Measuring yourself without shaming yourself: the quantified-self trap

Self-tracking promised self-knowledge. What often arrives instead is a little
judge in a widget: a score going down, a streak broken, a red arrow where your
Tuesday used to be. The quantified-self movement was built to make behaviour
visible; most fitness and social dashboards quietly turned visibility into
evaluation - and evaluation is where the psychology curdles.

Goodhart's law, applied to yourself

Goodhart's original observation was about economics: when a measure becomes a
target, it ceases to be a good measure.¹ Streaks were meant to
describe habit; they became the thing users protect, sleeping beside their
phone to keep a flame alive. Follower counts were meant to describe reach;
they became the thing people buy, beg and restructure their faces for.

The mechanism is well documented in self-regulation research: surveillance of
the self by a metric narrows behaviour toward the metric rather than toward
the underlying goal. When you watch a number, you optimise the number - and
the number is only a shadow of what you cared about (connection, expression,
curiosity). This is the trap:

1. You measure to understand.
2. The measure needs a scale to be readable.
3. The scale implies good and bad.
4. You start serving the scale.

A 0-100 "engagement score" completes step 3 by design. It has baked into it
someone's opinion of how much you should post, reply, be replied to. The
number looks like a thermometer - neutral, clinical - but thermometers don't
judge you for running a fever.

Self-concept injury: when the dashboard talks back

Why does it matter whether the number has an opinion in it? Because metrics
about social behaviour feed directly into self-concept - your working theory
of who you are. Psychology has long observed that self-discrepancies (the gap
between your actual self and the self you believe you should be) generate
distress, and that the size of the reference point drives the size of the
pain.² A neutral count creates a small reference point ("I posted
40 times"). A scored grade creates a large one ("I am at 62/100 - a C").

The result is a two-sided failure that shows up in every quantified-self study
worth reading:

• Under-performance shame - the number is low, and because the number is

  the target now, you are low. Deltas get read as verdicts, not observations.

• Over-performance capture - the number is high, and you become afraid to

  change behaviour in any way that might lower it. The streak ends you, not the
  other way around.

Both outcomes are forms of the same error: treating a descriptive statistic as
an evaluative one. And both are designed in whenever a product renders your
life as a single graded scale.

The refusal this product was built on

This is not a philosophical preference floating above the code - it is a
decision in the ledger with observable consequences. The decision D5 bans
composite 0-100 scores outright;³ the practical consequence across
Insights, Compare and Features is counts, medians and deltas only - score tiles
are nowhere to be found. Compare states in its own copy that it shows a
snapshot, not a prediction of where your numbers will be.⁴ There are
no forecast rows, no "you should post more" nudges, no red/green grades on your
own history.

What replaces the score is context you supply:

• Counts with their units and time windows visible, so "12 messages" can be a

  quiet week or a reunion, and the app does not decide which.

• Deltas between two real exports - two measurements, not a projection.
• Medians and distributions instead of averages-as-targets, because a median

  describes what happened, it does not imply what should happen next.

• Your own reading of the timeline - the year-in-review and heatmap exist to

  jog memory, not to rank it.

Using your archive without becoming its subject

Practical rules that follow from the psychology and from what the code allows:

1. Read trends over windows, not spot values. A single count invites

   judgment; a span of counts invites observation. The heatmap and timeline
   views exist for exactly this.

2. Never average yourself across people you don't know the context for.

   Median reply time across ten conversations tells you about your ten
   conversations, not about your character.

3. Treat surprise as information, not verdict. "I didn't realise I

   searched for that in 2022" is the archive doing its job. "I'm at 40/100"
   is an artefact this product refuses to produce - by policy, not by omission.

4. If a metric starts changing your behaviour, shrink its surface. That is

   Goodhart working as predicted. The tool gives you filters and windows, not
   streaks and badges, precisely so the option to shrink is always available.

Related reading

The parts-we-refused-to-build post for all seven refusals with their enforcing
code; the counts post for how every metric is derived and why units and windows
are always shown; the year-in-review post for how the same numbers are staged
as memory rather than as a grade.

¹: Goodhart, C. - "Problems of Monetary Management: The U.K.
Experience" (1975), on measures that cease to work once they become targets;
restated as Goodhart's law in the quantified-self literature.
²: Higgins, E. T. - self-discrepancy theory (A Theory of
Cognition, Affect, and the Self, 1987): distress scales with the gap between
actual and reference selves - cited here to explain why a graded self-metric
hurts more than a descriptive one.
³: The project decision ledger D5 bans composite 0-100 scores;
observable consequence documented in content/blogs/2026-10-07-the-parts-we-refused-to-build.md
(section 5, with its scores footnote in that post).
⁴: Compare view copy, apps/web-next/src/views/Compare.tsx - the
explicit statement that Compare shows a snapshot between two exports rather
than a prediction.