The Manifesto
Seven shifts.
Individual → Team. Trait → Situational. Hire → System. One view → Many. Score → Rubrics. Point → Range. Asocial → Social.
These are the moves the field has to make if it wants to take human performance seriously. Performix is built on the assumption that all seven are correct — that the current practice is wrong in seven specific, demonstrable ways, and that an instrument built on the corrected practice is materially more useful than one built on the dominant practice.
The rest of this page is the argument. The full tables are at the bottom.
M1
The conjunction problem.
See the proofHide
M1
The conjunction problem.
A capable person who has no clarity about what success looks like, no reason to care, and no permission to do the work as it should be done will not produce performance — no matter how capable they are. This is the conjunction problem. Capability becomes realized performance only when alignment, motivation, and support are also present. Drop one of the four, and the system stalls.
This is not a metaphor. It is the load-bearing model behind every diagnostic Performix produces. We call the four conditions CAMS — Capability, Alignment, Motivation, Support. In any given team, in any given quarter, one of the four is doing less work than the others. That one is the binding constraint. Identifying it is the entire job.
The seductive alternative — score people and rank them — assumes capability is the constraint. Sometimes it is. Usually it isn't. A team of capable people producing mediocre results is, by orders of magnitude, the more common pattern. The mediocre result is almost never explained by the capability ranking of the people in the room.
M2
What the field gets wrong.
See the proofHide
M2
What the field gets wrong.
Most performance tools rank people. They rank them on competency, on engagement, on goal completion, on calibration scores, on 360 inputs. Ranking is mechanically what they produce, regardless of what they're called. The output is a list.
This presumes the constraint is individual. If you can identify the lower-performing people and improve them — or replace them — performance will follow. That presumption is false often enough that the entire ranking apparatus is a category error in most of the situations it gets deployed in.
The other category error is the dashboard. A dashboard is a confession that the diagnostician doesn't know which signal matters, so it shows all of them and lets the reader sort it out. Reading a dashboard is performance-of-rigor, not rigor. The reader leaves with a feeling, not a constraint.
The corrected posture is to treat performance loss as a property of the system the people are in — and to identify the one condition that, relieved, would unblock the rest.
M3
Why employees can't say it.
See the proofHide
M3
Why employees can't say it.
FEEDBACK, WHEREVER YOU WORK
The same rating lives wherever you work — on a document, on a CAMS statement, on whether an intervention fits. You answer in place; there's never a separate survey.
Almost every signal that would let a leader diagnose a team's binding constraint sits in the heads of the people on that team. They know whether the goal is unclear. They know whether the rewards are misaligned. They know whether the tools are broken. They know whether the manager is the bottleneck.
And they cannot safely say so. Every channel that looks safe carries attribution risk. A one-on-one is attributed to the speaker by definition. An engagement survey aggregates at small enough N that a manager can guess which two people answered candidly. An exit interview happens after the speaker has already lost the room. An anonymous tip line is anonymous in name only — anyone who has worked in an organization knows this.
The thing that unlocks the channel is not asking better questions. It is a different privacy primitive: cryptographic tokenization, minimum-N gates, suppression rules — engineered so that the speaker cannot be re-identified and cannot believe they will be re-identified. Protected feedback. The diagnostic is impossible until that primitive is in place. With it, the diagnostic becomes the easiest part.
M4
Not "what's wrong" — "what's binding."
See the proofHide
M4
Not "what's wrong" — "what's binding."
A leader who asks "what's wrong on this team" gets a laundry list. The laundry list is the problem. Twenty-four things are imperfect; acting on twenty-four things is acting on none of them. The list does not tell the leader where to start, and so the leader does not start.
The discipline is to ask a different question: of the things that could be relieved, which one — if relieved — would unblock the others. That question has a single answer. Almost always. Not because the team has only one issue, but because most issues are downstream of a smaller number of conditions, and one of those conditions is binding.
In economics this is called the theory of constraints. In organizational practice it is mostly absent. The reason it is absent is that finding the binding constraint requires the kind of survey-plus-system signal that no off-the-shelf tool produces. Performix produces it. That is the product, restated.
M5
What corrected practice looks like.
See the proofHide
M5
What corrected practice looks like.
Support is the binding constraint for the Platform Group.
Support is starving primarily on two subconstructs: cross-functional response time and tooling fit. The capability and motivation evidence is strong; alignment is within normal range.
Convene a 45-minute working session with engineering and product to identify the top three cross-functional handoffs introducing the most delay; assign a single owner per handoff with a two-week resolution target.
Remeasure Support in 4 weeks (12 items, 8 min).
Corrected practice is a diagnostic, not a dashboard. One question per team: which condition is starving, and what is the accountable action. That question gets a card, not a worksheet. The card names the binding constraint, names the recommended action, names who owns it, and names when to remeasure.
The dashboard mindset asks "what is the data telling us?" — and then everyone leaves the room with a different answer because the data is ambiguous and the room is political. The diagnostic mindset asks "what is the onething we will do?" — and the answer is unambiguous because the diagnostic is structured to settle on one.
Remeasurement is the close of the loop. A diagnostic without remeasurement is performance theater. A pulse against the same dimension, four to twelve weeks later, tells the team whether the action worked. If it did, the binding constraint moves to the next condition. If it didn't, the action was wrong — or the diagnosis was. Both are useful. Both are actionable. Neither is a quarterly survey re-run.
M6
What this is NOT.
See the proofHide
M6
What this is NOT.
The category, and the posture opposite it
- notAn engagement survey→Measures conditions, not a feeling
- notBI→Shows a binding constraint, not a chart
- notSurveillance→Reduces attribution risk to engineering-grade zero
- notAn AI HR copilot→Runs psychometric methods, not prompt-completion
- notCoaching→Operates on the team-as-system, not the individual
- notConsulting→A product on a cadence, not custom recurring work
The category Performix sits in is unfamiliar enough that the question we get most often is what it isn't. Six clarifications.
It is not an engagement survey. Engagement is a feeling. Performix measures conditions. The two correlate, but they are different objects.
It is not BI. BI shows data; Performix shows a binding constraint. The output of BI is a chart. The output of Performix is a decision.
It is not surveillance. Surveillance increases attribution risk. Protected feedback reduces it to engineering-grade zero. They are opposites.
It is not an AI HR copilot. An AI HR copilot generates text in response to prompts. Performix runs psychometric methods on precomputed substrates. The output is measurement, not commentary.
It is not coaching. Coaching operates on the individual. Performix operates on the team-as-system. The diagnostic that points to "Support" doesn't tell anyone to be coached; it tells the org to fix what the team needs to do its job.
It is not consulting. Consulting is custom and recurring. Performix is a product and runs on a cadence.
M7
AI in its rightful place.
See the proofHide
M7
AI in its rightful place.
There are two confused things people are reacting to when they hear "AI for performance." This is the corrected posture on both.
AI is not the diagnostic.A language model cannot tell you whether a team's binding constraint is Capability or Alignment, because a language model is not a measurement instrument. It generates plausible-sounding text in response to prompts. Plausible is not measured. The diagnostic is run by real psychometric methods — IRT for adaptive item selection, MaxDiff with balanced incomplete block designs for forced-choice trade-offs, Monte Carlo with EVPI/EVSI for value-of-information gating, Wilson confidence intervals for the result. None of those are LLMs.
AI is, however, a first-class consumer of the substrate around it. The data work that surrounds a diagnostic — connecting source systems, mapping fields to a common shape, generating candidate items, extracting evidence from documents, summarizing team narratives — is exactly the work AI agents are now good at. Performix is designed for AI to do that work, and to do it against typed, machine-readable contracts rather than against fragile pixel-scraped surfaces.
Both things are true at once. The engine is not AI. The platform is, deliberately, AI-native around the engine. The mistake the field is making is conflating the two — using AI where measurement is required, and using human consultants where pipelines should be agents. We reverse both.
M8
Seven shifts in the philosophy of performance measurement.
See the proofHide
M8
Seven shifts in the philosophy of performance measurement.
Philosophy of performance measurement
- 01Performance is best understood at the individual level→At the team level
- 02Performance and talent are global, persistent traits→Situational — shifting with tasks and conditions
- 03Hiring higher-quality people produces better performance→Performance is a system with constraints
- 04Performance can be safely evaluated from a single perspective→Only by combining many diverse perspectives
- 05A 1-to-5 score is a measure of performance→Measurement requires task-oriented rubrics
- 06Performance measures are expressed as single scores→Expressed as confidence ranges
- 07Evaluation of human performance is uninfluenced by sociological conditions→Must take social-psychological dimensions seriously
Each of these is contested somewhere in the field, and each is settled in the research. The instrument we build is the instrument that takes the settled positions seriously and operationalizes them.
M9
Seven shifts in the analytics.
See the proofHide
M9
Seven shifts in the analytics.
Analytics of performance measurement
- 01Data first, theory after→Theory first; data operationalized in a tool
- 02Dashboard→Diagnostic — many KPIs collapsed to one binding constraint
- 03Point estimate→Confidence range (Wilson CIs, Monte Carlo bands)
- 04On-demand calculation hidden behind UI→Precomputed metrics accessible to algorithms and AI
- 05Fixed item bank→Adaptive narrowing — IRT-selected by what we already know about your team
- 06Anonymous opinion→Protected feedback — cryptographic primitive with min-N gates
- 07Systems analytics and surveys disconnected→Fused with behavioral science across both
These are the moves on the methodside that parallel M8's moves on the philosophy side. A serious philosophy of performance measurement that runs on dashboards-of-KPIs cannot work. The two have to move together. Performix is the instrument that moves them.
Each shift, drilled down — the analytical move shown as a diagram, with the surface that proves it in the product.
Data first, theory after→Theory first; data operationalized in a tool
Product capture · TBD
A live screenshot of the CAMS model behind /diagnose goes here once the app capture pass lands.
The construct comes first. CAMS (capability · alignment · motivation · support) is a theory of why capable people don't produce; the tool operationalizes it. Data is collected to test the theory, not mined for a story after the fact.
Dashboard→Diagnostic — many KPIs collapsed to one binding constraint
Product capture · TBD
A live screenshot of the live binding-constraint card on /diagnose goes here once the app capture pass lands.
A dashboard shows every KPI and asks the leader to decide. The diagnostic collapses them to the one binding constraint — argmin over the reliable CAMS dimensions — and applies resource only there.
Point estimate→Confidence range (Wilson CIs, Monte Carlo bands)
Product capture · TBD
A live screenshot of an index meter with its CI on a results surface goes here once the app capture pass lands.
A single number hides its own uncertainty. Every score ships with its interval — Wilson CIs on proportions, Monte Carlo bands on simulated outcomes — so a difference that isn't real reads as not real.
On-demand calculation hidden behind UI→Precomputed metrics accessible to algorithms and AI
Product capture · TBD
A live screenshot of the substrate read by the Insight Player + report engine goes here once the app capture pass lands.
Calculations buried behind a UI can't be reused. Performix precomputes its metrics into a substrate that algorithms, AI, and every surface read from the same source — measurement first, presentation second.
Fixed item bank→Adaptive narrowing — IRT-selected by what we already know about your team
Product capture · TBD
A live screenshot of adaptive item selection on /diagnose goes here once the app capture pass lands.
A fixed questionnaire asks everyone everything. Adaptive narrowing (IRT) selects the next items from what's already known about the team — fewer questions, more signal.
Anonymous opinion→Protected feedback — cryptographic primitive with min-N gates
Product capture · TBD
A live screenshot of an n-gated, suppressed cell on /attrition goes here once the app capture pass lands.
Anonymity is a promise. Protected feedback is a primitive: responses are tokenized and rollups are gated by a minimum-N floor, so a cell below the threshold is suppressed, never shown.
Systems analytics and surveys disconnected→Fused with behavioral science across both
Product capture · TBD
A live screenshot of fused systems + survey inputs in a composed insight goes here once the app capture pass lands.
Systems analytics and surveys usually live in separate tools. Performix fuses them under one behavioral-science model, so observed behavior and self-report inform the same diagnosis.
The commitment
One condition, one action, one owner, one remeasurement.
The cost of getting performance measurement wrong is not academic. It is the cost of a team of capable people producing mediocre results, quarter after quarter, while the organization tries one more ranking exercise, one more engagement survey, one more dashboard. Each round is expensive and produces no constraint. The leader leaves with a feeling.
What we are committing to is the opposite of a feeling. Each diagnostic Performix produces names one condition, names one action, names one owner, names one remeasurement. If it is wrong, it is wrong in a way you can detect — the remeasurement will say so. If it is right, the team moves. That is the deal.
The seven philosophical shifts and the seven analytical shifts are not slogans. They are commitments. The rest of this document shows what honoring them looks like in operator experience, in the literature, and in the platform we are building. Read whichever serves your question. The goal is not to persuade. The goal is to be exact.