---
name: markup-favorability-classify
description: "Use when a marked-up/redlined contract comes back and each tracked edit must be classified favorable, unfavorable, or neutral to ONE named party and ranked by impact — the base model reasons edit-by-edit but flips whose seat it argues from, so the same change gets called \"favorable\" in one line and \"unfavorable\" in the next. Pins the perspective test (label relative to the named party only: reduce their obligation/liability or add their rights/recovery = favorable; the reverse = unfavorable; no allocation shift = neutral), the payor-vs-claimant flip on caps and thresholds, the mutual→one-sided rule, and a fixed impact ladder (I1 economic/risk-allocation, I2 exit/control, I3 operational, I4 administrative) to order the edits. Do NOT use for rating a clause's raw risk/escalation tier, extracting fields, plain-language summary, checking deviation from your own playbook, or redrafting a clause."
metadata:
  decimalai-skill-id: "9eeedbe7-dd4f-458f-8d07-756df62c482e"
  decimalai-version-id: "775ffd60-54c9-4075-930e-bc9677e877ac"
  decimalai-version: "1"
  decimalai-body-sha256: "2d1d8b82a81e881ef95ea0bfdb79b347b7dac73a61a70fc4c633df32423aaca7"
  decimalai-name: "markup-favorability-classify"
  decimalai-format: "1"
  decimalai-invocation: "model"
  decimalai-category: "legal"
  decimalai-source: "https://app.decimal.ai/s/markup-favorability-classify@1/SKILL.md"
---
# Classify each redline edit by favorability to one named party, then rank by impact

Given a marked-up contract and a **named party** (the one whose seat you sit in), process every tracked
edit through the same four steps. The base model has all the pieces; what it does inconsistently is hold
the perspective fixed and rank consistently. Do both.

For each edit:

1. **What lever moved?** Name the thing the edit changes — an obligation, a liability, a right, a
   protection, a threshold, an amount, a term.
2. **Which direction, for the named party?** Apply the perspective test below — always from the named
   party's seat, never the "obvious" one.
3. **Label** favorable / unfavorable / neutral.
4. **Assign an impact tier** (I1–I4) so the edits can be ordered.

## The perspective test — hold the seat fixed

Everything is judged relative to the **named party P**, not to whoever the clause seems to be "about."

- **Favorable to P** = the edit *reduces* P's obligation, liability, risk, or cost, **or** *adds* P's
  rights, protection, optionality, or recovery.
- **Unfavorable to P** = the edit *increases* P's obligation, liability, risk, or cost, **or** *removes*
  P's rights, protection, optionality, or recovery.
- **Neutral** = the edit shifts no allocation between the parties: a defined-term cleanup, a
  cross-reference or renumbering fix, formatting, a symmetric change that hits both parties' deadlines
  the same way, or an update to a party-agnostic mechanic.

### The flips the base model gets wrong

- **A cap or damages ceiling that is *lowered* is favorable to the party that would PAY and unfavorable
  to the party that would RECOVER.** Fix which seat P is in before you label. Same edit, opposite label
  depending on the named party — this is the single most common mislabel.
- **Deleting a duty P owes = favorable to P; deleting a protection P holds = unfavorable to P.** A
  deletion is not favorable by default.
- **Moving a threshold, cure period, or notice window** — direction depends on who must clear it. A
  *longer* cure period is favorable to the party at risk of breaching and unfavorable to the party
  waiting to terminate.
- **Mutual → one-sided:** converting a reciprocal right or protection into a one-way one is unfavorable
  to the side that LOSES the reciprocal half, even if the surviving words look unchanged.
- **Genuinely bilateral edits are neutral** — a change that cuts both ways symmetrically allocates
  nothing.

## Impact ladder — rank the edits after labeling

Assign each edit one tier, then order the review by tier first (I1 before I4), and by the size of the
shift within a tier. Neutral edits carry no impact and sort to the bottom.

| Tier | What's at stake | Typical edits |
|------|-----------------|---------------|
| **I1** | Core economics & risk allocation | liability cap, indemnity scope, cap carve-outs, payment amount, IP ownership |
| **I2** | Exit & control | termination rights, renewal / auto-renew, assignment & change-of-control, exclusivity |
| **I3** | Operational obligations & protections | SLA, warranty, cure period, notice window, audit rights, confidentiality scope |
| **I4** | Administrative & enforceability | governing law / venue, definitions, boilerplate, severability, notice mechanics |

An administrative edit can still be directional (a venue moved to the counterparty's home state is
unfavorable to the away party) — it just stays at I4. Direction and impact are independent axes: set the
label with the perspective test, set the rank with the ladder.

## Output shape

For each edit: the label, the tier, and one clause naming the lever and its direction for the named
party. Then a single ordered list, highest impact first. When asked which of two edits matters more,
the higher tier wins; ties break on the magnitude of the shift.