# Selected research-route decisions

Purpose: show different recorded outcomes for selected research branches. These are bounded transcriptions and summaries of retained project records, not a new assessment of the underlying research or a census of every attempted idea. The original working records remain private.

## Embodied Energy

The literature-position finding recommends: “do NOT mint this as a standalone research paper.” Its reason is that the proposed contribution largely restates existing literature. It recommends retaining the boundary-normalisation work as a method example. This is the recorded recommendation; the map has not repeated that literature search.

## Critical Minerals

The project record labels the standalone publication “FOLDED” and gives “novelty gate failed” as its reason. The same record retains it as a method demonstration. This is evidence of the route decision, not a fresh check of its market figures or source claims.

## DRACO-lite

The README labels the project “STATUS: PARKED” and describes a “working skeleton, not an active project”. It does not record a failed novelty check. The map therefore shows a parked branch without inventing that reason.

## Dolma and FineWeb-Edu

The FineWeb-Edu handover states: “This work began as a Dolma 3 auditability arm of *The Model Is a Dependency*. It is not that.” It records the earlier design premise as void because the proposed corpus/model pair no longer corresponded. The map records that supersession; it has not rechecked the external dataset’s present state. A superseded arm is distinct from the later published FineWeb-Edu study.

## Verification and limits

The extracts were checked against the retained project records for the stated decisions. No fresh literature search, dataset audit or evaluation was run. The selected cases do not estimate how often ideas are stopped, and a parked project may be resumed only through a separate author decision.

## AI assistance

NM AI Research selected the permitted examples and makes substantive and publication decisions. OpenAI GPT-6 prepared these summaries from the records; Claude Opus proposed wording and reviewed an earlier candidate. Both providers are subjects of the wider portfolio.