Write Down the Dead End

The Attempt That Passed Every Check

Before the layout mode from the last lesson existed, somebody tried to fix the packing problem within the existing mechanism. The idea: instead of padding the host — which lengthens the document — constrain the host shell to the height left over, give it its own scroll, and drop the padding entirely.

Check every geometric property and it holds. The shell ends exactly where the panel begins. The shell owns its scrolling rather than the page. The document stops lengthening. Nothing is under the panel. By every measurement anybody had defined, the problem was solved.

It was shown to the person who had raised the original complaint, who looked at it and said it still plainly reads as overlap.

Why the Measurements Were the Wrong Instrument

The verdict is not aesthetic fussiness — it identifies a real difference the checks were blind to. Clipping content into a shorter box and moving content out of the way produce the same geometry and different experiences. In the second, the cards visibly relocate; you watch the layout respond. In the first, everything stays exactly where it was and a window closes over the bottom of it.

The measurements had been derived from the mechanism rather than from the goal. Each one was a true statement about boxes, and the goal was about what the page appears to do when the panel opens — which no box measurement captures.

When something satisfies every check and still fails, the checks are measuring an adjacent question — and that is the finding. The reflex is to argue the result, or to add one more check. The useful move is to notice that the check set was derived from the mechanism you happened to be using, and to go find the property that actually distinguishes the outcomes. Here it was 'does the layout visibly respond', which no measurement of final geometry can express.

The Part That Matters Most: It Was Written Down

The rejected approach is recorded, in the architecture document, with its full mechanism, the verdict, and an explicit instruction not to re-attempt it.

That is unusual and it is deliberate. The natural fate of a failed attempt is deletion — it did not work, so it leaves no trace, and the record shows only what shipped. The problem is that the approach is reachable: it is the first thing a competent person tries when they encounter this problem, precisely because it satisfies every obvious check. Deleting it guarantees somebody rebuilds it, gets the same clean measurements, is equally satisfied, and is equally wrong.

A dead end that is easy to reach and expensive to disprove deserves a paragraph, and the paragraph has to say plainly that it was tried and rejected. Otherwise the next session does not read it as a warning — they read it as an idea nobody has had yet.

Code

The rejected approach, recorded as rejected·text

REJECTED - do not re-attempt. Written down BECAUSE it is the first
thing a competent person tries, and because it passes every check.

  THE IDEA: stop padding the host (which lengthens the document);
  instead constrain the shell to the height that is left and give
  it its own scroll.

    body:has(.panel.mode-dock) .app-shell {
      min-height: 0;
      height: calc(100vh - var(--panel-offset-bottom, 0px));
      padding-bottom: 0;
      overflow-y: auto;
    }

  EVERY GEOMETRIC CHECK PASSES:
    [x] the shell ends exactly where the panel begins
    [x] the shell owns its own scroll, not the page
    [x] the document stops lengthening
    [x] no content is underneath the panel

  AND IT STILL READS AS OVERLAP IN USE.

  WHY: clipping content into a shorter box and moving content out
  of the way produce THE SAME GEOMETRY and DIFFERENT EXPERIENCES.
  With flow layout the cards visibly relocate - you watch the page
  respond. Here nothing moves; a window closes over the bottom of
  it. The check set was derived from the MECHANISM, not from the
  goal, so every check was a true statement about boxes and the
  goal was about what the page appears to DO.

  THE ANSWER IS the flow layout mode (previous lesson). This is not.


Why this block exists at all: the natural fate of a failed attempt
is deletion, and deleting this one guarantees that the next person
rebuilds it, takes the same clean measurements, is equally
satisfied, and is equally wrong. Recorded so it reads as a warning
rather than as an idea nobody has had yet.

Exercise

Recall an approach your team tried and abandoned in the last year that a newcomer would plausibly propose again. Write the record: what it was, what checks it passed, why it was rejected anyway, and one line saying it was attempted rather than merely considered. Put it where the design lives, not in a closed ticket.

Hint

The best test for whether it is worth recording: would you be annoyed to see it proposed again next quarter? If yes, that annoyance is a prediction that it will be — and the paragraph is how you answer it once instead of every time.

The Attempt That Passed Every Check

Why the Measurements Were the Wrong Instrument

The Part That Matters Most: It Was Written Down

Code

External links

Exercise

Progress

Comments 0