ADR-D22: Thesis as a concise research diary

Status:

Accepted

Date:

2026-05-05

Phase:

F-THESIS

Context

Two failure modes for a software-platform thesis: a chronological log of every iteration (archaeology, no narrative), or a clean post-rationalised systematisation (loses the reasoning that justifies the canonical pipeline). The journey itself (KNN baseline -> 22 features -> 52 features -> trazabilidad crisis -> leakage discovery -> selective-rerank canonical) is part of the contribution.

Decision

Concise research diary. Each pivot that taught something earns 1-3 pages. Distilled history, not raw archive. Chapter 6 (evaluation) is the journey, written in prose. The insights appendix (D30) carries brief notes on ideas that were probed and discarded without reaching the chapter.

Consequences

• Volume target 60-75k words is reachable without padding.

• Every pivot link to its operational record (Job findings, see D11).

Resolution

Closed.