> ## Documentation Index
> Fetch the complete documentation index at: https://vaquill.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Drift Suggestions

> Where your team's real redline behavior has diverged from what the playbook says, with the language they actually accepted

A playbook records what you decided your positions should be.
Your redline history records what your team actually did.
Drift suggestions is the gap between the two, per clause, with the accepted language attached so the conversation has evidence in it.

Find it on the **Insights** tab of the playbook editor.
The panel stays hidden until there is enough signal to say something.

## What a suggestion looks like

Each row names a clause, an action, a signal strength, how many redline decisions it is based on, and the acceptance rate.

| Action                  | When it appears                                                 | What it is telling you                                                                 |
| ----------------------- | --------------------------------------------------------------- | -------------------------------------------------------------------------------------- |
| **Consider relaxing**   | Your team accepted the counterparty's language on most redlines | The playbook may be more conservative than your real practice                          |
| **Consider tightening** | Your team rejected most redlines on the clause                  | The playbook may be more permissive than your real practice                            |
| **Review alignment**    | Acceptance is mixed                                             | Worth a senior read on whether the position matches the deals you are actually closing |

Suggestions are ordered by signal strength first, then by how far the acceptance rate sits from the middle, so the strongest and most lopsided cases come first.
Use the window selector to look at 30 days, 90 days, 180 days, or a year.

## Expand a row

Opening a suggestion gives you three things.

<Steps>
  <Step title="The summary">
    A plain count, for example that your team accepted 9 of 11 redlines on this clause.
    No interpretation beyond the action label.
  </Step>

  <Step title="Your current playbook position">
    The preferred position as written today, so you can read it against the behavior without switching tabs.
  </Step>

  <Step title="What your team actually accepted">
    Up to three representative samples of the final language that came out of those negotiations.
    This is the part that makes the suggestion actionable: it is real clause text from your own deals, not a paraphrase.
  </Step>
</Steps>

## Add an accepted position as a fallback

Each sample has an **Add as fallback** action.
It appends that language to the clause's fallback ladder, leaving your preferred position untouched.

This is the safe direction for a learning loop to move.
It records "we have accepted this before and it was fine" without quietly rewriting what you ask for first.

<Note>
  Applying a suggestion writes a version snapshot before it changes anything, so it is undoable from the History tab.
</Note>

<Tip>
  Treat these as conversation starters for a senior review, not auto-update recommendations.
  A clause where you always accept the counterparty may mean your position is too strict, or it may mean your team has been conceding something it should stop conceding.
  The data cannot tell you which.
</Tip>

## Limitations

* A clause needs a handful of recorded decisions in the window before it produces a suggestion. Quiet clauses never appear.
* The action is a simple threshold on acceptance rate, not a statistical inference.
* Suggestions are derived from decisions your team recorded on contract reviews. If nobody records outcomes, the panel stays empty forever.
* Samples are representative final texts, chosen heuristically. They are not the most common position, and no numeric distribution is computed from them.
* Applying a suggestion only ever adds a fallback rung. It never edits the preferred position, the walk-away floor, or any governance field.
* Acceptance is counted per decision, so a handful of unusual deals can dominate a clause with low volume.
* Nothing is applied automatically. Every change is a click you make.

## Related

<CardGroup cols={2}>
  <Card title="Playbook Health" icon="chart-line" href="/docs/guides/playbooks/playbook-health">
    The same signal at playbook level, plus coverage gaps and staleness.
  </Card>

  <Card title="Clause Rules" icon="list-check" href="/docs/guides/playbooks/clause-rules">
    How fallback ladders behave once you add a rung.
  </Card>

  <Card title="Versions, Sharing, and Export" icon="clock-rotate-left" href="/docs/guides/playbooks/versions-and-sharing">
    Undo an applied suggestion from version history.
  </Card>

  <Card title="Contract Review" icon="file-contract" href="/docs/guides/contract-review">
    Where the accept and reject decisions are recorded.
  </Card>
</CardGroup>