# Tailoring an Existing Setup

> Start from your own Le Mans Ultimate setup, answer four questions, and let the Setup Interview derive a new version using bounded relative changes.

The **Setup Interview** tailors a Le Mans Ultimate setup that is already in your
private vault. It never creates a setup from Braking Lab's reference files. Your
own `.svm` stays the base, and a successful interview appends a new version to
the same setup.

> Setups is in **Public Beta** and requires a personal **Plus or Ultra** plan. A Team seat alone does not grant access.

## Before You Start

Upload or select an LMU setup that you own. The engineer also needs the exact
context of that version: track or track type, dry or wet, qualifying or race,
and its risk posture. The source context is explicit; the interview does not
silently guess a neighboring condition or session.

The selected version must be a readable `.svm` whose car identity matches the
setup. If ownership or identity cannot be verified, no version is written.

## The Four Target Questions

Your engineer asks all four in one message:

| Question | Answers | What it decides |
|----------|---------|-----------------|
| **Session** | Qualifying · Race | What the new version is for |
| **Condition** | Dry · Wet | Which exact target evidence cell applies |
| **Risk** | Safe · Balanced · Aggressive | The requested stability margin |
| **Balance** | Stability · Neutral · Rotation | Whether the car should plant or rotate |

"I don't know" maps to the conservative **Safe** risk and **Neutral** balance.
For dry evidence with Safe and Performance anchors, choose **Balanced**
explicitly when you want a computed middle. That midpoint is used only when
every option stop it introduces was observed and certified for the exact car
variant; otherwise your engineer asks you to choose Safe or Aggressive and
writes nothing. Wet evidence may have one unspecified risk cell instead of Safe
and Performance siblings. In that case the risk axis is unavailable and no
Balanced middle is invented.

## References Teach Moves; They Never Ship

Braking Lab's protected reference library is used only to learn bounded,
relative movements between a matching source cell and target cell. Those
movements are applied to the adjustable values in **your** setup.

The reference file itself is never copied into your vault. Its identity, path,
hash, text and absolute values are never returned. This also means two drivers
with different owned bases do not converge onto a hidden shared file: the same
relative movement starts from each driver's own values.

After the relative movement, reviewed tailoring rules may refine the candidate
for the requested risk and balance. The complete candidate is validated once
against the LMU setup rules before anything is saved.

## What Stays Yours

The derived version keeps the selected setup's lineage and remains under the
same setup record. Its parent is the exact version you selected. The driver's
LMU header, vehicle identity, Notes and line-ending style are preserved; only
declared adjustable parameters may change.

If every proposed move is inapplicable, outside the reviewed bounds or cancels
out, the interview reports that no safe delta landed and writes nothing.
For a dry Balanced build, this includes a midpoint that would require an
unobserved option stop. Braking Lab never interpolates across that gap.

## What Your Engineer Reports

The result is deliberately relative. Expect:

- the categorical source and target cells;
- which adjustable parameters moved and by how many option steps;
- which proposed moves were dropped, with stable reasons;
- whether reviewed risk, balance or session transforms were applied or skipped;
- the new child version and its download, when URL signing succeeds.

The report never exposes a protected reference or its absolute setup values. A
missing download URL after a successful build only means temporary URL signing
failed; the child version already exists and should not be created again.

In the app, the new version carries an **Interview build** badge, and the setup
shows your answers as its **Interview profile**. Later iterations read that
profile back instead of asking the four questions again.

## Exact Coverage, No Silent Fallback

Both source and target cells must exist for the same verified LMU car variant.
There is no nearest-match search and no silent dry/wet or qualifying/race
fallback.

Coverage is limited to Braking Lab's protected reference library. Not every LMU
car, track type, condition and session has references. Your engineer maps the
track you name to a track type through a reviewed list. If it does not know the
track, it asks you which type fits instead of guessing.

When your exact ask is not covered:

- your engineer tells you plainly and leaves the vault unchanged;
- it logs the missing combination, so Braking Lab knows what to curate next;
- it never uses another car's references;
- it may offer a variant for the same car, such as the dry cell when you asked
  for wet, but only builds it if you explicitly agree.

The regular remix route remains available whenever you want to make a focused
hypothesis directly from your own setup and telemetry.

## iRacing

The Setup Interview does not apply to iRacing. Its native `.sto` is encrypted,
so iRacing changes are delivered as garage recipes derived from your own garage
export. See [iRacing setups](/en/docs/ai-setups/iracing-setups).

## Related

- [Remixing setups](/en/docs/ai-setups/remix) — make a focused change to an owned version
- [Le Mans Ultimate setups](/en/docs/ai-setups/lmu-setups) — upload the `.svm` that becomes your base
- [Setups and your sessions](/en/docs/ai-setups/sessions) — turn subsequent runs into evidence