R&D — Projects | The Undercut Wiki

The Projects tab is the control room for every running R&D project — this season’s and next season’s, in one table. It does not start anything; it lets you re-allocate, watch and cancel. It is also where the R&D pool’s real state is visible: capacity, how much is locked, and how much is free.

How to open it

Car icon → Projects (tab 4).
Both Design tabs carry a Go to Projects → shortcut at the bottom.
Locked during pre-season (current_phase < 6).

The R&D pool — this tab’s currency

Projects run on the R&D department pool (the flask), not the Engineering pool (the hammer):

capacity = 20 + rnd_level × 8         (lv1 28/GP · lv5 60/GP · lv10 100/GP)
allocated = Σ wu_per_turn over in-season projects + next-year projects
free      = capacity − allocated

In-season and N+1 projects share the same capacity. The Work Order panel on this tab shows allocated / capacity WU/GP — free X — N project(s) running, plus a one-line summary of each in-season project: Component (invested/total WU @ n/GP — g GP). The bar recolours amber past 60% and red past 85%.

An allocation is locked from the moment a project launches until it completes or is cancelled. That is the whole scarcity model: capacity is per-GP throughput, and a project holds its slice of it.

The Active R&D Projects table

Column	Meaning
Part	Icon + name + a year badge — blue `{year}` for this season’s car, purple `{year+1}` plus an `N+1` chip for next season’s
Project	`New part (Low/Medium/Big)` or `Fix trait: {name}`
Progress	`invested / total WU (P%)` with a bar — blue for in-season, purple for N+1
Allocation	`[−] N WU/GP [+]` and `~G GP` remaining
Action	`Cancel`

gps_remaining = ceil( (total_wu − wu_invested) / max(wu_per_turn, 1) )

A completed N+1 project is rendered differently: a green full bar reading ✓ Complete, a dash where the allocation controls were, and Ready next season instead of Cancel. Its wu_per_turn was already zeroed and returned to the pool — a finished design must not keep reserving throughput. This also self-heals older saves whose completed projects were still holding capacity.

Re-allocating

new_alloc = max( wu_per_turn + min(delta, free), 1 )

+ adds throughput, clamped by the free capacity you actually have.
− always works, down to a floor of 1 WU/GP — a project can be slowed to a crawl but never fully parked.
The ~G GP estimate updates immediately.

The two allocation controls are separate (rnd_alloc_± for in-season, nyalloc_± for N+1) but they draw from and return to the same pool.

More WU/GP = fewer GPs, linearly. There is no efficiency penalty for concentrating or spreading — only the opportunity cost of what else could hold that capacity.

Cancelling

Cancel removes the project and frees its allocation immediately. The invested WU are lost. There is no partial refund and no way to resume from where you were.

For a mandatory next-year regulation part, cancelling late in the season is close to fatal: with no completed design at the rollover, the component is force-homologated as a rushed Poor part with zero head-room. See R&D — Design.

The accrual rule

Every GP — and every pre-season turn — advance_rnd_projects runs:

wu_invested = min( wu_invested + wu_per_turn, project.wu_cost() )

wu_cost() is the project’s own persisted total_wu (the continuous slider size), falling back to the legacy tier bucket only when total_wu == 0 on an old save.

The accrual-bucket bug, and why the rule is written this way. Accrual used to cap at the tier bucket (kind.wu_cost() — Low 80, Medium 150, Big 240) rather than the project’s real size. A slider-sized 121-WU “Low” project therefore froze at 80/121 and could never reach completion. Carried-over N+1 regulation redesigns hit this every single time and rolled straight into rushed Poor parts at the next CarReview. If you are reading an old guide that says “size your project to the tier boundary”, ignore it — the cap is the project’s own cost now.

Pre-season turns counting as GPs is what saves most carried-over redesigns: a project that was 60% done at the rollover routinely finishes before round 1.

What happens on completion

New part → version bump, quality tier rolled, quality budget applied with a random perf/head-room split, a fresh trait set rolled, the result registered as a KnownDesign blueprint, and an email with the spec card. No physical part is created; go to R&D — Production.

N+1 new part → the allocation is freed and an email says the design is ready, but the blueprint itself only lands at the season rollover.

Trait fix → see below.

Fix a Problem — trait remediation projects

Trait fixes are launched from the Design {year} tab but live here like any other project. They are the cheapest thing R&D sells.

cost = 40 WU        (Yellow trait)
cost = 120 WU       (Red trait — 3×)

What can be fixed

A row appears only when all of these hold:

The trait is fully Discovered (a Suspected ”?” is not enough).
The component is not the engine — it is supplier-rolled each season with its own vintage traits, and is never player-designable.
The discovery’s version matches the component’s current default design version. A fresh N+1 roll has its own undiscovered traits; last year’s v1 knowledge must not resurrect a fix offer on a design you no longer produce, nor leak that the new roll carries the same flaw.
The default design still carries that trait at Yellow or Red level.
No other project is running on that component, and free capacity > 0.

Rows are deduplicated per (component, kind) — after a fix, the same problem used to appear twice (once per version).

What a fix actually does

The result badge is always a green ✓ Solved: a fix resolves the trait outright. Yellow and Red both go away — a Red project simply costs 3× at launch. (The older flow downgraded Red → Yellow and demanded a second project.)

The mechanics on completion:

The fix is sourced from the real fitted part carrying the trait — not from the blueprint. The blueprint keeps the full base head-room, so sourcing from it handed you a fresh ceiling: you would re-discover head-room you had already spent improving, and pocket a bonus.
The trait is removed from that stats copy.
Race wear is cleared (current_perf = initial_perf, health = 100) so produced copies start clean at the developed level — not worn, not re-rolled.
A new design version is created, guaranteed to outrank every existing version for that component (max(existing_max, version_counter) + 1), so it becomes the default and is what Production builds.
Trait-discovery knowledge is carried forward. Every surviving trait’s km and state are copied onto the new version. Without this, fixing one trait forced a full re-discovery of every other known trait on the part.
An email arrives: “{Component} redesign ready”, with a before/after trait chip.

The part on the car does not change

A fix produces a blueprint. The fitted parts keep the trait — and keep suffering its effect — until you produce a copy of the new version and fit it. That is deliberate.

What the AI does

AI teams run the same project system with tighter rules:

Maximum 3 concurrent in-season projects.
Design tier is capped by budget tier: Large → Medium, Medium and Small → Low. No Big designs, ever.
Allocation is auto-sized to finish in roughly the tier’s classic delivery delay (total_wu ÷ delivery_delay), locked against capacity like yours.
Candidates are picked in a fixed order: reliability redesign (a fitted part below the finance model’s reliability_floor — bypasses the championship-hope gate, because a fragile part costs points and rush-replacement cash), then a trait remediation (only when the finance plan allows discretionary work — it buys no raw pace), then the weakest fitted part on pace (gated by the model’s hope_floor, so a prudent manager pivots to next year’s car early and an aggressive one keeps developing a lost season for constructors’ prize money).
Mandatory N+1 designs are commissioned before the improve pass, most important part first (by new-part cost: body > floor > gearbox > … > fuel lines), with adaptive pacing ceil(pending / cycles_left) so a Full reset fits the remaining GPs. An underfunded team secures the impactful parts and eats the Poor penalty on the cheap ones.
The AI cannot do reliability improvement at all — that lever is player-only.
A bankrupt team develops and delivers nothing.

Advanced

Capacity, not cash, is the bottleneck. Design projects cost no money — only R&D throughput. If the tab says no WU, the fix is to cancel something or raise the R&D department, never to earn more.
Front-load, then taper. Push a project to full allocation to get the blueprint into your hands early, then drop it to 1 WU/GP on anything that no longer matters. The floor of 1 lets you keep a slow project alive at almost no cost.
A trait fix is the best WU-per-point in the game — when the trait is bad. 40 WU removes a Yellow Overheating or Twitchy outright. Compare: 40 WU of Improve on a level-5 department buys under three steps.
But a fix is not free of side effects. It replaces your default design with a version sourced from the current fitted part’s consumed state. If a much better spare of the old version exists, check the Production preview before switching over.
Watch the year badge. The table mixes both seasons. Cancelling what you think is a spare in-season project may be the mandatory N+1 redesign for a regulated part.
A completed N+1 project is free capacity you may not have noticed. It shows ✓ Complete and has already released its allocation — check the free number before assuming you are full.
The estimate is honest only if nothing else changes. ~G GP uses the current allocation; adding a new project steals free capacity, not existing allocations, so running estimates stay valid until you re-allocate.

Common mistakes

Cancelling to “recover” WU. Invested work units are gone. Only the per-GP allocation returns.
Launching a fix on a Suspected trait. Only fully discovered traits are offered. Run a focused Free Practice or Private Testing stint first — see Free Practice.
Expecting the trait to disappear from the car. It disappears from the design. Produce and fit a copy.
Trying to fix an engine trait. The engine is supplier-rolled and never designable; those rows are filtered out on purpose.
Spreading 28 WU/GP across four projects. At 7 WU/GP each, a 160-WU design takes 23 GPs — longer than the season. Fewer, faster projects beat many slow ones.
Leaving a mandatory N+1 project at 1 WU/GP. It will not finish by the rollover, and the rushed Poor part has zero development head-room for the whole of next season.