# Dry-run — Idiopathic Pulmonary Fibrosis (IPF)

Validation that the pipeline is genuinely **disease- and modality-agnostic**: the planning logic
was walked on IPF (a fibrotic disease, deliberately unlike EoE) without running compute, to
confirm no EoE-specific assumptions leak into the generalized flow.

## What was checked
1. **All composed skills exist** — 27/27 referenced catalog skills resolve; 0 missing.
2. **All composed connectors exist** — `mcp-clinical-genomics`, `mcp-drug-regulatory`,
   `mcp-omics-archives`, `mcp-chembl`, `mcp-clinical-trials` all present.
3. **Stage map resolves** for IPF (stages 1–5 walked; inputs/skill/gate per stage below).
4. **Modality router diverges from EoE** — the pivotal agnosticism test.

## Modality routing (Stage 5) — IPF target shapes
The EoE program locked a **vaccine** (antigen-specific tolerance). For IPF, the same router
produces different modalities purely from target biology — confirming no EoE lock-in:

| IPF-plausible target | Router output | Design skills |
|---|---|---|
| Secreted pro-fibrotic ligand (IL-13 / TGFβ-type) | `ligand_trap_binder` | proteinmpnn/solublempnn/ligandmpnn → boltz/chai1/esmfold2/alphafold2/openfold3 |
| Cell-surface receptor (integrin αvβ6-type) | `antibody` | proteinmpnn/solublempnn/ligandmpnn → boltz/chai1/esmfold2/alphafold2/openfold3 |
| Intracellular fibrotic driver w/ chemistry | `small_molecule` | diffdock → boltz/chai1 |

## Stage map + gate schedule (dry-run, stages 1–5)
| # | Stage | Composes | Gate |
|--:|-------|----------|------|
| 1 | Indication & unmet need | indication-dossier + clinical-genomics/drug-regulatory | SCOPE LOCK (hard) |
| 2 | Patient priorities / PFDD | patient-centered-market-and-survey | PRIORITY RANK (soft) |
| 3 | Literature + pipeline map | systematic-review-orchestration + clinical-trials/clinical-genomics/drug-regulatory | WHITESPACE CONFIRM (hard) |
| 4 | Data mining → targets | omics-target-mining + omics-archives/clinical-genomics | TARGET EVIDENCE (soft) |
| 5 | Target & modality selection | rubric + clinical-genomics/chembl | TARGET/MODALITY LOCK (hard) |

## Result
- Stage map resolves end-to-end with correct skill/connector composition and the right gate at
  each stage. Gate schedule matches `checkpoint_map.csv` (incl. Stage 4 TARGET EVIDENCE soft gate).
- Modality router is data-driven, not EoE-anchored: three IPF target shapes yield three distinct
  modalities, none of them the EoE default.
- Dossier state threads correctly (init → stage updates → `next_stage` advances to 6 after 1–5).

**Conclusion:** the pipeline generalizes. No EoE-specific content is hard-coded into the flow;
disease, target, and modality are all parameters resolved from data + the human gates.