The LIA project: one system, five milestones
420-302-VA · LIA PROJECT · FALL 2026

Stage 3 of 5 · Delivering · the five gates

The five milestones

Each milestone is a gate with three faces: what you demonstrate in class, what the repo holds as evidence by end of day, and what evaluation focuses on. The faces repeat from D1 to D5 with rising stakes, so learn the pattern once. Exact required contents, submission format and grading come from the brief on Omnivox, which governs; this page is the logic those requirements follow.

How to read this page

Every milestone's evaluation asks one underlying question, and knowing it is how you prepare without guessing: D1, is this idea a project; D2, does the architecture hold; D3, is the method surviving contact with a hard week; D4, is the system real; D5, can you stand behind it in public. Weights follow the questions: the three 5 % gates buy early correction while change is cheap, and the 25 % at the end pays for what the corrections made possible.

D1 · The proposal · W11, Nov 16 · 5 %

FaceD1
DemonstratesThe idea, as a system: the six slots filled in one line each, the schematic of the interconnections, and the pair's plan, rung list ranked, Week 13 pre-planned small
Repo evidenceThe proposal document per the brief in docs/; draft requirements passing the five qualities, each with a verification-table row; pair agreements in the README
Evaluation focusCompleteness of the six slots, requirement quality (testable, unambiguous, atomic, necessary, feasible), and scope realism against the four remaining weeks

D1's classic failure is an idea pitched as a feeling ("something with plants") instead of six filled slots; its classic quiet strength is a sharp slot 5, since a pair that already knows what it will fight has already designed its demo. Launch context: Week 11's project page.

D2 · The walking skeleton · W12, Nov 23 · 5 %

FaceD2
DemonstratesOne real value crossing every interface, live, in the three-step shape: cause something physical, show it travel, name the contracts that carried it
Repo evidenceThe skeleton tagged (git tag d2-skeleton, pushed); docs/topics.md current with every topic, payload and route; the D2 analysis per the brief
Evaluation focusEnd-to-end truth over polish: did the value genuinely cross all contracts, and does the written analysis match the running system; a diagnosed partial, honestly half-split, beats an undemonstrated whole

Thin is the assignment: hard-coded thresholds and unlabelled numbers are fine today and scheduled rungs tomorrow. The full argument and the demo shape: Week 12's skeleton page, taught the day D2 lands.

D3 · One rung, half-width week · W13, Nov 30 · 5 %

FaceD3
DemonstratesThe skeleton plus exactly one thickness, chosen tiny but real in advance, demonstrated after the final examination that shares the day
Repo evidenceThe rung's commit history and tag; old rungs re-verified (the regression check); the verification table gaining its first green rows; the journal's stand-up lines across the squeezed week
Evaluation focusMethod under pressure: a small rung fully landed, nothing broken behind it, and a plan that visibly anticipated the half-width week rather than colliding with it

The one scheduling trap of the whole project

The final examination (15 % of the course) and D3 share Monday, November 30. This has been announced since the project opened; the rung for this week is chosen in the Week 12 studio, not during exam study. A pair arriving at D3 with an over-sized rung chose to, a week earlier.

The day itself, review hub included, both proofs and the mode switch between them: Week 13’s hub, with its D3 page carrying the demonstration’s craft.

D4 · The heavy milestone · W14, Dec 7 · 15 %

FaceD4
DemonstratesThe system substantially complete: the rung list's high-value features landed, the full sensing-deciding-acting-showing loop running against its disturbances, services keeping it alive without a terminal
Repo evidenceThe verification table largely green, each row demonstrable on request; deploy/ holding the unit files; documentation current enough that the system is replicable from the repo, the Week 1 promise kept
Evaluation focusThe largest single project grade, weighted toward working breadth and verified requirements: what runs, what is proven, and what the repo would let a stranger rebuild

D4 is where the early method pays or its absence bills: a pair that climbed rungs arrives assembling evidence, not features. The remaining gap to D5 is deliberately small, because the week after D4 belongs to rehearsal.

The build week itself, ranking, rhythm, scoreboard, demonstration and freeze: Week 14’s hub, with its D4 page carrying the demonstration’s craft.

D5 · The public demonstration · W15, Thu Dec 10 · 10 %

FaceD5
DemonstratesThe complete system, publicly: the live demo in its rehearsed shape, cause the disturbance, show the system answer it, walk the dashboard, with both partners presenting and answering questions
Repo evidenceThe frozen, tagged final state the demo runs from; the completed verification table; the project documented to its end, journal included
Evaluation focusThe demonstration itself and the understanding behind it: does the system do what the requirements claim, live, and can each partner explain any layer when asked

Thursday, December 10 follows a Monday schedule; the session is demonstrations, in public. The freeze rule and the rehearsal protocol that make this day calm are on the evaluation page, and the brief governs format, timing and grading.

Check yourself

Why are the three checkpoints worth 5 % each when they plainly take serious work, and what are they buying you?
They are priced as forcing functions, not as products. Each checkpoint forces a decision while changing it is cheap: D1 forces the idea into testable shape before code exists, D2 forces every interface risk into the open before features stack on them, D3 forces the method to survive a hard week before the heavy milestones. What they buy is the 25 % at the end: D4 and D5 grade the system those early corrections made possible, so the checkpoint work is paid, mostly, two gates later. A pair optimizing each gate locally misses that the weights describe one connected investment.
Your D3 demo is the skeleton plus one rung, and another pair shows five new features, three of which crash. Reason about the evaluation focus.
D3's question is whether the method survives pressure, and its evidence row asks for old rungs re-verified and nothing broken behind the new work. One rung fully landed, regressions checked, journal showing the week was planned around the exam: that answers the question yes. Five features with three crashing answers it no twice over, the rungs were cliffs (the system is not demonstrable end to end) and the half-width plan was ignored. Breadth is D4's focus, a week later, with triple the weight waiting for exactly that work, done on rungs that hold.
What single repo artifact appears in every milestone's evidence row, and why is it the one markers can trust most?
The verification table (with its companion, the tagged history). It appears as draft rows at D1, first checks at D2, first greens at D3, largely green at D4, complete at D5, so it records the project's whole truth over time. It is trustworthy because it is checkable on demand: any green row is an invitation to say "show me", and a pair keeping it honest has been rehearsing the D5 demonstration since November, one row at a time.