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

Stage 5 of 6 · The milestone · 15 %, demonstrated

D4 day

Monday is not a build day; that is the first and most violated rule of heavy milestones. If the week ran its loop, Monday opens with a system that has been whole since Sunday's last join, a table green with honest edges, and a pair whose job today is presentation of true things. The class runs studio-style with demonstration slots per the announcement; this page is what to walk in with and the craft of showing breadth in minutes.

The weekend run-up

Two rituals close the build days, both scheduled on the week's strip for Sunday. First, the last join and the full regression lap: every re-run command in docs/verification, executed against current main, with the table's dates updated where the laps are the evidence. Second, the stranger test below, because replicability is a face of this milestone, not a garnish. What Sunday is not for: new rungs. A rung started Sunday is a ninety-ninety trap with one day of runway, the Friday-feature row wearing weekend clothes; the list's remaining rows become Monday's named cuts, and named cuts score better than broken attempts.

The demonstration: anatomy of breadth

D2 demonstrated one thing walking; D3, one rung under pressure. D4's axis is breadth, many requirements proven, which changes the craft: you cannot give each of eight rows the three-minute cause-travel-name treatment, so the table itself becomes the spine and depth happens only where the marker points. Whatever slot length the announcement sets, the proportions hold:

A horizontal bar divided into four segments as fractions of any slot. First, about fifteen percent: frame with the table, the system in one sentence, the table on screen, greens, ambers and cuts visible. Second, about half the slot: rows live on request, the marker picks rows, you run each chain, cause the condition, show the proof. Third, about a quarter: the evidence walk, tags d2 to d4, journal, deploy unit files, the stranger-test result. Fourth, about ten percent: state the freeze, what is now locked and what the runway holds. Under the bar: the table is the script; you rehearsed it every day at station five. Frame Rows live, on request Evidence walk Freeze system in onesentence · table up the marker picks rows; each runs its chain: cause the condition, show the proof, name the date tags · journal · deploy/stranger-test result what lockstonight ~15 %~50 %~25 %~10 % The table is the script, and you rehearsed it daily: every station-5 update this week was one row of this demo, practised once. Ambers and cuts go in the frame, stated by you before they are found by anyone; that order is the difference between method and excuse.
Spine, then depth on demand. The green segment is deliberately the widest: breadth is proven by letting the marker choose where to zoom, which only a true table survives. "Demonstrable on request" was the brief's phrase all along; today it is simply read aloud.

If the week went sideways

The honest-demo rule has not expired and will not: demonstrate what walks, state what does not with its diagnosis and its half-split evidence, show the triage decision in the journal. A smaller true table outranks a larger painted one on every face D4 marks, and telling the teacher early remains part of the method being graded. The failure mode with no defence is silence plus surprise.

The stranger test

D4's evidence face includes the repo "replicable from the repo, the Week 1 promise kept", and there is exactly one honest way to test a claim about strangers: simulate one. On Sunday, one partner plays the stranger: clone the repo fresh into a new directory, and follow the README with your hands and nobody's memory, config_example.py to real config, services installed from deploy/, the system brought up from nothing. Every place the stranger stalls is a documentation bug, and documentation bugs found Sunday are rungs; found by a marker, they are marks. The test doubles as insurance you already want: it is a dress rehearsal of the single-board-hostage recovery, proving the pair can rebuild on fresh hardware, three days before a public demo in which hardware is allowed to die.

The evidence, by end of day

  1. The tag: git tag d4, pushed, on the main that was demonstrated, joining d2-skeleton and d3-rung in the paper trail.
  2. The table, final for D4: dated greens, noted ambers, named cuts, exactly as shown, matching the format.
  3. The D4 document per the brief, in docs/: what was demonstrated, the cuts and their reasons, the stranger-test result.
  4. The journal's week: the daily entries and DEBT lines that make the week auditable, the method visible in writing.

Exit: calling the freeze

The last act of D4 day is one sentence said out loud in the pair, with a timestamp: "Frozen, as of now." From that sentence until Thursday's demo, the change rules of the next page apply. Calling it explicitly matters because an uncalled freeze is a negotiation that reopens every evening; a called one is a fact the tired version of you on Wednesday can lean on. The runway it protects is only three days long, which is exactly why it needs protecting.

Exit checklist