Illustrated book — state of work

Branch: claude/brain-connects-narrative-5qo317 · Last updated: 2026-08-21

The ask this branch answers: “BRAIN and CONNECTS struggles to explain strategy and impact for the work and investment to the public. Can we put together a compelling book — illustrated cartoon? With clear evidence and accomplishments so far.”

What exists and works

/book/ — “The Last Great Map” (book/index.html, one self-styled page on the default layout, linked from the About dropdown and the homepage impact section).

Eleven chapters, each with a hand-drawn inline SVG cartoon panel, public-friendly narrative, and an evidence box whose every claim resolves to a numbered source:

#	Chapter	Carries
1	The uncharted country	86B neurons; the connectome as the missing map
2	Why bother	GBD 2021: 3.4B people, #1 cause of ill health; repair-without-schematic framing
3	The strategy	2013 tools-first bet; >$3B, 1,300+ projects
4	Proof it works	Worm (15 yr by hand) → fly (139,255 neurons, 287 co-author proofreaders); ~1,000× per-neuron speedup (labeled as our arithmetic)
5	A grain of sand	MICrONS 1 mm³ (2 PB, function+wiring); H01 human fragment
6	Dividends	BICCN atlas (32M cells, 5,300+ types); DBS, dopamine sensor; AI both directions
7	From wiring to health	Retina motion rule (2011); like-to-like cortex rule (2025); H01 axon whorls; adaptive DBS FDA approval (Feb 2025); NIH speech BCI (Mar 2025); Alzheimer’s/epilepsy baselines as prospects
8	Why is this so hard	The four chores: slice / photograph / store / untangle; proofreaders as the scarce resource
9	BRAIN CONNECTS	11 awards, ~$150M, 40+ institutions; ELI5 cards for the three routes; HI-MC flagship
10	Limits	Bargmann & Marder; map of roads not traffic; genome-project analogy (provenance flagged)
11	Room for your name	EyeWire 200k players; FlyWire citizen co-authors; NeuroTrailblazers as workforce arm

Plus: eight-number spread, 1986→2028 timeline, “A note on the receipts” (release-specific counts caveat), 27 numbered sources, print CSS, CTA. Shared cartoon characters (Maya, Pip) live in one SVG <defs> block and are <use>-referenced by panels.

Validation: validate_frontmatter.rb, check_site_links.rb, check_anchor_links.rb all pass on the built site. Build recipe on this container: export PATH=/opt/rbenv/versions/3.1.6/bin:$PATH (Gemfile pins Ruby 3.1.6), bundle install, LANG=C.UTF-8 LC_ALL=C.UTF-8 bundle exec jekyll build. Panels were visually verified with headless Chromium screenshots.

Editorial ground rules established (keep these)

Every evidence-box claim needs a numbered source; derived arithmetic and analogies are explicitly labeled as ours (per-neuron speedup, slice-per-hair, HGP comparison).
Numbers describe a specific data release, not the animal — stated in ch. “A note on the receipts.”
Claims this repo’s own fabrication audit flagged as uncited are deliberately not used: the “$40M” figure, “50× larger”, the “91-beam” count, the conflicting 4 nm / 8 nm HI-MC resolutions, the repo-only CONNECTS acronym expansion. If these get sourced upstream, they can come back.
Health claims are split into delivered results (cited) vs. baselines/prospects (future-tensed, hedged) — never promise a cure by a date.
Newly verified sources beyond the repo: NINDS CONNECTS announcement, NIH awards release (>$3B / deliverables), GBD 2021 / WHO, MICrONS Nature 640 (2025) + Ding et al. wiring rule, Medtronic adaptive-DBS FDA approval (Feb 2025), NIH Research Matters speech-BCI (Mar 2025).

Requested next (pending, in priority order)

The inverse chapter (task: “Show the inverse. Disease, computation. We know stuff. Stuck. Imprecise. People dying. No map!! Need to do the following…”). Mirror of ch. 7: stand on the clinic bank and the computation bank looking back at the fog where the map should be. Structure: what medicine already knows (genes, cells, symptoms) → where it is stuck without circuit sight (trial-and-error meds, ~30% drug-resistant epilepsy [verify via WHO fact sheet], no Alzheimer’s cure) → same inversion for AI (20 W vs gigawatts; the one wiring sketch we copied launched CNNs; no further blueprints) → ends in a concrete numbered “what must be done” list mapping to CONNECTS routes + workforce + open data + sustained funding [sources 8, 15–17 already in the book]. Suggested placement: after ch. 7, before “Why is this so hard” (renumber 8→9 etc. — badge numbers are hardcoded spans).
Hitchhiker’s Guide + PDF edition (“combine this with a hitchhiker’s guide to brain mapping with summaries of all the key reference things — a pdf/book that takes the key site points”). Plan: new /book/guide/ reference companion (dataset cheat-sheets, pipeline stages from /datasets/workflow/, glossary from the site dictionary data, acronym decoder, key-papers reading list from content-library/journal-papers/), then print /book/ + /book/guide/ to PDF with headless Chromium, merge, commit as a downloadable asset linked from the book page.
Technical gap audit (“audit site for technical gaps”). Sweep technical-training/, content-library/, modules/ for missing/thin topics and internal inconsistencies; deliver docs/TECHNICAL_GAPS.md with prioritized, file-referenced findings. Known seeds from the earlier fact-extraction pass: HI-MC resolution stated as 4 nm vs 8 nm in different pages; mSEM-IBEAM vs ATUM described as the acquisition method in different pages; H01 z-resolution 33 vs 30 nm; MICrONS 80k neurons vs 200k cells and “2021 bioRxiv” vs “2025 release” not reconciled; C. elegans gap junctions 900 (text) vs 600 (caption); two different titles for the same H01 DOI and for Lappalainen et al. 2024.

Session-container notes for whoever resumes

Screenshots and Playwright scripts live in the session scratchpad (ephemeral).
Local preview: python3 -m http.server 8123 from _site/, then screenshot with Playwright using the pre-installed Chromium (NODE_PATH=$(npm root -g)).
The earlier fact-extraction summary (per-case-study numbers, contradictions, and which claims are uncited) is reproduced in essentials above; the case-study pages themselves are the ground truth to re-check against.