HUMANS ≠ METRICS. Universe ↗
A FIELD GUIDE TO THE UNCOUNTED VOL. 01 / 2026

HUMANSARE NOTMETRICS.

A number can be true, useful, and still not the whole truth.

THE CENTRAL QUESTION ↓

When a dashboard improves, what else might be changing beyond its frame?

Enter the laboratory Download the free field guide Branded PDF · 16 metrics · practical worksheets
SCROLL TO EXPERIMENT
00 / A necessary distinction

“Measurement is a flashlight, not a portrait. The problem begins when we mistake the lit patch for the whole room.

NOT ANTI-DATA. PRO-CONTEXT.
01 / Interactive experiment

THE VISIBILITY
ENGINE.

Run a deliberately simplified thought experiment. Change a fictional organisation's target pressure. The visible metric might rise—even as other things worth caring about move differently.

CHOOSE A FICTIONAL WORLD
FICTIONAL / ILLUSTRATIVE NUMBERS
EXPERIMENT 001

The exam factory

A school wants more students to pass. That can be a genuinely useful goal. What if passing becomes the only goal anybody is asked about?

35%
Learn & observeTarget dominates

Drag or use arrow keys. Values illustrate one possible trade-off, not a forecast or a general law.

THE RULE

Notice the difference between what is rewarded and what was wanted. Then ask whether the omitted indicators are actually being measured.

PUBLIC PERFORMANCE DASHBOARDNorth Street School / fictional
TOY SCENARIO
HEADLINE KPI

Students passing the exam

82%
Students passing82%
?—
?—
?—
PRESSURE SWEEP ILLUSTRATIVE MODEL / 0–100% SCALES
Hypothetical measures across target pressureThe visible metric rises as pressure increases in this invented example. A companion outcome may rise slightly and later decline. Values are not forecasts.
Headline proxy Companion outcomeCURSOR = YOUR SETTING
03 DIMENSIONS NOT IN THE HEADLINE
MODEL STATUS: ILLUSTRATIVE, NOT PREDICTIVENOT AN ASSESSMENT OF PEOPLE

The lesson is not that the left-hand number is fake. It may be perfectly accurate. The lesson is that improving an observed proxy does not, by itself, establish that the underlying goal improved. Different mechanisms—and different evidence—are possible.

Why distributions matter ↗
02 / The instrument library

SIXTEEN SIGNALS.
SIXTEEN SHADOWS.

Each indicator earns its place because it answers a real question. Each needs a companion question before it is asked to judge a person or institution.

SELECT A CARD TO INSPECT WHAT THE NUMBER CAN'T TELL YOU USEFUL SIGNAL   BLIND SPOT
16 of 16 instruments

Sixteen instrument notes are loading. The core premise: no number contains its own context.

The right response is not to discard a number; it is to name its scope, add context, and invite correction from people affected by its use.

03 / Four illustrated case files

WHEN THE TARGET
TAKES THE WHEEL.

These are original, fictional illustrations of plausible incentive mechanisms—not reports of specific organisations, experiments, or universal effects.

CASE 01 — THE CLASSROOM

A school that mastered the test

The school's pass rate rises after teachers align lessons with the exam. Some of that may reflect better teaching. But a school rewarded only for pass rates could narrow its curriculum, drill familiar question formats and spend less time on students who need open-ended help.

Missing column: Can students explain a novel problem six months later? Who stopped receiving attention?

BETTER MEASUREMENT Pair assessment results with new-task performance, curriculum breadth and subgroup outcomes.
CASE 02 — THE SERVICE DESK

The fastest unanswered question

A team halves its time to first reply. For urgent enquiries, this is valuable. But if agents must optimise only that stopwatch, they may send early acknowledgements while difficult problems remain untouched or bounce between queues.

Missing column: Was the problem resolved without reopening? Could the person explain their need without repeating it?

BETTER MEASUREMENT Track resolution quality, reopened cases, complexity and respectful customer feedback.
CASE 03 — THE CREATIVE COLLECTIVE

Audience growth, shrinking room

A creative collective grows by publishing what gets shared most widely. More reach may bring collaborators, income and opportunities. But if every discussion is judged by follower growth, quieter experiments and difficult ideas can become harder to justify.

Missing column: Are members learning, dissenting and participating freely? Is attention converting into meaningful exchange?

BETTER MEASUREMENT Combine reach with repeat participation, voluntary feedback and room for experimentation.
CASE 04 — THE WORKPLACE

Everyone stayed. But why?

A company celebrates high retention. It might indeed reflect trust, good conditions and satisfying work. Or some employees may be staying because options elsewhere are scarce, or because their teams are too stretched to search. The same rate admits different stories.

Missing column: Do people have agency, fair workloads and genuine reasons to stay?

BETTER MEASUREMENT Study retention alongside job quality, confidential feedback and voluntary versus constrained staying.
04 / A pocket statistics clinic

THREE WAYS
TO MISREAD A NUMBER.

Technical vocabulary is not a prerequisite for statistical caution. Try the two small experiments, then borrow a dashboard audit you can use anywhere.

LESSON 01 / DENOMINATORS

“90% success!”
Of whom?

A percentage always divides one count by another. Change the question, and the apparent winner can change too.

Program A9 successes / 10 participants
90%
Program B75 successes / 100 participants
75%
A has the higher rate; B helped more people in total. Neither observation tells you who was eligible, how participants were selected or how “success” was defined.

Both programs and figures are invented. Different denominators answer different questions.

LESSON 02 / SELECTION BIAS

The survey was positive.
Who replied?

Imagine 100 members: 50 liked an event, 50 did not. But 45 of the first group and only five of the second return the survey.

AMONG 50 RESPONDENTS90%
LIKED THE EVENT
The response rate is 50%. A 90% positive result among respondents does not equal 90% positive among all members. Here, it was 50% overall.

Invented population and response pattern. Real surveys require appropriate sampling and nonresponse analysis.

LESSON 03 / DASHBOARDS

Ask six questions
before trusting the green arrow.

A dashboard is an argument about what matters. A better dashboard shows its seams.

0 of 6 questions considered. This checklist is for thinking, not scoring people.
05 / Put the idea to work

WRITE A BETTER
METRIC MEMO.

Make one decision more thoughtful. Write down what you want to improve, what you actually count, what remains unseen and what evidence would change your mind. Everything stays in this browser; no details are submitted.

Local only. No accounts, scoring or tracking.

A TAKEAWAY WORTH KEEPING EDITION 01 · 2026
SAHIL TAYAL / FIELD GUIDES
H≠M
HUMANS
ARE NOT
METRICS.

THE MISSING COLUMN

16 SIGNALS
04 CASES
WORKSHEETSsahiltayal.com
06 / Your free companion publication

TAKE THE
QUESTION
WITH YOU.

Not a one-page summary. A substantive, printable field guide for students, educators, founders, managers, researchers and anyone who has ever been reduced to a dashboard.

  • 16 indicator autopsies: what each signal can and cannot say.
  • Four detailed fictional case studies with counter-explanations.
  • Worked examples of denominators, sampling and incentives.
  • Reusable audit questions, conversation scripts and blank decision worksheets.
↓ DOWNLOAD THE COMPLETE FIELD GUIDEPDF • Works offline • Free to read and share with attribution

By Sahil Tayal · Part of sahiltayal.com · No form, paywall or email required.

07 / The thing worth keeping

A BETTER
DASHBOARD
HAS DOUBT
BUILT IN.

Counting can protect people. It can reveal exclusion, track resources, identify preventable harm and hold institutions accountable.

But a proxy is an answer to a narrow question. Grades can tell us something about performance on an assessment. Revenue per worker can tell us something about output under a chosen accounting convention. Neither tells us the full worth of a student or employee. Nor should either pretend to.

The craft is to use several kinds of evidence, inspect who is missing, clarify the denominator, learn from exceptions and ask whether people can challenge the interpretation. A wise dashboard is not silent about what it cannot see.

MEASURE CAREFULLY. INTERPRET HUMBLY.
08 / Research desk

SHOW YOUR
WORK.

Fictional scenarios are labelled as such. The following readings support the methodological distinctions—not the invented numerical outcomes of the simulations.

  1. [01]
    OECD — How's Life? 2024, state of well-being

    Why averages can mask distributional differences, including deprivation and group gaps.

  2. [02]
    OECD — Guidelines on Measuring Subjective Well-being (2025)

    Survey design, question wording, multiple dimensions and reporting beyond a mean.

  3. [03]
    OECD — PISA 2022 Reader's Guide

    How test score scales work and why their units require careful interpretation.

  4. [04]
    Our World in Data — Work and Employment

    Why statistical definitions and administrative versus survey coverage matter to comparisons.

  5. [05]
    Our World in Data — The story of the COVID-19 dataset

    How real data work involves reporting gaps, inconsistent sources and transparent limitations.

  6. [06]
    Bradley et al. — Big surveys can still be biased

    Research example: a large sample does not automatically fix selection or nonresponse bias.

Evidence note. The simulations and all four case files on this page are authored teaching examples. They do not represent a real organisation, empirical causal estimate, or claim that a metric inevitably produces a bad outcome. Source links provide further reading, not post-hoc proof for fictional events.

09 / Open correspondence

Have a thought, challenge,
or idea? Write to Sahil.

Disagreement, better evidence and excellent counterexamples are welcome.

Write to Sahil Download the free PDF field guide Explore the Sahilverse EMAIL OPENS YOUR OWN COMPOSER · NO MESSAGE IS SENT AUTOMATICALLY