The folder structure of the content repository, drawn as a proposal for one decision session. Decided on 2026-10-01, all four as recommended, and carried out the same day. The repository already has an index of where each thing is. What it lacks is a settled answer to where each kind of thing goes: a note, a piece of research, a proposal, canon, and the method that says how one becomes the next.
Section 01 is the chain from note to method, and where it breaks. Section 02 is where the method guides should live. Sections 03 and 04 are the two folders whose names say more than their contents support. The decisions are yours, and they are listed at the end.
Material moves through five stages. Only one step between them is designed: the research decision of 2026-08-04 (D-121), under which generated research is read by you and only then ratified into canon. The step into research, the stage between research and canon, and the step from practice to method have no rule. The rules that do exist are written in five different places.
Proposed: one short document at the repository root owns the chain. It names the stages and the folder that holds each, says who moves material from one stage to the next, and defines the status line every file carries. The writing_notes README, the index warning and the master plan point to it instead of each restating part of it. The three decisions stay in the decision log as the record of why.
Methods for the whole project go at the root as project_*.md. They are
the project's layer of the shared writing methodology, which looks for them there, as do the review
skills and the Uniflection project. The prose-quality review looks for the prose guidelines
in the root and nowhere else. A method for one subsystem stays with that subsystem,
because you reach it from that subsystem's own entry point.
Three guides are out of place: two sit among the scene plans, and one sits inside the findings it produced. One method has no guide at all.
The index heads the research folder with a warning that everything in it is machine-generated and unratified. That is true of most of it, but not all. The folder also holds work you or an agent authored, work checked against its sources, and evidence from your own photographs. The index works around this by tagging each entry's status in italics.
Proposed: keep the folders by topic, and put a status line at the top of every file. Splitting the folders by kind is the alternative. But a status written in the file stays with it wherever it is cited. That guards against the hazard the index already warns about, where each citation makes a claim read more settled than the last.
| Kind | Example in research/ today | Status line |
|---|---|---|
| LLM output | the sixteen deep-research conversations, the chat extracts, the 2026-08-03 resource wave | generated |
| Authored analysis | the gathering model; the arable and wild-food compilations; the extraction syntheses and their reconciliation | authored proposal |
| Checked against sources | the verification of canon contradictions C-1 and C-2, with sources opened | checked |
| Primary evidence | what the labels in your museum photos say, with photo paths and corrections | evidence |
| Authored story material | the historical families written as story strands | authored proposal |
The research decision routes ratified material into canon/. The master plan and the
project instructions say a file's folder alone confers no canon status. The two fit together only if
canon/ holds ratified documents, one version of each. Today it does not.
| In canon/ | What it is |
|---|---|
discussion_over_anthracite_and_ilmenite.md | The metallurgy production plan again, without its table of contents, beside metallurgy_production_plan_anthracite_ilmenite.md. |
grand_strategy_framework_established.mdgrand_strategy_framework_[established].md | Two diverging versions of one framework, 867 and 850 lines. |
knowledge_diffusion_…_complete.md | Beside …_complete_refactored.md, which is nearly twice as long. |
spain_sovereign_…_complete.md | Beside …_canon_refactored.md, six lines apart. |
sa1628_system_prompt_web_based_2026-01-12.md | A chat assistant's system prompt that summarises the project. A briefing, not canon. |
None of these needs a ruling of its own. They are listed so they can be approved as a batch, or one at a time.
| Kind | What | Fix |
|---|---|---|
| broken link | The writing pipeline links three files that don't exist: the old user-level writing pipeline and scene-plan guidelines, and a hyphenated name for the website build guide. | Point at the methodology's fiction guides and the underscored build guide. |
| broken link | The writing pipeline and the project instructions spell the scene-plan guide with hyphens. The real file uses underscores. | Corrected as part of the move. |
| stale route | The index sends active work to NEXT_SESSION.md and to a BACKLOG.md that doesn't exist. Session state moved to Workstate on 2026-07-18. | Point the index at Workstate. |
| misplaced | narrative/planning/ also holds the substrate-viz build, a subgroup-matrix generator, and two specs: the substrate visualization and the Day-0 timeline rebuild. | Specs to specs/; the viz build beside the substrate. |
| misplaced | At the root: five worker briefs, at least two still in flight, and spec_brief.md. | Finished ones to brief_archive/. |
| misplaced | At the root: a session-9 handoff manifest. | _audit/, with the other close manifests. |
| misplaced | At the root: the research prompts for open design questions. | Into the research folder. |
| clutter | At the root: 10 screenshots; two old checklists, one with three open items; and --help.bands.md, a population-build output saved under a mistyped filename. | Screenshots and checklists out of the root; the mistyped output deleted or moved to its run. |
Decided 2026-10-01: all four as recommended. The rules for commissioning research became a section of the chain document, project_canon_chain.md. The subgroup-matrix generator stays in narrative/planning/, because the substrate engine loads it by that path.
Recommended: the root, as project_*.md.
The alternative is a methods/ folder. That means editing the review skills for both Claude Code and Codex, and the methodology's routing. Either way, the project instructions' pointer to the scene-plan guide changes, and this decision approves that edit.
Recommended: move it to the root as project_historical_people_research.md.
Last session proposed narrative/planning/, which would put a method among the scene plans. Its findings stay where they are.
Recommended: one root document owns it, and every file in research/ carries a status line.
Yes, no, or change it. If yes: whether the research-commissioning rules go in the same document or get their own guide.
Sections 04 and 05.
Approve them as a batch, or one at a time. The doubled canon documents need your call on which version stands.
Drawn on 2026-10-01 from the repository as it stood before the decision, and kept as the record of
it (D-135). index.md is the map of where things are now. How the substrate is operated is on the system page.