principle · witnessed, not described

Witnessed, not described

Sit next to the person while they do it. Description is downstream of seeing.

People are unreliable narrators of their own work — not because they lie, but because they cannot see themselves. A grader does not notice that they alt-tab to a spreadsheet to fix Hebrew names because by year three of the workaround it has become invisible to them. Asked what slows your grading down?, they will not say Hebrew names. Asked can you walk me through what you did at 09:14?, they will tell you precisely, because the question anchors them to the witnessed moment.

This is the chain's first input discipline. Discovery does not run on what people say they do. It runs on what we saw them do.

The three-rank order

The corpus puts inputs in this order:

1. Observation in the field. Sit next to the named person, in their environment, while they do the activity. Take time-stamped notes. Ask only post-hoc questions, anchored to specific timestamps.
2. Recorded session of the activity. Same activity, captured. You miss the environment, the interruptions, the side conversations. You keep the workarounds.
3. Interview. A description of the activity. Useful as fill-in. Never load-bearing.

A brief built on observation has a different shape than one built on interview. The brief built on interview tends to produce features that solve the spoken problem and miss the witnessed one.

When observation feels too expensive

It is almost always cheaper than the alternative. A 90-minute observation session that prevents a six-week feature build solving the wrong problem is the cheapest investment the chain makes. The math is not subtle; the discipline is being willing to spend the 90 minutes when the team is impatient to start.

If the named person is in a regulated environment you cannot enter, the corpus's fallback is recorded session with their permission, then colleague-shadow, then interview anchored to specific recent moments. What does not substitute for observation: a stakeholder's description of the named person's work. That is description twice removed.

What surfaces when you witness

Three classes of signal — all expensive to extract by interview, free at observation.

• Workarounds. The bridges the person built over gaps the system left. Workarounds are evidence of pain that has already been measured by the person paying it.
• Friction. Pauses, alt-tabs, the let me find that, the post-it. Often unmentioned because internalised.
• Domain language. The actual words the person uses. Usually different from the words on screen. Always different from the words in the marketing deck.

A brief that captures these three is the brief that Volume V's check will be able to read. Without them, the prediction is a description of a system change. With them, it is a description of a change in a named person's day.

See also

• Canon — Before We Build · Observation
• Principle — Person-first
• Practice — Running an observation session (coming)
• Clinic — A brief that didn't witness