research/¶
The audit trail behind daydreamd. Nothing here is edited after the fact; corrections go in new files.
| File | What it is | Date |
|---|---|---|
00-reference-dojo-md.md |
The publishing pattern we modelled on (Eduard Cristea's dojo.md), what it lacks, what we added | 2026-09-13 |
01-publishing-practice.md |
Evidence-backed checklist for research-grade open-source repos (MUST / SHOULD / NICE), exemplar repos, gotchas | 2026-09-13 |
02-sanity-review-2026-09-13.md |
Adversarial review of the approach: ten objections, internal contradictions in the brief, the minimum viable experiment, the honest claim scope | 2026-09-13 |
03-prior-art-refresh-2026-09-13.md |
Literature and repo sweep for mid-July to mid-September 2026; what is still unclaimed; naming availability | 2026-09-13 |
BRIEF-v3-2026-07-25.md |
The founding brief, unchanged. Start at "SETTLED POSITION". Contains claims that were made and then killed; kept so nobody re-litigates them | 2026-07-25 |
DREAMING-lineage-2026-07-24.md |
The first research pass: academic names for the idea, prior systems, white space, build plan | 2026-07-24 |
Reading order for a newcomer: BRIEF (SETTLED POSITION only), then 02, then 03, then ../PREREGISTRATION.md, then ../paper/paper.md.
The three 2026-09-13 reports were produced by parallel agent runs on Claude Fable 5.1 and are reproduced verbatim, including any typographic habits of the model. Documents authored for the repo itself follow the style rules in ../CONTRIBUTING.md.
- 05-prior-art-execution-verified-code-2026-09-13.md: sweep on execution-verified daydreaming over a codebase (the v0.2 direction); component-by-component claimed/unclaimed verdict and a v0.2 eval design.