# Claim-to-defense evidence compiler

Compile a dissertation as a network of bounded claims, reproducible artifacts, adversarial tests, and honest residual limitations.

## Instructor edition

## 1. Compile the central claim graph

**Task type:** derivation

Express a proposed dissertation contribution as a graph linking premises, methods, artifacts, intermediate results, and bounded conclusions. Reconstruct the strongest central result from the graph and identify single points of failure.

### Deliverables

- Claim-dependency graph
- Reconstruction of the central result
- Single-point-of-failure and limitation register

### Scoring criteria

- Claims trace to evidence: 8 points
- Reconstruction is technically complete: 7 points
- Failure points and scope are honest: 5 points

### Solution outline

- Give every empirical and inferential edge an artifact or explicit assumption.
- Recompute the central result from archived inputs.
- Narrow the headline claim if it depends on an untested or nonportable edge.

## 2. Run the robustness matrix

**Task type:** analysis

Evaluate the central conclusion across data exclusions, calibration choices, model families, priors or thresholds, and plausible alternative explanations. Distinguish specification sensitivity from genuine contradiction.

### Deliverables

- Prespecified multiverse or robustness matrix
- Effect and uncertainty summary across variants
- Decision table for stable, conditional, or unsupported claims

### Scoring criteria

- Variants cover material choices: 7 points
- Uncertainty and multiplicity addressed: 7 points
- Claim status follows the matrix: 6 points

### Solution outline

- Separate defensible alternatives from arbitrary perturbations.
- Show the full distribution of results rather than selecting favorable specifications.
- Carry conditionality into the abstract and public summary when the result is not stable.

## 3. Design the hostile defense and release

**Task type:** design

Create an adversarial defense protocol in which reviewers challenge provenance, calibration, inference, alternative models, and reproducibility. Pair every challenge with an artifact, bounded response, or planned concession.

### Deliverables

- Challenge-response matrix
- Reproducibility and publication package manifest
- Correction, embargo, and post-publication monitoring rules

### Scoring criteria

- Strongest objections are represented fairly: 7 points
- Responses point to inspectable evidence: 7 points
- Release and correction rules are operational: 6 points

### Solution outline

- Assign reviewers to attack different dependency classes, not just presentation.
- Permit revise, narrow, or withdraw outcomes rather than forcing defense of every claim.
- Release data, code, environment, negative results, and limitations to the maximum ethically and legally allowed.

---

Evidence rule: distinguish calculation, model-dependent inference, experimental observation, and unresolved claim in every response.