D4 build week: the table turns green
420-302-VA · WEEK 14 · FALL 2026

Utility · read Wednesday even if the week feels fine

When the week fights back

Build weeks fail in patterns, and every pattern below was predicted by a page this course already wrote; the table mostly routes you back to the argument you need at the moment you stop believing it. The one new structure is the triage checkpoint: a ten-minute decision procedure the pair runs midweek on purpose, because the alternative is running it Sunday night by accident.

The midweek triage

A flowchart for the midweek checkpoint. Start: triage, Friday at the latest, ten minutes. First decision: is any old rung red right now? Yes leads to a stop box: nothing new; half-split back to green first, Week 12's rule. No continues. Second decision: will the remaining rungs honestly fit the remaining days, with estimates doubled? Yes leads to keep the rhythm, re-confirm Sunday's last join. No leads to the cut box: cut from the bottom of the ranked list, journal each cut with its reason, never cut quality from kept rungs; then re-rank what remains. A third check beside the flow: hardware or partner at risk? If yes, run the clone recipe and raise it with the teacher today. Triage · Fri at the latest · 10 min any old rungred right now? yes STOP: nothing new half-split back to green; main must walk first (W12) no do the remaining rungs fit the days left, estimates × 2? yes Keep the rhythm re-confirm Sunday as the last join + stranger test no Cut from the bottom of the list, journal each cut never quality from kept rungs · then re-rank what remains Hardware or partner at risk? clone recipe today + tell the teacher today
Ten minutes, three questions, standing decisions. The order matters: regressions outrank schedule (a list re-ranked on a broken main is fiction), and the estimates-doubled test is Hofstadter's law made procedural. The side check runs regardless of the other answers; risk raised early is the method being graded.

Symptom, diagnosis, move

SymptomWhat is actually happeningThe move
One rung is at triple its estimate and growingThe ninety-ninety rule found its meal: the invisible slice (the join, the edge cases) was never in the estimateSlice thinner: land the smallest demonstrable piece as done-done, re-rank the rest of the rung as a new, honest entry; Rank
"We're behind, let's skip the re-runs this week"Trading minutes of visible cost for an invisible break whose next full exercise is the demo itselfRefuse; script the lap under ten minutes instead. The economics are on Rhythm, and they do not change when tired
A join broke a rung from last weekA shared contract shifted, the predicted n(n−1)/2 mechanism; normal, and why station 4 existsThe triage's first rule: nothing new until green. Half-split with the W12 ladder, smallest fix, full lap
The table is filling with amber, not greenRungs are being declared before the definition of done; "basically working" has crept backConverting two ambers to green outranks starting anything new; each amber's note names its missing link, which is the to-do
Partner out sick midweekThe silent-pair risk, materialized; the plan assumed twoWIP = 1 makes the handoff clean: one rung's state to transfer. Re-run triage solo with estimates honest, cut early, and tell the teacher today, not Monday
"We found a much better architecture"The money-pit quadrant talking, mid-week, wearing insight's clothes: a proven system traded for an unproven one in proof weekJournal it as a DEBT/IDEA line with pride, build nothing of it before D5; the one exception (it blocks a high-value rung) is argued on Rank
Demo board behaving strangely, or diedThe single-board hostage scenario, on scheduleThe clone recipe is the stranger test run for real: fresh SD, config from the example, services from deploy/. Run it now, not Sunday; flag hardware with the teacher today
README drifting behind the systemReplicability, a D4 face, decaying silently while features shineDocs are rungs with value density like any other; one ranks in the top half, and Sunday's stranger test is its verification; D4 day
Third evening in a row on the same already-green rowThe depth trap: polishing R1's fifth coat while R4 stays red; effort without table movementReread the breadth figure, then pull the top red high-value row; the table, not satisfaction, is the scoreboard
"The freeze wastes two whole days, we could add one more rung"The risk curve being argued with, from its cheapest-looking pointThe curve and the tie-breaker question are on Freeze; the two "wasted" days are buying the only product left, trust in what is already green
Rehearsal keeps being postponed for "real work"Rehearsal mis-filed as optional polish instead of as the highest-density remaining rungIt is on the runway schedule with dates (Tue, Wed) precisely so it cannot be postponed twice; zero risk, direct D5 value, nothing else left competes
Everything works, but only on your home benchWorks-on-my-bench: hidden dependencies on your network, your paths, your luckThe stranger test and the demo-conditions rehearsal exist for this; hardcoded addresses found Sunday are config fixes, found Thursday they are the demo

Before you ask for help

  • The triage's three answers as of today: regressions state, fit-with-doubled-estimates verdict, risk flags.
  • For a technical wall: the half-split evidence, two probes minimum, per the Week 12 ritual.
  • For a schedule wall: the ranked list with the proposed cut line drawn, so the conversation starts at the decision.