can't be the bottleneck — Capability, Alignment, and Motivation fade when Support is the clear gap.

Layer 2 drills into Support subconstructs — cross-functional response time rises above tooling fit and resource sufficiency. Layer 3 is deeper analysis, but only when it's worth it: a quick remedy (convene a working session on top handoffs) ranks high value · low cost; an org-wide cross-team comparative study ranks low value now · high cost — and stays queued until the math changes.

02 · Worked example

Support at 58 — narrow to handoffs, not a rep-replacement program.

See the proof

demo team · Enterprise Sales East (illustrative)

1 · Short CAMS instrument names Support as the binding dimension — Capability trails West by only 4 pts, Support by 30 pts.
2 · Narrow to Support subconstructs — cross-functional response time selected over tooling fit and resource sufficiency.
3 · VOI gate ranks deeper tracks: quick remedy (high value · low cost) before org-wide comparative study (low value now · high cost).

Seeded demo path — illustrative sequence, not a live customer run. Support score 58 and East-vs-West spreads are from the shipped demo fixtures.

In the built-in demo team Enterprise Sales East (seeded sample data, not a real customer), the engine returns a Support score of 58 and names Support as the binding constraint — not because Support "felt low," but because it's the widest gap: East trails the demo West team by 30 points on Support and only 4 points on Capability.

The verdict ("Support, not skill") is earned by that spread, which is why the recommended action targets operating cadence — deal desk, escalation paths, manager bottlenecks — not training or hiring.

03 · The methods underneath

IRT, MaxDiff, Monte Carlo/EVPI-EVSI, Wilson — not a language model dressed as one.

See the proof

IRT (item response theory)
Selects the fewest, most informative questions — adaptive item selection from what you already know.
MaxDiff
Forced trade-off choices reveal what a team truly prioritizes among competing conditions.
Monte Carlo · EVPI/EVSI
Estimates the value of asking one more question before spending the respondent's attention.
Wilson intervals
Reports a confidence range around a score, not a single false-precision point.

Real psychometric methods · owned upstream · consumed via typed contracts

The engine is genuine psychometrics — the math behind standardized testing and clinical measurement. IRT picks the fewest, most informative questions. MaxDiff forces trade-off choices to reveal what a team truly prioritizes. Monte Carlo with EVPI/EVSI computes the value of asking one more question before asking it. Wilson intervals report a confidence range, not a single suspiciously-precise number.

These are real psychometric methods, engineered into Performix's measurement engine — not a language model dressed as one. As the engine improves, the diagnostic improves.

04 · Current status

Illustration wired; live adaptive engine coming online.

See the proof

The adaptive narrowing flow is designed, illustrated, and referenced across the explainer movements (manifesto shift #5, walkthrough Q3, research roadmap). The CAMS diagnostic runs end-to-end on seeded demo data today through Performix's adaptive psychometric engine. When live responses are connected, the same narrowing and VOI gating activates — no UI rebuild.

05 · New words

The terms, defined.

See the proof

Adaptive narrowing: Each next question or analytic step is chosen from prior responses — start at CAMS minimum, drill only where the data points.
Value of Information (VOI): Whether to gather more data is itself a decision with an expected value (EVPI/EVSI) — ask the next question only when the information it buys is worth its cost.
IRT: Item response theory — math that selects the fewest, most informative questions to reach a reliable score.
Probabilistic narrowing: De-prioritizing branches that probabilistically cannot be the binding constraint, so respondent attention stays on what matters.
Typed contract: A strict, machine-checked description of what a service accepts and returns — apps rely on it without copying internal algorithms.

Keep reading