quantitative-decision-gate

Quantitative Decision Gate

Wrong skill? Design evaluation -> debate. Creative options -> brainstorming. Stuck -> think-twice.

Announce at start: "I am using the quantitative-decision-gate skill to evaluate options quantitatively."

When to Use

• Before asking the user "should I...?" or "do you want me to...?"
• When choosing between 2+ implementation approaches
• When deciding whether to continue, pivot, or stop a task
• When evaluating trade-offs (speed vs quality, scope vs time)

Example

# Example: Deciding between extract-method vs inline
echo "Decision: refactor approach"
echo "| Dimension    | Wt   | Extract | Inline |"
echo "| Impact       | 0.35 | 8       | 4      |"
echo "| Effort (inv) | 0.25 | 6       | 9      |"
echo "| Risk (inv)   | 0.25 | 7       | 8      |"
echo "| Reversible   | 0.15 | 9       | 9      |"
echo "| Weighted     |      | 7.35    | 6.85   |"
echo "Margin: 7% -> AUTO-SELECT extract-method"

The Decision Matrix Protocol

Step 1: Frame

State: "I need to decide: [decision]". List options (min 2, max 5).

Step 2: Score (4 mandatory dimensions)

Dimension	Weight	Option A	Option B
Impact	0.35	8	6
Effort (inv)	0.25	9 (low)	4 (high)
Risk (inv)	0.25	7 (low)	8 (low)
Reversibility	0.15	9	5
Weighted		8.15	5.85

Score 1-10 (10=best). Effort is INVERSE (high score = low effort).

Step 3: Decision Rule

• Margin > 10% -> AUTO-SELECT. Proceed without asking.
• Margin < 10% + reversible -> AUTO-SELECT with note.
• Margin < 10% + irreversible -> Present matrix to user.

Step 4: Log

Record: decision, scores, 1-sentence rationale.

---

Anti-Patterns

Anti-Pattern	Detection	Correction
Asking user a quantifiable decision	"should I" + options exist	Build matrix, auto-select
Victory-claiming without data	"better" without numbers	Produce the numbers
False ceiling diagnosis	"impossible because X"	Try 2 more approaches first
Single-option framing	"Should I do X?" only	Generate >=1 alternative
Gut-feel decision	"feels right"	Score it. Numbers confirm or deny

Failure Modes

Failure	Detection	Recovery
Matrix not followed	Decision != highest score	Re-examine scoring
All options tie	Within 5%	Add context-specific 5th dimension
Scoring bias	One option all 9s/10s	Red-team: argue FOR the loser
Decision not acted on	TODO but no action	Execute or defer with reason

Companion Skills

• think-twice: When matrix reveals a trap
• debate: Multi-option design evaluation
• brainstorming: Generating options for the matrix
• plan-and-execute: Executing the chosen option
• failure-autopsy: Post-mortem if chosen option fails