NN — ⟨experiment title⟩

Project: ⟨Pnn⟩ · Milestone: ⟨n⟩ · Date started: ⟨YYYY-MM-DD⟩

Sections 1–8 are written and committed BEFORE the experiment runs. Field-by-field guidance: notebook-template.md. A completed example: example-filled.md.

1. Problem

⟨What breaks in the world if this does not exist? State it with a quantity. Do not name a technology.⟩

2. Constraints

Hard (physics, hardware, the data I have):

Chosen (design decisions I could revisit):

Assumed (believed but not verified — the interesting ones):

3. Existing Approach

⟨How is this normally solved? State exactly how much you have read.⟩

I stopped reading at: ⟨specific point, deliberately, so section 4 is mine⟩

4. My Naive Design

⟨Components, data structures, algorithm. Written before reading further.⟩

My reasoning: ⟨why each choice — this is what section 11 will diagnose⟩

5. Predictions

⟨Each must be specific enough to be marked confirmed or falsified without argument.⟩

6. Hypothesis

H:

Falsifier:

7. Experimental Setup

• Hardware: ⟨model, cores, RAM, load average at run time⟩
• Software: ⟨language and version, key library versions⟩
• Data: ⟨source or generator, size, distribution — and how you verified it⟩
• Parameters: ⟨swept, and held fixed⟩
• Seed:
• Command:

Setup validation: ⟨how you confirmed the independent variable actually varies⟩

8. Baseline

⟨What am I comparing against, and why is it the fair comparison? Prefer a degenerate configuration of my own system.⟩

9. Results

⟨Tables and plots. Units, n, uncertainty. No interpretation.⟩

Verdict on each prediction

10. Surprises

⟨What contradicted section 5, and by how much. Quantify each gap.⟩

11. Failure Analysis

⟨Which specific design decision caused it, and the causal chain. Not a category — a mechanism, specific enough to be wrong.⟩

12. Next Experiment

E1 — ⟨the smallest experiment that resolves the most uncertainty⟩ Cost: ⟨hours⟩ · Prediction: ⟨...⟩ If confirmed: ⟨...⟩ · If not: ⟨...⟩

Then, in order:

• E2 —
• E3 —

13. Generalization

Should hold when:

Should not hold when:

Predicts about systems I have not built: ⟨a falsifiable claim⟩

14. Reproducibility

commit  :
command :
seed    :
runtime :
output  :
expect  :  ⟨headline number ± tolerance⟩