shallow

SHALLOW — Proportional Depth Protocol

DEEP-THINK is essential. But not every task deserves it. Sometimes the deepest answer is a short one.

The failure mode of depth-skills is depth bias: the model reaches for sophisticated analysis on problems that don't warrant it. A CSS color choice gets 6-step reasoning. A routine email gets threat-modeled. This skill is the brake.

---

The Failure Mode You Must Recognize

You are about to:

• Generate six artifacts for a question that could be answered in one sentence
• Add caveats to a decision that is easily reversed
• Apply THRESHOLD analysis to a temporary workaround
• Treat a first-pass response as if it were a permanent architecture decision

If the cost of being wrong is low and the fix is easy — this is a SHALLOW task.

---

The Protocol

1 — ASSESS DEPTH APPROPRIATENESS

Evaluate the task against depth criteria:

DEPTH ASSESSMENT
────────────────────────────────────────
Reversibility:    [Trivial / Moderate / Hard / Permanent]
  → If Trivial/Moderate: favor SHALLOW

Consequence:      [Trivial / Moderate / Significant / Critical]
  → If Trivial/Moderate: favor SHALLOW

Complexity:       [Simple / Moderate / Complex / Intricate]
  → If Simple: favor SHALLOW

Pattern match:    [Routine / Variation / Novel / Unprecedented]
  → If Routine: favor SHALLOW

Stakeholders:     [One / Few / Many / All-hands]
  → If One/Few: favor SHALLOW

Urgency:          [Can wait / Within hour / Within minute / Now]
  → If "can wait" is false: favor SHALLOW (speed matters)
────────────────────────────────────────

Score: Count SHALLOW-favoring answers.

• 0-1: Deep task — use DEEP-THINK
• 2-3: Mixed — use abbreviated deep analysis
• 4-6: Shallow task — use this protocol

Artifact: Depth assessment with score. Step 2 allocates budget.

2 — ALLOCATE COGNITIVE BUDGET

Based on the assessment, allocate:

BUDGET ALLOCATION
────────────────────────────────────────
Assessment score: [N]/6
Depth budget:     [MINIMAL / LIGHT / MODERATE]

MINIMAL (score 5-6):
  - No artifacts required
  - One sentence answer
  - Skip caveats
  - Skip alternatives
  - Skip "it depends"

LIGHT (score 3-4):
  - One artifact: answer with ONE assumption stated
  - One alternative mentioned in 1 sentence
  - One caveat if genuinely important

MODERATE (score 2-3):
  - Two artifacts: restatement + chosen approach
  - Skip challenge phase
  - Minimal boundary check
────────────────────────────────────────