docs: add proposal 129 — prompt dentist (pre-pipeline prompt enrichment)

[HejEgonBot]

Proposal 129: The Prompt Dentist — Pre-Pipeline Prompt Enrichment

Empirical finding from the SpicedSnackCo_v1 pipeline run (2026-03-29) that challenges Proposal 128's conclusion that the Execute Plan section is always template-driven autopilot.

Key finding

An operationally dense prompt (3 SKUs, packaging specs, CT/RI market, DTC + farmers markets channel) produced 274 WBS tasks with domain-specific descriptions — not the ~32K generic block seen in P128's comparison runs. The quality ceiling is primarily a prompt limitation, not a pipeline limitation.

The gap

Most user prompts are missing critical dimensions that the pipeline needs to produce specific plans:

• Location/jurisdiction — without it, the pipeline guesses. US vs UK vs Canada = different regulations, suppliers, and market dynamics
• Budget/scale — "$500 farmers market table" vs "$5M manufacturing line" — the pipeline can't distinguish without a signal
• Product specifics — generic "snack business" gives no grounding for downstream tasks
• Target market & channel — who buys this and how?
• Timeline — 3 months vs 18 months changes every dependency chain

A prompt missing location should never score above "fair" regardless of other detail — location is load-bearing.

Proposal

A pre-pipeline "prompt dentist" that:

1. Scores the prompt across key operational dimensions
2. Asks 5-8 targeted questions to fill gaps
3. Enriches the prompt before the expensive pipeline runs
4. Provides a quality forecast

Cheapest intervention with the highest quality multiplier.

Relationship to other proposals

• Complements 128a (quality score) — dentist forecasts pre-pipeline, 128a measures post-pipeline
• May reduce need for 128c (two-tier Execute Plan) if enriched prompts naturally produce the context layer
• Recommended priority: before 128c, alongside 128a

Full analysis and evidence in the proposal doc.

[HejEgonBot]

PAT scope test