idea:forge

Forge Idea

Read everything accumulated about an idea — docs, evaluations, pushback sessions, predictions — and synthesize it into a single consolidated summary. This is the compounding step: it makes all the accumulated knowledge legible.

Workflow

Step 1: Load Everything

Read <argument> to identify the idea folder. Search for it under ideas/.

ideas/*{argument}*/

Read ALL files in the folder:

• Idea docs: 00-overview through 05-experiments (including their changelog entries)
• Evaluation files: all evaluation-*.md files (sorted by date for trajectory)
• Pushback sessions: all pushback-session-*.md files
• Prediction docs: all pushback-predictions-*.md files

Step 2: Build the Score Trajectory

Parse YAML frontmatter from all evaluation files. Build a timeline:

## Score Trajectory

| Date | Combined | VC | Market | Weakest Dimension | Deal-breakers |
|------|----------|-----|--------|-------------------|---------------|
| 2026-03-19 | 42/100 | 34 | 50 | Team (1/5) | Team |
| 2026-03-20 | 58/100 | 52 | 64 | Traction (2/5) | None |
| 2026-03-22 | 74/100 | 70 | 78 | GTM (3/5) | None |

**Trend:** Improving (+32 points across 3 evaluations)
**Biggest improvement:** Team (1→4, after adding CTO background and advisor plan)
**Persistent weakness:** GTM — still no proven cold acquisition channel

If only one evaluation exists, show it as the baseline with no trajectory.

Step 3: Synthesize Findings

Read all pushback sessions and compile:

What's been validated:

• Claims that received "Verified" verdicts with High confidence
• Assumptions marked as validated in 03-assumptions.md

What's been refuted or remains risky:

• Claims that received "Refuted" or "Unresolved" verdicts
• Assumptions still untested or invalidated
• Deal-breakers from evaluations (even if resolved — note they were once flagged)

Key pivots and evolution:

• Compile from changelog entries across all docs
• Track how the thesis evolved from the original pitch

Step 4: Write the Forge Summary

Create forge-YYYYMMDD.md in the idea folder:

# Forge: <idea name>
Date: <date>
Evaluations analyzed: <count>
Pushback sessions analyzed: <count>

## Idea in One Paragraph
<Synthesized from 00-overview, incorporating all evolution>

## Score Trajectory
<table from Step 2>

## What's Validated
- <validated claim/assumption with source>

## What's Still Risky
- <unvalidated claim with what would test it>

## Key Pivots
1. <date>: <what changed and why>

## Open Questions
<Compiled from all sessions and docs>

## Pitch-Ready Summary
<3-5 bullet points a founder could use to pitch this idea right now, grounded in validated evidence>

## Verdict
**Idea strength:** [Strong / Promising / Needs work / Reconsider]
**Confidence level:** [High / Medium / Low] — based on ratio of validated vs assumed claims
**Biggest remaining risk:** <one sentence>
**Recommended next action:** <one specific thing to do>

Step 5: Present and Suggest Next Steps

Output the forge summary, then:

Forge complete! Here's where this idea stands.

→ /idea:pushback {idea-name}   — if there are still untested claims to challenge
→ /idea:evaluate {idea-name}   — if you've made changes since the last score
→ /idea:new {new-idea-name}    — if the forge reveals a pivot worth exploring as a new idea

Principles

• Synthesize, don't summarize. A summary just shortens. A synthesis connects dots across sessions and finds the pattern.
• Show the trajectory. The score timeline tells the story of how the idea evolved — it's the most valuable output.
• Ground in evidence. Every "validated" claim must cite where it was validated (evaluation score, pushback verdict, experiment result). No unsourced confidence.
• Be honest about what's still assumed. The ratio of validated to assumed claims is the real confidence metric.
• Make it pitch-ready. The forge output should be something a founder can hand to an advisor or investor as a concise status update.