A 'Final Draft' That Wasn't Final

Both papers took big steps forward this week, and in hindsight I declared victory too early.

Lost in the Middle

I assembled a full draft of the reproduction paper and called it the final draft. The project kept going for another four months, so that label didn't age well. What the draft did do well was put the whole argument in one place, which made its weak points obvious: every result still came from the heuristic model.

RAG saturation

I expanded the manuscript with a related-work section, formal definitions for the three diagnostics (score margin, score entropy and rank overlap), a table of toy-corpus results, and a bibliography. Formalizing the score margin as an equation later exposed a mismatch with the code. That's exactly why I wanted the definitions written down.

Reflection

'Final draft' is a status I now reserve for when the evidence is final, not when the document is complete.