Week 10 Studio: Numbers without sources

Week 1 turned inward. Every number, phrase and movement on your own screen, traced back to the event that produced it, or deleted.

13 May 2027

In week 1 you stood outside an interface you did not build and asked what could possibly support what it said.

Today you do it to your own. It is not the same exercise, and the difference is the point of the week.

What changed since week 1#

In week 1 you had a recording. Today you have the source, the state enumeration from week 2, the denominator decision from week 3, the estimator you chose in week 5 with its measured instability, and every failure state you can reproduce on demand.

So the question sharpens. It is no longer what could support this claim? It is which of my promises actually hold? The verdicts change with it:

Losing the third verdict is the week working. “I cannot tell” was an honest answer in week 1 and is an evasion today.

What replaces it is a fourth verdict week 1 had no use for: kept with reason. Some claim on your screen has no source event, you know it, and you are keeping it — a rounded figure, a minimum display duration, a reassuring phrase during verification. That can be a good decision. It stops being one the moment it is unexamined.

By the end of this week#

You can:

• inventory everything a stateful interface renders, across all its states and both viewports, before judging any of it;
• trace a displayed value to the event that produces it, and identify the ones with no such event;
• distinguish a defensible “kept with reason” from an excuse, and write the former;
• compare your own analytic position now against the one you had in week 1, and say what changed.

The idea#

No lecture this week. This section and the 14:00 framing are the teaching.

Inventory before judgement. The instinct is to go straight to the suspicious number. Resist it: the claims you are least likely to examine are the ones you wrote so early you no longer see them. Enumerate everything first — every string, every number, every animation, in every state, at both viewports — and only then start tracing. Most people find their worst row in something they added in week 2 and never looked at again.

What counts as a source event. Same definition as week 1: something the process reports, at a moment you can point at in code. Three near-misses that are not source events:

• A value derived from a timer rather than from progress. A bar that advances on setInterval is reporting that time passed, while appearing to report that work happened.
• A constant that was measured once, in week 5, on your machine. It was evidence then. It is a hardcoded number now, unless the code re-measures.
• A default that arrived with a library and was never chosen.

Kept with reason, versus an excuse. The test is whether a reader could argue with it. Not a reason: “it looks better”, “users expect it”, “it felt too abrupt”. A reason:

The verification state completes in under 80 ms. A state that flashes for 80 ms is harder to read than one held for 400 ms, so this state is displayed longer than it exists. The cost is that a user reading the elapsed counter will see it disagree with the state duration by up to 320 ms.

That names the deception, the justification and the price. A marker may still disagree, and that is fine — a reason you can lose an argument about is still a reason.

View this week’s slides — the 14:00 framing runs from these. They carry the worked example and the exercise brief. On a phone they are small — the deck is built for a projector, and this page carries the same material as prose.

In the room#

Thursday 13 May, 14:00–17:00. No lecture this week.

What you start with#

• Your complete prototype through week 9.
• Seven planted claims with no source event, of varying obviousness, in the stage-10 checkpoint. You will not be told how many you found.
• Your week 1 claims table, which you read again at 16:30.
• An empty page at /claims/, which ships as part of Assignment 3.

What you build#

A claims register for your own prototype, as a page in your repo at /claims/. Build it where it will ship — it is a deliverable of Assignment 3.

A worked example#

One row, fully worked, in the verdict week 1 had no use for:

The verification state completes in under 80 ms. A state that flashes for 80 ms is harder to read than one held for 400 ms, so this state is displayed longer than it exists. The cost is that a reader comparing it with the elapsed counter will see them disagree by up to 320 ms.

It names the deception, the justification and the price. A marker may still disagree; a reason you can lose an argument about is still a reason.

Not reasons: “it looks better”, “users expect it”, “it felt too abrupt”.

The 80 ms figure is the simulator’s verify stage, not a measured service.

Done when#

• Every distinct thing your interface renders has a row. Walk every state with the simulator controls, at both viewports, before you call the inventory done.
• No row has an empty verdict.
• Every kept with reason row names the deception, the justification and the cost.
• Anything marked removed is actually gone from the interface, not just from the register.

The mistake to expect#

Going straight to the suspicious number.

The claims you are least likely to examine are the ones you wrote so early you no longer see them. Most people’s worst row is something added in week 2 and never looked at since. Inventory everything before judging anything.

The second: marking a row removed in the register while it is still on the screen. The register describes the deployed interface, not your intentions.

Before next week#

About two hours.

1. Finish the register and deploy it. Week 11 tests the interface the register describes, so they have to agree.
2. Choose the one state you want watched in week 11. The default is the stall, because it is where your messages are least tested. Note the two or three things you expect a person to say when they hit it — written down before Thursday, so you find out whether you were right.

If you are joining late#

git checkout stage-10 gives you a complete prototype through week 9 — states, denominator, estimator, indeterminate handling, failure states, cancel and retry — and seven planted claims with no source event, of varying obviousness. It is a floor, not an answer. You will not be told how many you found.

What this week cannot tell you#

Whether the claims that survive are understood.

You have established that everything on your screen has an event behind it. You have established nothing about whether a reader takes from it what you meant. A claim can hold perfectly and still mislead: “resumes from stage 3” is impeccably sourced and means nothing to someone who has never seen your stage names.

That gap is week 11, and it is the only week of the course where the evidence comes from outside your own head.