nhouseholder/fpf-hypotheses

Propose Hypotheses Workflow

Execute the First Principles Framework (FPF) cycle: Context → Hypotheses → User Input → Verify Logic → Validate Evidence → Audit Trust → Decision → Present.

Problem Statement: $ARGUMENTS

---

Parallel Evaluation Template (used in Steps 5-7)

For each hypothesis file in [source_dir], launch a parallel fpf-agent (sonnet[1m]):

• Read the hypothesis from [source_dir]/<hypothesis-id>.md
• Evaluate against [criteria]
• Rate on 0-1 scale; pass if score >= [threshold]
• [action]: move passing files to [dest_dir], failing to .fpf/knowledge/invalid/, appending evaluation notes
• Wait for all agents, confirm files moved correctly

| Step | Source | Dest | Threshold | Criteria | Action | |------|--------|------|-----------|----------|--------| | 5 | L0 | L1 | 0.6 | Internal contradictions, logical fallacies, unsupported leaps, circular arguments, missing causal links | Move + add verification notes | | 6 | L1 | L2 | 0.5 | Empirical/theoretical support, evidence quality, counter-examples, evidence strength | Move + add validation notes | | 7 | L2 | — | — | Source reliability, confirmation/survivorship bias, anchoring, epistemic humility → compute R_eff | Write audit to .fpf/evidence/audit-{id}-{date}.md, reply with R_eff + weakest link |

---

Steps

Step 1: Initialize (Main Agent + FPF Agent)

1. Create directory scaffold:

   mkdir -p .fpf/{evidence,decisions,sessions,knowledge/{L0,L1,L2,invalid}}
   touch .fpf/{evidence,decisions,sessions,knowledge/{L0,L1,L2,invalid}}/.gitkeep
   

2. Launch fpf-agent (sonnet[1m]): Analyze problem statement, identify domain/constraints/stakeholders/success criteria/assumptions. Write context summary to .fpf/context.md.

Step 2: Generate Hypotheses (FPF Agent)

Launch fpf-agent (sonnet[1m]): Generate 3-5 competing hypotheses. Each gets:

• Unique ID (H1, H2...), title, kind, scope
• Core claim, reasoning, key assumptions, testable predictions

Write each as separate file to .fpf/knowledge/L0/. Reply with summary table (ID | Title | Kind | Scope).

Step 3: Present Summary (Main Agent)

Read all L0 files, present summary table, ask user: "Would you like to add any hypotheses of your own?"

Step 4: Add User Hypothesis (Conditional Loop)

If user says yes: launch fpf-agent (sonnet[1m]) to format user's hypothesis into standard template with ID/kind/scope/assumptions/predictions. Write to .fpf/knowledge/L0/. Loop back to Step 3. Exit when user declines.

Step 5: Verify Logic

Apply Parallel Evaluation Template with Step 5 parameters.

Step 6: Validate Evidence

Apply Parallel Evaluation Template with Step 6 parameters.

Step 7: Audit Trust

Apply Parallel Evaluation Template with Step 7 parameters.

Step 8: Make Decision (FPF Agent)

Launch fpf-agent (sonnet[1m]): Create Design Rationale Record (DRR) comparing all L2 hypotheses on R_eff, evidence strength, risk, feasibility, and constraint alignment. Select best hypothesis, justify decision, document why alternatives were rejected.

Inputs: .fpf/knowledge/L2/, .fpf/evidence/ Write DRR to .fpf/decisions/. Reply with comparison table (Hypothesis | R_eff | Weakest Link | Status) + recommended decision + rationale.

Step 9: Present Final (Main Agent)

1. Read DRR from .fpf/decisions/, present results
2. Suggest next steps: implement selected hypothesis, /fpf:status, /fpf:actualize
3. Ask user if they agree; if not, relaunch Step 8 with user's modifications