crypto-dashboard-build

Build source for the Crypto Value-Capture Watchlist dashboard.

This repo exists for one reason: so a scheduled cloud session can rebuild and republish the dashboard every day, unattended, with no desktop connected — which was impossible while the dashboard was a single 124 KB hand-edited HTML file.

Why the split

A cloud session can read a project doc, but only into its context. Re-emitting a 124 KB file to disk means retyping it, which is both expensive and a corruption risk — so the publication rule was "stand pat", and the dashboard drifted days behind the data.

Splitting the file fixes the actual constraint:

	size	changes	who edits it
shell.html	~70 KB	rarely	a human, deliberately
data/*.js	~53 KB across 16 files	every run	the scheduled run

A run fetches the shell mechanically (git clone), edits only the few kilobytes that actually changed, and rebuilds. Nothing is ever retyped.

Layout

shell.html          engine: HTML, CSS, chart code, law comments. Data constants replaced
                    IN PLACE by /*__CONST:NAME__*/ markers — nothing reordered.
const_meta.json     how each constant is re-emitted (exact declaration spelling/spacing)
data/               the daily payload — one file per constant
  BAKED_ASOF.js  LAST_KNOWN.js  TOKENS.js  HISTORY_CSV.js  BACKFILL_CSV.js
  UNLOCKS.js  NETFLOW.js  IN12.js  NET_MODEL.js  DILNEG.js  EST_GROSS.js
  SPARKS30.js  BENCH.js  BENCH2.js  BH.js  RANKS.js
  _html.json        rendered prose that carries data claims (stamps, hints, axis notes)
  _comments.json    per-constant basis annotations
dist/dashboard.html the built artifact — this is what Notion fetches
tools/              split.py · build.py · validate_build.js · validate_prose.py · scope_gate.py · publish.sh

Daily run

git clone https://github.com/Elokane/crypto-dashboard-build.git && cd crypto-dashboard-build
# edit only what changed, e.g. data/LAST_KNOWN.js, data/TOKENS.js, data/BH.js
GH_TOKEN=… GH_REPO=Elokane/crypto-dashboard-build ./tools/publish.sh \
  --stamp "build v19.2 · board v14.1 · baked 2026-07-28"

publish.sh builds, validates, refuses to push if a credential appears in tracked content, pushes, and prints a commit-pinned raw URL:

https://raw.githubusercontent.com/Elokane/crypto-dashboard-build/<sha>/dist/dashboard.html

Pinning to the SHA matters — raw.githubusercontent.com caches by branch ref, so a branch URL can serve a stale build for minutes. The commit URL is content-addressed. The script then re-fetches that URL anonymously and compares SHA-256 against the local build, so a private repo or a propagation lag fails loudly instead of silently publishing yesterday's numbers.

Guarantees

• Lossless split. build.py on unmodified data reproduces the original file byte-for-byte. That was the acceptance test for the split, not an approximation.
• Fail-closed build. Unfilled marker, missing data file, orphaned data file, or more than one <script> block → hard error, nothing written.
• Validated publish. validate_build.js evaluates the built engine in a DOM stub and asserts the data invariants: every seat has a price basis and a rank, ranks are a contiguous 1..n, benchmark series are parallel and strictly increasing in date (a duplicated append is the failure mode there), no NaN or negative in any numeric field, B/H scores in range, IN12 and NETFLOW in-legs agree.

Verified against four injected faults — syntax break, out-of-range score, duplicated benchmark date, missing data file — all caught, none published.

What this is not

validate_build.js checks whether the file we are about to publish is internally coherent. It says nothing about whether the numbers are right — that is validate_prints.py's job, upstream, and neither substitutes for the other. validate_prose.py checks that prose restating a number still matches the number, and scope_gate.py dash (added 2026-08-24) enforces the Scope law on every rendered string: word budgets per field and a forbidden-pattern list (run narration, dated run-stamps, verification chatter, engine instructions, internal ids). Rendered prose is a current statement of fact, rewritten each run — never an accumulation of dated notes.

Research notes, not investment advice.