Shadow redline (internal)
What this page is — the capability that privately reviews the counterparty's latest round for the liabilities they slipped in, and tells you the counter-move for the next one.
What it is for — because the dangerous changes in an incoming round are never the obvious ones, and finding them by reading is exactly the task attention fails at.
The problem it solves — the damaging change in an incoming round hides among harmless edits and gets accepted with them.
Route: /org/papers/documents/<document>/ai → Negotiate → Shadow redline (internal) ·
Permission: Use AI drafting/extraction/Q&A for documents. ·
Utility: papers_shadow_redline · Cost: 450 credits estimated, 1,800 maximum.
1. What it is
Shadow redline does asymmetric review. It compares their proposed round against your current text and surfaces every hidden liability it introduces or preserves against you.
The things it is built to catch are the ones that survive a careful read:
| The quiet move | What it looks like |
|---|---|
| The widened indemnity | Three deleted words — "to the extent caused by" |
| The cap that lost its teeth | Still there, now excluding the claims that matter |
| The shifted burden of proof | "shall demonstrate" moved from one party to the other |
| The deadline trap | A notice window shortened just enough to be missable |
| The ambiguity | Wording that will be read their way when it matters |
Note "or preserves": a liability they left untouched from a previous round, which you conceded and forgot, is reported alongside the new ones.
This analysis is written to be candid about your own exposure and your negotiating moves. It is your side's privileged working product.
It must never be shared with the counterparty, pasted into a document they can see, or attached to
a message that leaves your organisation. Act on it through your normal redlines — not by forwarding
the findings. The (internal) in its name is doing real work.
| Reads | Candour | |
|---|---|---|
| Shadow redline | Their round against your current text | Blunt, internal-only |
| Risk analysis | The whole document | Shareable |
| Version diff | Two versions, neutrally | Shareable |
| Redline grade | One edit against your playbook | Shareable |
Version diff tells you what changed. Shadow redline tells you what it costs you — the same change, read adversarially.
2. Why you would use it
The failure mode this addresses is specific: a round comes back, it looks broadly acceptable, three clauses have moved, and the one that matters is the one whose change was smallest.
- Small edits carry the largest liabilities. Deleting four words from an indemnity is a bigger commercial event than rewriting a whole schedule, and it is the one a reader's eye passes over.
- It reviews with unlimited patience. Round six gets the same scrutiny as round one, at 2am, under deadline. That consistency is the point.
suggested_responsemakes it actionable in one line. Each finding arrives with the counter-move for the next round, so review output converts straight into your redline.- It surfaces what you already conceded. Liabilities preserved from earlier rounds are the ones institutional memory loses first.
- It is the cheapest insurance in the negotiation. 450 credits against an uncapped indemnity is not a close comparison.
3. What you provide
| Input | Required | Values | Notes |
|---|---|---|---|
| Perspective | Optional — defaults to internal / our side | our_side, counterparty, neutral | Whose interests the review defends |
You rarely change this. The entire purpose is a candid one-sided read for your own team, so the default is correct almost always.
Set it only when you are deliberately reviewing on behalf of a specific entity — for example reviewing a licence round from your subsidiary's exposure as licensee rather than from the parent's.
Results are cached per perspective, as with Risk analysis and Fairness / balance.
This compares their proposed text against yours. Without a counterparty round loaded there is nothing to shadow-review — run it on a document mid-negotiation, not on a draft you wrote.
4. What it reads automatically
| Read from the document | Used for |
|---|---|
| Our current text — your side's current version | The baseline being protected |
| Counterparty's round — their proposed text, from the prior or incoming version | The thing being reviewed |
| Title & family — e.g. "Licence Agreement" | What liabilities are expected in this kind of document |
| Parties | Telling our side from theirs |
The first two rows are the capability. Everything it reports is a difference between them, or a liability standing in both.
5. What you get back
hidden_liabilities[], ordered highest severity first. An empty array is a real result: their
round is genuinely clean.
| Field | Type | Values | Means |
|---|---|---|---|
title | string | — | Short name of the liability |
severity | enum | high, medium, low | How badly it can hurt you |
clause_ref | string | empty if unclear | The section it hides in |
rationale | string, 1–3 sentences | — | What they changed or kept, and how it can hurt you in practice |
suggested_response | string, one sentence | — | The counter-move to make in the next round |
rationale is where the value concentrates. It does not say "the indemnity is broad"; it says which
words moved and what that conversion means.
6. Worked example
A licensor receives round three back from the counterparty. Nothing looks alarming — the changes are small and mostly tidying.
Input: perspective left at the default, internal / our side.
Top entry of hidden_liabilities[]:
Indemnity quietly widened severity:
highclause_ref: "9. Indemnification" rationale: They deleted "to the extent caused by", converting our proportionate indemnity into a full one — we'd now cover losses even where they are partly at fault. suggested_response: Reinstate the causation qualifier and add a comparative-fault carve-out in the next round.
What the licensor does with it. Five words were deleted from a clause nobody flagged. The consequence is that a loss the counterparty partly caused becomes wholly the licensor's — which, on a licence with no cap on indemnity claims, is unbounded.
The response goes into the next redline as written: reinstate the qualifier, add the carve-out. The findings themselves stay internal — the counterparty sees a redline, not this analysis.
7. Running it
- Open the document in Orbit Papers with the counterparty's latest round loaded.
- Open the AI Assistant drawer, or go to the document's AI workspace.
- Choose Shadow redline (internal) under Negotiate.
- Confirm the Perspective — the default is right almost always.
- The internal-only findings appear in a result tab. Keep them on your side.
Run it on every incoming round before you respond. That is the whole discipline: the value comes from it being routine, not from remembering to use it on the rounds that feel risky.
Then act through your normal redlines. Do not forward the findings.
Available for auto-run from the type's AI Config tab, where on_round is the correct and only
sensible event — it fires exactly when a round arrives. See
AI configuration.
8. The admin contract
| Must be true | Where | What happens if it is not |
|---|---|---|
| Your role holds Use AI drafting/extraction/Q&A for documents. | Role editor | The AI Assistant button does not appear |
The papers_shadow_redline utility is active | Orbit AI Flow → utilities | "This utility is currently disabled" |
| The utility is enabled for your organisation | /org/ai-utilities | Absent from Negotiate, with no error |
papers.ai_monthly_credit_cap not yet reached | System Config | "monthly AI credit cap reached: n of n credits used this month" |
| A counterparty round exists on the document | Redlines and rounds | There is no incoming text to compare against, and it has nothing to review |
| Result visibility is restricted to internal roles | Role editor | The candour that makes this useful becomes a disclosure risk. Do not grant Use AI drafting/extraction/Q&A for documents. to counterparty-facing roles |
That last row is the one to get right before enabling this capability. Everything else here is a configuration question; that one is a confidentiality question.
For a confidential document, auto-run is default-denied — a platform administrator must set
papers.ai_confidential_cloud to the literal allow, and the block appears only in the server log.
A manual press still works, and where a local model key is configured the run is forced on-prem and
audited on the timeline as ai_local_routed.
9. Don't confuse this with…
| Version diff | Neutral account of what changed. This is an adversarial account of what it costs you |
| Risk analysis | The whole document, shareable. This is their round, internal-only |
| Redline grade | Judges one edit against your playbook. This sweeps the round for what you missed |
| Compromise draft | Produces wording to share. This produces analysis never to be shared |
10. Troubleshooting
| Symptom | Cause |
|---|---|
| It found nothing | Either their round is genuinely clean, or no incoming round is loaded (§3) |
| It has nothing to compare against | No counterparty round exists on the document. Run it mid-negotiation, not on your own draft |
| A finding refers to something they did not change this round | Deliberate — it reports liabilities preserved as well as introduced (§1) |
A liability has no clause_ref | It arises from the interaction of several clauses rather than one |
| The findings read harshly | They are written to be candid for internal use. That is the design — do not share them (§1) |
| Switching perspective returned instantly | Results are cached per perspective; that is a previously saved run |
| The capability is missing from Negotiate | Role lacks Use AI drafting/extraction/Q&A for documents., or the utility is off for the organisation (§8). The document type does not gate it — AI Config controls auto-run only |