Skip to content

feat(crosscheck): make the site the EVerest OCPP study, arranged like balfrontower.org - #146

Merged
hnipps merged 39 commits into
mainfrom
worktree-everest-site
Oct 30, 2026
Merged

hnipps merged 39 commits into
mainfrom
worktree-everest-site

Conversation

@hnipps

@hnipps hnipps commented Oct 29, 2026 •

Copy link
Copy Markdown
Member

Tier: 2
Intent: intent/2026-10-28-everest-site-brief.md
Spec: intent/2026-10-28-everest-site-brief.md

Why

The old landing page asked readers to wade through too much text before they knew what it was about. The site now presents one thing, the EVerest OCPP 1.6 study, arranged like balfrontower.org: a few one-line sections that open in place, questions that filter the evidence, and a grouped list of evidence entries that each carry their strength.

What changed

  • site/index.html is the study. everest.html forwards old links to it, fragment included.
  • Each question shows a short answer and the evidence for it. A second filter narrows by strength (proved, tested, decided by Harry, assumed, not yet reached).
  • The Crosscheck "Counterproof" mark, a glossary with EVerest and OCPP, and an updated DESIGN.md and PRODUCT.md.

Second round (commit 4c4dccf)

Two design critiques (27/40, then 29/40) drove this round.

  • The first screen names the cause: EVerest restarts the schedule "a leap second early".
  • Both filters live in the URL fragment (#q-proof&s=tested), so Back, refresh and shared links keep them. Strength counts warn of dead ends, and a bad link is reported under the Questions label.
  • Each answer exists once, under its question in the evidence list, so readers without JavaScript see it. check-site fails if a question and its answer drift apart.
  • The answers to "What went wrong?" and "Does a one-second error matter?" link the filed EVerest issue #3323.
  • The prose is plain, with one idea per answer paragraph. Every human decision is credited to Harry, and Harry's first-person account is set apart as a quote.
  • The title is now "EVerest and OCPP 1.6: a formal study". check-public-site follows it.
  • Accessibility: numeric ledgers stay tables on phones, standalone links have 44px hit areas, entry titles are headings, and strength buttons read their counts.
  • Tested and Not yet reached have new icons, the filter row lost its redundant labels, and the copyright sits centred at the page bottom.

Header type (commit e3c5b4f)

The 28px wordmark and the 32px title were nearly the same size, so the brand competed with the study's name. The lockup now has an 18px mark and a 20px wordmark at weight 520, over a light 36px title (weight 320) that scales down on phones. Harry picked it from three variants in Impeccable's live mode.

Findable, shareable and readable by agents (commit ee50c93)

This commit follows intent/2026-10-30-site-discoverability.md. It changes only each page's <head> and adds generated files.

  • Each page has a canonical URL, Open Graph and X card tags with a 1200x630 preview card, favicons, and schema.org JSON-LD: a TechArticle for the study and a DefinedTermSet for the glossary.
  • The build writes robots.txt, sitemap.xml and llms.txt. One origin constant in scripts/build-site.mjs feeds every absolute URL.
  • check-public-site.mjs refuses a public URL whose pages name another origin. check-site.mjs requires absolute links into the origin to resolve, and each card to match its declared size.
  • Headless Chrome renders both pages pixel-identical with and without the new head block, at 1440px and 390px.

Glossary and phone layout (commit e82bb1a)

  • The glossary reads like the study page. It uses the evidence-list layout. A 28px "Glossary" heading sits with a plain "Back to the study" link. The terms are grouped as on the study page: EVerest and OCPP, Proof, Tests, Assumptions and People.
  • Finding a term. A jump list covers all thirteen terms, and each term links to itself.
  • Plainer longer accounts. They are one idea per paragraph. "Profile", "warning" and the smallest example are explained, and each line is measured in its own font size.
  • A route in from the study. Three evidence entries link to the glossary account for their term. A ?explain= link for an unknown term forwards to the matching glossary entry.
  • One name for the model. "Reference program" is now "reference model", the term Lean-based differential testing uses, as in Cedar's Lean model and Resonate's reference model. "The model" stays as its short form. Term ids are unchanged, so existing links keep working.
  • Phones. Every section's first line shows in full. The label runs into its sentence on the same line, and the plus marks hang in the page margin, so every text block has the same 24px margin on both sides. Evidence text used to sit 42px from the left.
  • Matching shared parts. The study title breaks onto the same lines on both pages. The shared title rule now sets a balanced wrap whatever tag it sits in.

Scope

  • Every number still renders from site/everest-study.json. Copy was regrouped and rewritten for plainness, not changed in substance.
  • The brief now records that sections start closed and that paragraphs hold one idea each.
  • Deployment to the home server is not part of this PR.

Blast Radius

Only the static site under site/ and its generated files, its Node build and check scripts, and the design, product and intent notes change. No plugin, MCP server or CGV code is touched, and no protected surface.

Verification

  • node scripts/build-site.mjs --check, node scripts/check-site.mjs and node --test scripts/site.test.mjs scripts/check-public-site.test.mjs all pass (16 tests).
  • A headless-Chrome script (.impeccable/verify-filters.mjs, not committed) passes 36 of 36 behaviour checks at 1440px and 390px. It covers deep links, Back, strength counts, the bad-link notice on a phone, answers without JavaScript, phone tables and screen-reader names.
  • A second headless-Chrome script (.impeccable/compare-globals.mjs, not committed) finds the lockup, title, footer and copyright identical on both pages at 1440, 700 and 390px. It also measures every glossary block at 24px from both edges at 390px.
  • The Impeccable detector reports no findings on glossary.html. On index.html it reports one "flat type hierarchy" warning, which is a false positive: the scan cannot read the title's min(36px, 8.4vw) size, which renders at 36px. The overlay's 14 phone hits are ledger headers clipped to 1px for screen readers on purpose.
  • Design critiques: the study page went from 27 to 28 to 29 out of 40. The glossary scored 22 out of 32 before its last round of fixes.
  • Known gap: at 1440x900, after choosing "What went wrong?", the first evidence title sits at y=926, just below the fold.

🤖 Generated with Claude Code

Canvas and others added 30 commits October 28, 2026 13:59
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The home page is now the case study, rebuilt around index tabs and
questions that filter the evidence by topic and by strength. everest.html
forwards old links, and the footer names Crosscheck as the tool behind
the study. Every number still renders from everest-study.json.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The finish review found short answers and the Repeat tab that read
stronger than the study. They now carry the signed-reading premise, the
per-leap-second drift, vacuous passes, the generator bounds and the full
trusted base. Repeated table captions become screen-reader only, prose
keeps a 70ch measure, and replay commands wrap under themselves. PRODUCT.md
and the journal now describe the case-study site.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drop the hero and the nav header. The page opens with a title line, five
one-line sections that open in place, and a Filter row of questions and
evidence strengths with drawn icons. The evidence becomes a grouped list
of collapsed entries, so the default page fits in about two screens.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two new opening lines say what EVerest and OCPP are, quoting and linking
their own documentation, so readers need no prior knowledge. Choosing a
filter now keeps the filter row at the top of the screen instead of
jumping; without JavaScript each entry links back to the filters. The
page margins are wider.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… CLI

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…tower.org

Cap the content at 1200px with gutters up to 240px, narrow the label
column, and set each answer's question above its text. Phones keep
24px gutters.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The About section already introduces both, so the separate opening lines
go. The glossary gains an EVerest entry, and both entries carry the
fuller account and links to their own documentation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The mark, two proof signs that face each other but do not meet, replaces
the placeholder ticks above the study title as a lockup linking to the
repository, and sits before the Crosscheck link in the footer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…he serif

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Raise the title to the 32px headline size (26px on phones) so it no
longer competes with the 22px wordmark, set the wordmark and mark in ink,
and cap an opened section's first line at a reading measure.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… no longer has

A browser audit of every page and state found rules for the old masthead,
landing intro, button, install steps, publication path, clock hero and
three table variants matching nothing. Removing them changes no pixel on
the study, glossary, phone or filtered views.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DESIGN.md and its sidecar now document the Balfron-style page as built:
the lockup, one-line sections, filter row, answer band and grouped
evidence entries. The hero, index tabs, masthead and filled button are
gone from the system, and three unused colour tokens leave the stylesheet.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A cold load of index.html#q-proof left the answer scrolled off-screen because the browser's own fragment scroll to the no-JS anchor spans ran after the script anchored. With JS the anchor ids are removed, scroll restoration is manual, and the first anchor runs after the load event inside a frame.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e page to fit

Choosing a question opens every visible matching entry; clearing it closes only the entries the filter opened. The 100vh minimum height on the evidence list is replaced by a measured padding on body, set in anchor() to exactly what the target needs to reach the top of the screen.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ossary

The Finding line is never clamped and its two times carry the weight, with the highlight on 10:59:59 only. A1, A2, D4, I4, N4 and Q4 get a plain gloss supported by the pinned spec and report. The glossary gains schedule anchor, signed reading, vacuous pass, tolerance and correspondence review, and drops Dafny, CGV and byfuglien, which the study page never uses. The glossary title names the study.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ach strength once

The eight questions fall into three labelled groups. The strength filter reads as secondary in muted ink. The evidence-at-a-glance entry leads the list in its own Overview group, and the no-JS jump for the tested question now lands on the test results. One table in help.js names each strength, and index.html and the build script use the same names.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… phone

Help terms leave the section summaries. The answer count is announced as a polite status. Question links reach 44px and help terms get a larger hit area without moving text. Not yet reached is drawn as three dots so it no longer resembles the dashed assumed circle. The answer band has an Answer label, and phones keep the back-to-filters links when JavaScript runs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The pinned spec lets any limit conform when the model gives none, and
the caveat that uses the gloss is about exactly those instants.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
hnipps and others added 9 commits October 29, 2026 15:38
…closed

Shorten the About line so it no longer truncates, and drop the reading
caps that made the About and Finding lines wrap when opened.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Link the Open Charge Alliance's OCPP 1.6 download from About, from the
evidence that says the standard's wording admits another reading, from
the glossary and from the footer. People now closes the evidence list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ne edge

On wider screens the labels sit in a 92px margin column, so every first
line and the paragraphs under it share one left edge, and an opened
section gets room to end before the next. Phones keep the stacked label.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The credit, this study's links and links that leave the site now sit in
three labelled columns, stacked in that order on phones, in place of
one wrapping row. X and GitHub name whose accounts they are.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ithout script

Two design critiques (27/40, 29/40) drove this pass on the EVerest study page.

- The first screen names the cause (a leap second early); a hint under Questions says how to use them.
- Both filters live in the URL fragment, strength counts warn of dead ends, and a bad link is reported where it can be seen.
- Answers exist once, under their question, so readers without script see them; check-site fails if a question and its answer drift.
- The answers that matter name the filed EVerest issue #3323.
- Prose is plain, one idea per answer paragraph, and every human decision is credited to Harry; Harry's first-person account is set as a quote.
- Tested and Not yet reached have new icons; the filter row lost its redundant labels; the copyright sits centred at the page bottom.
- Numeric ledgers stay tables on phones, standalone links have 44px hit areas, entry titles are headings, and strength buttons read their counts.
- The title is EVerest and OCPP 1.6: a formal study, and check-public-site follows it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… study title

The 28px wordmark and the 32px title were nearly the same size, so the brand competed with the study's name. Chosen in live mode: an 18px mark with a 20px wordmark at weight 520, over a light 36px title (weight 320, opsz 72) that scales down on phones. The glossary shares the lockup and changes with it. DESIGN.md records the new values.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e by agents

Each page now carries an absolute canonical URL, Open Graph and X card
tags with a 1200x630 preview card, the Counterproof mark as favicon.ico,
favicon.svg and apple-touch-icon.png, and schema.org JSON-LD (a
TechArticle for the study, a DefinedTermSet for the glossary). The build
also writes robots.txt, sitemap.xml and llms.txt. The JSON record is
linked as an alternate of the study page.

One origin constant in scripts/build-site.mjs feeds every absolute URL.
check-public-site.mjs refuses a public URL whose pages name another
origin, and byte-checks the new files. check-site.mjs requires absolute
links into the origin to resolve in the release and each card to match
its declared size. scripts/render-site-images.mjs draws the icons and
cards with ImageMagick and headless Chrome, using the site's own fonts.

Only <head> changes. Headless Chrome renders index.html and
glossary.html identically with and without the new head block, at 1440
and 390px wide (0 differing pixels; a one-letter title change moves
7,390).

Intent: intent/2026-10-30-site-discoverability.md

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…he phone layout

- The glossary is set as the study page's evidence list: a Glossary heading on the content column, terms grouped as on the study page (EVerest and OCPP, Proof, Tests, Assumptions, People), a jump list of all thirteen terms, terms that link to themselves, and a plain Back to the study link.
- The glossary's longer accounts are plain and one idea per paragraph. Three evidence entries on the study page link to them, and each line is measured in its own size.
- Reference program becomes reference model, the term Lean-based differential testing uses; the model stays as its short form.
- On phones every section first line shows in full, the label runs into its sentence on the same line, and plus marks hang in the page margin so every text block has the same 24px margin on both sides.
- The study title wraps the same on both pages: the shared title rule sets a balanced wrap, whatever tag it is in.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@hnipps
hnipps merged commit 78d4a84 into main Oct 30, 2026
8 checks passed
@hnipps
hnipps deleted the worktree-everest-site branch October 30, 2026 19:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant