Coverage roadmap

This page tracks known gaps in the book — topics that belong here and are not yet written. It exists so that contributors can see where the holes are without having to audit 80-odd pages, and so that the reasoning behind each gap is recorded rather than rediscovered.

How the gaps were identified¶

The book was compared against a full research-lifecycle model — a nine-stage pipeline (propose, initialize, curate, analyze, checkpoint, QC, export, publish, disseminate) running inside a continuous administrative track covering funding, ethics, deadlines and credit.

The comparison produced one clear finding: the book covers the middle of the lifecycle well and both ends barely at all.

Lifecycle:  Govern → Initialize → Curate → Analyze → Checkpoint → QC → Export → Publish → Disseminate
Coverage:   ░░░░░░   ████████     ██████   ███████   ░░░░░░░░░░   ███  ░░░░░░   ░░░░░░░   ░░░░░░░░░░
            └─ thin ─┘                                            └────────── thin ──────────┘
            └──────────── administrative track: thin throughout ────────────┘

The Start Here pathway reflects the same shape: it begins at “find data” and ends at “share your work”, which skips what must be settled before data access and stops short of what turns an analysis into a citable research product.

Documentation of tools is genuinely strong. Documentation of process is where the gaps are.

A second axis: analysis and craft¶

The audit above asked whether every stage of the lifecycle was covered. A later pass asked a different question — whether the analytical and craft topics a scholar needs are present regardless of stage — and found a second set of holes, recorded as priorities 8–12 below.

Two findings stand out. The Research Methods section is entirely machine learning: there was no classical statistical inference anywhere in a book about analysing clinical neuroimaging cohorts. And cross-cohort pooling is a stated premise of several CNS projects while site harmonization, the thing that makes pooling valid, had no coverage at all.

Note that research_topics/ and modalities/ pages are deliberately short — they are link-dense hubs, not content pages. Their length is not a gap.

Priority 1 — Provenance and data versioning¶

The largest single hole, now partly filled.

Before this, DataLad appeared in four pages purely as a name in a list, despite being the mechanism most of the reproducibility advice elsewhere in the book depends on.

Priority 2 — Before you touch data¶

Nothing in the book covered this stage. A new scholar’s first real blocker is usually data access, and the book documented which datasets exist without ever describing how to get into one.

Priority 3 — Turning an analysis into a research product¶

The book stops at “share your work”. The steps that make work citable and reusable are absent.

Priority 4 — Curation depth¶

BIDS is explained as a standard, but not as a task you have to perform.

Cross-cohort pooling is a premise of several CNS projects, and variable harmonization is the unglamorous step it depends on.

Priority 5 — Administration and credit¶

Priority 6 — AI in the research workflow¶

The existing AI models section covers how large language models work. It does not cover working with them, which is now part of most people’s daily practice and has unresolved implications for provenance.

Priority 7 — A quality framework¶

The book already argues for most of STAMPED without naming it — containers for portability, provenance for tracking, BIDS for modularity. Naming the framework would let those arguments reinforce each other instead of appearing as unrelated advice.

Priority 8 — Statistics and inference¶

The largest hole on the second axis. Research Methods covers deep learning and five ML method pages; nothing covered inference. A new Statistics and Inference section was added to hold it.

Priority 9 — Ways to fool yourself¶

Pairs with imbalanced data, which already covers resampling inside the fold.

Priority 10 — Tools used everywhere and documented nowhere¶

Each of these appeared across the book only as a name in passing.

Priority 11 — Privacy and review¶

Priority 12 — Program and career¶

A different genre from the rest of the book and lower priority, but a fellowship is fixed-term and the transition out of it is a real part of the experience here.

Also outstanding¶

• Testimonial sections are empty. Every scholar page has a Tools & Resources I Used section awaiting its author. See the scholar page template. This is the highest-value outstanding item in the whole book and cannot be written by anyone other than the scholar concerned.

• Gourab K Sar’s page is not linked from any topic hub — his project needs a few more sentences before it can be categorized.

• Johanna Bayer’s page has no Project section; hers is currently embedded in the Background paragraph.

• Extending the pathway. Start Here has six steps. Once the governance and publishing pages exist, it should gain a step at each end.

• Reference management. Zotero, BibTeX and the citation workflow have no coverage. Deliberately not stubbed, because it overlaps Zettelkasten — the better move is probably a section there than a page of its own.

• COMPASS-ND and LORIS are named in Arman Hassanpour’s project but neither has a page, unlike the other portals in Databases.

Contributing¶

Every stub above states what it needs. Pick one, write it, open a PR — and if you have actually done the thing, write from that experience rather than from the documentation. The first-hand version is worth considerably more.