Skip to content

feat(web): the page fades in once on arrival — main and footer, opacity only, 700 ms under a still top bar - #96

Merged
AlecROndo merged 6 commits into
mainfrom
AlecROndo/anthropic-fade-in-effect
Nov 10, 2026
Merged

AlecROndo merged 6 commits into
mainfrom
AlecROndo/anthropic-fade-in-effect

Conversation

@AlecROndo

@AlecROndo AlecROndo commented Nov 9, 2026 •

Copy link
Copy Markdown
Owner

Summary

  • Entering the site, or moving between its pages, the content now fades in under a top bar that is already there, in the register of anthropic.com: subtle, once, and nothing travels.
  • One theme rule does it for all eleven surfaces: main, footer { animation: arrive 700ms var(--ease-out) both }. Every page renders one main and one footer, so the static pages get it through /theme.css and the apps when React mounts main. A tab or view changing inside a page does not replay it.
  • Opacity only, so the chip stays the one thing that moves on load. Under reduced motion the same fade is kept, by name, as the gentler variant rather than nothing.

Changes

  • web/rung0-viz/src/theme.css (and the four byte-identical copies): the arrive keyframe and rule, restated in the reduced-motion block; the header's third point now says the content fades in once.
  • web/DESIGN.md: "Nothing moves on load" becomes "nothing moves on load; the page fades in once"; the arrival joins the allowlist and the duration rule.
  • tests/test_ui_standard.py: main enters EXPLANATORY for running past 300 ms; test_the_page_fades_in_once_on_arrival holds its shape (one-shot, the ease-out token, a keyframe that moves nothing, kept under reduced motion); the allowlist's selector boundary is two-sided, so main does not exempt .split-main.

Test plan

  • uv run pytest — the full default suite passes
  • npm test in web/rung0-viz (the four other apps carry the same theme bytes; test_shared_chrome holds them equal)
  • Chrome, on the homepage, Rules and the built rung 0 app, with and without reduced motion: main.getAnimations() is the one arrive, 700 ms, one iteration; at 90 ms the content was at 0.64 opacity under a top bar at 1 (measured at 500 ms, before Alec asked for 700 on localhost)
  • Judged on localhost:4321 from the full Vercel build; 500 ms read a touch quick, so it is 700 ms
  • A phone and a throttled connection, where an app's slow mount shows bare paper before the fade starts

…ty only, 500 ms on --ease-out under a still top bar, one theme rule for all eleven surfaces

- `main, footer { animation: arrive 500ms var(--ease-out) both }` in the
  theme, copied to the five apps and linked by the static pages. Every
  surface renders one `main` and one `footer`, so the one rule is the whole
  site's arrival; in an app it plays when React mounts `main`, which is the
  first moment there is anything to see. A tab or view changing inside a
  page does not replay it, because `main` stays mounted.
- Opacity only: the chip stays the one thing that moves on load. Under
  reduced motion the same fade is kept, restated by name, as the gentler
  variant rather than nothing.
- `web/DESIGN.md`: "Nothing moves on load" becomes "nothing moves on load;
  the page fades in once", the arrival joins the allowlist, and the
  duration rule names it.
- `tests/test_ui_standard.py`: `main` enters EXPLANATORY for running past
  300 ms; `test_the_page_fades_in_once_on_arrival` holds its shape
  (one-shot, the ease-out token, a keyframe that moves nothing, kept under
  reduced motion); the allowlist's selector boundary is now two-sided, so
  `main` does not exempt `.split-main`.
- Checked in Chrome on the homepage, Rules and the built rung 0 app: at
  90 ms the content sits at 0.64 opacity under a top bar already at 1.
@vercel

vercel Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
drawmaha-solver Ready Ready Preview Nov 10, 2026 9:18am UTC

Request Review

@claude

claude Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Claude finished @AlecROndo's task in 40s —— View job


Review of PR #96: page fade-in on arrival

I read the diff only. I did not run the test suite or open the site in a browser.

The change is small and coherent. The arrive rule is opacity only, uses the ease-out token and is restated under reduced motion. The new test pins those properties. footer is a sibling of main in each site.tsx, so the two opacities don't compound.

Findings

  1. The main allowlist entry is broader than the fade it exempts. Severity: medium. Confidence: high.

    • The lookbehind added in tests/test_ui_standard.py stops main from matching .split-main. It still matches any selector containing the bare token main, for example main .panel or main.wrap .x.
    • A future rule such as main .foo { transition: opacity 600ms } would count as exempt from the 300 ms rule, the ease-out rule and the layout-property rule. The only check that the entry matches a rule that moves is the ghost test, and the main rule itself satisfies it.
    • A narrower fix is to exempt only a rule whose selector parts are exactly main and footer, or to special-case the arrival in Rule.explanatory.
  2. animation-fill-mode: both leaves an opacity animation attached to main and footer permanently. Severity: low. Confidence: medium.

    • The animation holds opacity 1 after it ends. Chrome keeps the element's stacking context for as long as the fill is active.
    • This could change z-order between main descendants (popovers, the solver's .sheet, sticky or fixed elements) and the top bar. The PR's Chrome check covers the homepage, Rules and the rung 0 app, not the solver sheet or the rung 3 deal view.
    • Using backwards instead of both would avoid it. That is enough here, because to is the element's natural state. The from frame still applies during the animation, which is the only time it matters.
  3. The test checks that the arrival rule exists, not that the reduced-motion copy is the one that wins. Severity: low. Confidence: medium.

    • The reduced-motion block repeats the same declaration. If a later reduced-motion rule set animation: none on main, the test would still pass, because _arrival_rules only looks for rules that animate.
    • A guard that no reduce rule gives main or footer an animation: none would close this.
  4. The allowlist comment line in tests/test_ui_standard.py is now very long. Severity: nit. Confidence: high.

    • The edit appended , and \main` not `.split-main`` to an already full line. Re-wrapping it would match the surrounding comment.
  5. Behaviour depends on an assumption the test doesn't enforce. Severity: low. Confidence: low.

    • Each page must keep exactly one main and one footer, and in the apps main must stay mounted across tab changes. The CSS comments and DESIGN.md both state this. If an app ever re-keys main or renders a conditional <main>, the fade would replay on every view change, and nothing would fail.
  6. Perceptual cost on large pages. Severity: informational. Confidence: low.

    • The wiki and the homepage are long. Fading the whole main as one layer promotes it to a compositor layer. This is probably fine, and your unchecked phone check is the right place to look.
    • Content is invisible for the first frame or so, until the animation's first paint. If the JS-rendered apps are slow to mount, the fade starts only after the mount. That is a flash of the empty paper, not a defect in this PR.

Docs and pedagogy

  • web/DESIGN.md and the theme.css header are consistent with the code.
  • The ease-out description in the CSS comment ("brings the words up fast and lets the last of the fade settle slowly") is accurate.
  • No game-theory or solver math is touched.

Not verified

  • I did not check that the five theme.css copies are byte-identical. The diff shows the same blob SHA for all five, so they are.
  • I did not run uv run pytest or npm test.

…t, so nothing under main is exempt; the fill is backwards, so the fade lets go when it ends; no sheet may quiet it and every surface renders one main and one footer

Review findings on #96, in order:

1. `main` as an EXPLANATORY key reached any selector containing the word,
   so `main .panel { transition: opacity 600ms }` would have passed the
   300 ms, easing and layout gates. A key that is itself a selector list
   now exempts only the rule whose selector is exactly that list;
   `_exempts` is the one matcher, shared by `Rule.explanatory` and the
   ghost test.
2. `both` held `to` after the fade; `to` is the page's own state, and an
   animation left filling forwards keeps a layer and a stacking context
   in some engines. `backwards`, and the test refuses a forwards fill.
3. A later `animation: none` on `main` or `footer` would have cut the fade
   without failing the arrival test: `test_no_sheet_quiets_the_arrival`,
   over every sheet and media block.
4. The allowlist comment re-wrapped.
5. The premise that one theme rule reaches every page is now held by
   `test_every_surface_renders_one_main_and_one_footer`: one `<main>`
   and one `<footer>` per surface, the app's `<main>` unkeyed and the
   first thing `App` returns.

Co-Authored-By: Claude <noreply@anthropic.com>
@claude

claude Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Claude finished @AlecROndo's task in 27s —— View job


Review of PR #96 after the second commit (a7298bd)

I read the diff only. I did not run uv run pytest or npm test, and I did not open the site in a browser.

The second commit addresses every finding from my first pass. I checked each against the code.

Earlier finding Status
main allowlist entry too broad Fixed. _exempts treats a key containing a comma as the whole selector list, compared with _parts(key) == _parts(selector). So main .panel and a bare main are no longer exempt, and the ghost test uses the same function.
fill-mode: both leaves the animation attached Fixed. Both the base and reduced-motion rules now use backwards, and the test rejects forwards and both.
Nothing stops animation: none on main or footer Fixed by test_no_sheet_quiets_the_arrival. It runs over every sheet and media block, and covers animation-name too.
Comment line too long Fixed. The allowlist comment is re-wrapped.
One main and one footer per page was not enforced Mostly fixed by test_every_surface_renders_one_main_and_one_footer. See 1 below.

Remaining findings

  1. The "mounted once, unconditionally" check is textual and brittle. Severity: low. Confidence: medium.

    • The test passes only if return ( and <> appear within the 5 lines above the first <main in App.tsx. A harmless reformat, such as a comment added above <main> or a different fragment style, fails the test. A real regression, such as {cond && <main>} or a <main> rendered by a child component, may not.
    • The <main\b and <footer\b counts are also plain regexes over file text. A <main inside a comment, a string or an inline script in a static page would be counted.
    • This is acceptable as a tripwire. The docstring should say it is a heuristic, or the test should be loosened to what it can really prove: one <main, no key=.
  2. The reduced-motion restatement is a duplicate that exists to satisfy the test. Severity: nit. Confidence: medium.

    • The theme.css comment says it is "restated so the standard's test can see it was kept on purpose". The block repeats the base declaration exactly. Outside the reduced-motion media query it is a no-op.
    • The new test_no_sheet_quiets_the_arrival already guards the property that matters, that nothing turns the fade off. Consider dropping the duplicate and the second reduced=True pass, which would remove 7 lines from each of 5 byte-identical copies.
    • If the duplicate stays, nothing wrong follows from it. The same declaration in two places can drift apart.
  3. web/DESIGN.md line 119 is no longer wrapped. Severity: nit. Confidence: high.

    • "The page's arrival 500 ms, once. Nothing else in the chrome over 300 ms per declaration — and the" runs past the file's wrap width. It should be re-wrapped to match the surrounding text.
  4. The CSS comment hedges on its own justification. Severity: nit. Confidence: low.

    • It says an animation left filling forwards "keeps the element a layer of its own in some engines". It is true that a fill-forwards opacity animation can keep a stacking context and a compositor layer. "In some engines" is vague. Naming the actual consequence, a stacking context that can change z-order against descendants, would teach more.
  5. Perceptual checks are still open. Severity: informational. Confidence: low.

    • Fading all of main as one layer on the long wiki and homepage pages may cost a frame or two on a phone. This is the PR's unchecked box.
    • In the apps the fade starts when React mounts main, not when the HTML arrives. A slow mount shows a flash of empty paper before the fade begins. That behaviour is by design, but it is worth checking on a throttled connection.
    • Static pages with no JavaScript get the fade from CSS alone. Pages with a hash link (#section) fade in at the scrolled position. That is probably fine.

Docs and pedagogy

  • web/DESIGN.md, the theme.css header and the test docstrings agree with each other and with the code.
  • The ease-out description is accurate.
  • No solver or game-theory math is touched, and no committed fixtures or figures are hand-edited.
  • The five theme.css copies share one blob SHA (916ef8c), so they are byte-identical.

Not verified

  • Test results, because CI runs no tests and I did not run the suite.
  • Behaviour in Chrome, in particular the stacking-context claim and the 500 ms feel.

No finding blocks the merge. 1 and 2 are the ones worth a look.

@AlecROndo

Copy link
Copy Markdown
Owner Author

Review findings dispositioned at a7298bd (fix(web): the arrival's allowlist key is the whole main, footer list …):

  1. Allowlist entry broader than the fade — fixed. The key is now the whole selector list main, footer; a key with a comma exempts only the rule whose selector is exactly that list (_exempts, shared by Rule.explanatory and the ghost test). Probed: main, main .panel, main.wrap .x, body.wide main.wrap are not exempt; main,\n footer and footer, main are; .sweep keeps its boundary semantics. Only the two arrival rules are exempt in the theme.
  2. both leaves a filling animation on main/footer — fixed. backwards, in the chrome rule and the reduced-motion restatement; the arrival test now refuses forwards/both. For the record, nothing under main today carries a z-index and the top bar is not sticky, so no z-order could have changed either way; the fill is now backwards because to is the page's own state and there is nothing to hold.
  3. Test finds a rule that animates but not one that quiets it — fixed. test_no_sheet_quiets_the_arrival, parametrized over every sheet and media block: no rule whose selector names main or footer sets animation or animation-name to none. Covers an app's own sheet too, which loads after the theme and would win.
  4. Long comment line — fixed, re-wrapped and reworded for the list-key semantics.
  5. One main, one footer, mounted once — fixed. test_every_surface_renders_one_main_and_one_footer: exactly one <main> and one <footer> per surface (five apps' App.tsx + site.tsx, the five generated pages, the archived Cover), the app's <main> unkeyed and the first thing App returns after <>.
  6. Perceptual cost on long pages — no change; informational. A whole-page opacity fade is one compositor layer for 500 ms and then released (see 2). The phone check stays on the PR's own checklist.

uv run pytest tests/test_ui_standard.py tests/test_shared_chrome.py tests/test_cover_page.py tests/test_agent_docs.py → 219 passed; the five theme.css copies share one checksum.

…an show; the arrival's comments name the stacking context a forwards fill would leave and why the reduced-motion block restates the fade

Second review pass on #96, in order:

1. `test_every_surface_renders_one_main_and_one_footer` no longer looks
   for `return (` and `<>` five lines above `<main` — a comment there would
   have failed it and `{cond && <main>}` passed it. It now strips comments,
   counts one `<main` and one `<footer` per shell, refuses a `key=` on
   `<main` and a conditional on its opening line, and its docstring says
   what escapes it.
2. The reduced-motion restatement stays: the standard quiets by name
   (`test_every_pulse_and_travel_is_quieted_under_reduced_motion` fails for
   an animating selector with no reduced-motion rule of its own), and the
   arrival's quiet form is the same fade. The comment now says so instead
   of "so the test can see it".
3. `web/DESIGN.md` step 4 re-wrapped.
4. The theme comment and the test docstring name the consequence of a
   forwards fill — `main` kept as a stacking context, under which nothing
   inside it could rise above the top bar — in place of "in some engines".

Co-Authored-By: Claude <noreply@anthropic.com>
@claude

claude Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Claude finished @AlecROndo's task in 34s —— View job


Review of PR #96 after the third commit (48d7923)

I read the diff of 48d7923 only. I did not run uv run pytest or npm test, and I did not open the site in a browser.

The commit answers the two remaining findings from my last pass that were worth acting on. The one-main-one-footer test is looser and says what it can't prove. The reduced-motion restatement stays, now with a stated reason. Comments and docs agree with the code. The five theme.css copies share one blob SHA (0ecfc91), so they are byte-identical.

Earlier finding Status
Textual "mounted once" check was brittle Fixed. It strips comments, counts one <main and one <footer, refuses key= on <main, and refuses a conditional on the opening line. The docstring lists what escapes it.
Reduced-motion restatement is a duplicate Resolved by explanation. The comment now gives a real reason: the standard requires every animating selector to have its own reduced-motion rule.
DESIGN.md wrapping Fixed.
Vague "in some engines" wording Fixed. The comment now names the consequence: a stacking context on main that stops a popover or sheet inside it from rising above the top bar.

Remaining findings

  1. The conditional check scans only the first line containing <main. Severity: low. Confidence: medium.

    • next(line for line in text.splitlines() if "<main" in line) takes the first matching line, and the regex &&|\?[^.]|\bif\b then runs on that line only.
    • A false positive is possible. A static page whose <main line has a ? in an attribute value or URL, such as <main>, would fail. A ? not followed by . is a loose proxy for a ternary.
    • The docstring admits that conditionals broken across lines escape it, so this is a documented limit. It is still the check most likely to fail for a reason unrelated to the arrival.
  2. Comment stripping can remove real code. Severity: nit. Confidence: low.

    • /\*.*?\*/ with DOTALL also matches inside string literals and regexes, such as a glob like "src/**/*.ts" in a shell file. It cannot hide a tag in the files tested here unless a /* appears before a <main with a */ after it.
    • The cost is low and it fails safe in the likeliest direction, a count of zero, which fails loudly.
  3. The stacking-context claim is stated as fact, but I did not verify it in the browser. Severity: informational. Confidence: medium.

    • Opacity below 1 creates a stacking context, and an animation with a forwards fill holds the end value. At opacity 1 the animation's effect still creates a stacking context while it fills. That matches the CSS Animations and Compositing specs as I remember them, so the new wording looks accurate.
    • The earlier reply said nothing under main carries a z-index today and the top bar is not sticky. The comment's example (a popover or sheet raised above the top bar) is therefore hypothetical. That is fine for a rationale, but it is a future-proofing argument, not a bug being fixed.
  4. Perceptual checks are still open. Severity: informational. Confidence: low.

    • The PR's unchecked box, the 4× Animations panel and a phone, is the right place to settle the feel of 500 ms and the cost of fading a long main as one layer.

Docs and pedagogy

  • The theme.css comments, the DESIGN.md text and the test docstrings agree with each other.
  • No solver or game-theory math is touched, and no committed fixtures or figures are hand-edited.
  • The commit subject is long, but it follows the fix(web): <the finding> convention.

Not verified

  • Test results, because CI runs no tests and I did not run the suite.
  • Behaviour in Chrome, including the stacking-context claim and the 500 ms feel.

No finding blocks the merge. Finding 1 is the only one worth a look.

@AlecROndo

Copy link
Copy Markdown
Owner Author

Second-pass findings dispositioned at 48d7923:

  1. Mount check textual and brittle — fixed. test_every_surface_renders_one_main_and_one_footer no longer requires return ( and <> within five lines of <main. It strips <!-- --> and /* */ comments, counts exactly one <main and one <footer per shell, refuses key= on <main and a conditional (&&, ?, if) on its opening line, and its docstring states what escapes it (a <main> from a child component, a conditional broken across lines). A tripwire that claims only what the text can show.
  2. Reduced-motion restatement is a duplicate — not changed, with a reason: the standard quiets by name. test_every_pulse_and_travel_is_quieted_under_reduced_motion fails for any selector that animates and has no reduced-motion rule of its own that quiets it (animation: none, or a keyframe with no transform). Dropping the restatement fails that test on all twelve sheets. The arrival's quiet form is the same fade, since a fade with nothing moving is already the gentler variant. The comment now says this instead of "so the test can see it was kept on purpose", and the two declarations are held equal by test_the_page_fades_in_once_on_arrival's reduced=True pass, so they cannot drift apart.
  3. DESIGN.md line 119 unwrapped — fixed, step 4 re-wrapped at the file's width.
  4. "In some engines" hedges — fixed. The theme comment and the test docstring now name the consequence: a forwards fill keeps main a stacking context, which flattens the z-order of everything inside it against the chrome outside it, so a popover or sheet raised inside main could no longer rise above the top bar.
  5. Perceptual checks — informational, no change; the phone and throttled-connection checks remain the PR's own open boxes.

uv run pytest tests/test_ui_standard.py tests/test_shared_chrome.py tests/test_cover_page.py tests/test_agent_docs.py → 219 passed; the five theme.css copies share one checksum.

…re `<main`, not any `?` on its line

Third review pass on #96: `&&|\?[^.]|\bif\b` over the whole opening line
would have failed a `<main` whose attribute carried a `?`, and `?[^.]`
was a loose proxy for a ternary. The check is now `(&&|\?|:)\s*\(?\s*<main`,
which is how JSX renders a tag conditionally and nothing else: `{open &&
<main`, `{x ? (<main`, `: <main` fail; `<main className={a ? "x" : "y"}>`
and a `?` in a URL do not. Probed on both sides before committing.

Co-Authored-By: Claude <noreply@anthropic.com>
@AlecROndo

Copy link
Copy Markdown
Owner Author

Third-pass findings dispositioned at the branch head (fix(web): the one-main test reads the conditional operator right before <main, not any ? on its line):

  1. Conditional check scans the whole <main line — fixed. The regex is now (&&|\?|:)\s*\(?\s*<main\b: a conditional operator immediately before the tag, which is how JSX renders a tag conditionally and nothing else. Probed: {open && <main, {ready ? (<main>, {x ? <main> : null}, : <main>} fail; <main className="wrap">, <main href="/x?y=1">, the rung-3 template literal and <main className={tab === "play" ? "a" : "b"}> pass. The loose \bif\b is gone.
  2. Comment stripping could remove real code — not changed, with a reason: the stripped text is only the five App.tsx + site.tsx pairs and the six static pages, none of which carry a glob or a /* inside a string; and as noted, the failure direction is a count of zero, which fails loudly rather than hiding a tag. The docstring already names the test a tripwire over source text.
  3. Stacking-context rationale is future-proofing — agreed, no change. The theme comment is written in the conditional ("would flatten", "could no longer rise") for that reason; today nothing under main carries a z-index, so backwards fixes no live bug, it keeps one from being possible.
  4. Perceptual checks — informational; the 4x Animations panel and phone check stay the PR's own open box for its author.

uv run pytest tests/test_ui_standard.py tests/test_shared_chrome.py tests/test_cover_page.py tests/test_agent_docs.py → 219 passed; theme.css untouched this round, the five copies still share one checksum.

@claude

claude Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Claude finished @AlecROndo's task in 18s —— View job


Review of PR #96 after the fourth commit (838853a)

I read the diff of 838853a only. I did not run uv run pytest or npm test, and I did not open the site in a browser.

The commit changes one assertion in test_every_surface_renders_one_main_and_one_footer (tests/test_ui_standard.py). The old &&|\?[^.]|\bif\b over the whole opening line is now (&&|\?|:)\s*\(?\s*<main\b. That answers my last finding 1: a ? in an attribute or URL no longer trips it, and the failure message and docstring match the new behaviour.

Findings

  1. : before <main can false-positive on legitimate JSX or HTML. Severity: nit. Confidence: low.

    • A : directly before <main also matches a TypeScript return-type annotation or a case x: <main…. Neither occurs in these shells today.
    • If one is added, the test fails loudly and names the line, so the cost is small.
  2. The check is still tied to the first line containing <main. Severity: nit. Confidence: medium.

    • It is the same documented limit as before. A <main that follows a conditional on the previous line, such as {open && ( then <main> on the next line, escapes it. That is the form Prettier produces for multi-line JSX.
    • The docstring says conditionals broken across lines escape, so this is stated. Prettier-formatted code is the likeliest way to hit it, though, so the tripwire is weaker than its wording suggests. Checking the non-blank text immediately before <main across lines (for example re.search(r"(&&|\?|:)\s*\(?\s*<main\b", text) over the whole comment-stripped text) would close it at no extra cost.
  3. No test for the test. Severity: informational. Confidence: low.

    • The commit message says both sides were probed by hand. Nothing in the repo pins those probes, so a later regex edit could loosen it unnoticed. A small parametrized check on the regex alone would hold them.

Docs and pedagogy

  • The docstring and the assertion message agree with the code.
  • No solver or game-theory math is touched. No fixtures or figures are hand-edited. Only tests/test_ui_standard.py changed in this commit, and the five theme.css copies still share one blob SHA (0ecfc91).

Not verified

  • Test results, because CI runs no tests and I did not run the suite.
  • Behaviour in Chrome, and the open phone and throttled-connection checks on the PR's own checklist.

No finding blocks the merge. Finding 2 is the one worth a look.

…ss lines, and the pattern's probes are a test

Fourth review pass on #96: the operator check ran on the one line that
opened `<main`, so Prettier's `{open && (` newline `<main` escaped it.
`CONDITIONAL_MAIN` now runs over the whole comment-stripped shell, and the
cases probed by hand in the last commit — four conditional renders, the
five real openings and a `?` in an attribute — are a parametrized test,
so a later edit to the pattern cannot loosen it unnoticed. The `:` stays:
a ternary's else branch is a real conditional render, and a `case x:`
before `<main` would fail loudly by line.

Co-Authored-By: Claude <noreply@anthropic.com>
@AlecROndo

Copy link
Copy Markdown
Owner Author

Fourth-pass findings dispositioned at the branch head (fix(web): the one-main test reads the conditional before <main across lines, and the pattern's probes are a test):

  1. : before <main can false-positive on case x: or a return type — not changed, with a reason: : <main is a ternary's else branch, a real conditional render, so dropping : would let {x ? null : <main>} through. Neither a case nor a return-type annotation precedes <main in any shell, and if one ever does the test fails loudly and quotes the match.
  2. Tied to the first <main line — fixed. CONDITIONAL_MAIN now runs over the whole comment-stripped shell text, so Prettier's {open && ( newline <main is caught; the docstring no longer claims a cross-line conditional escapes.
  3. No test for the test — fixed. The pattern is a module constant and test_the_conditional_main_pattern_reads_jsx_conditionals pins ten cases: five conditional renders (including the multi-line form) and five legitimate openings (the real App.tsx shape, the rung-3 template literal, a ternary inside className, a ? in a URL).

uv run pytest tests/test_ui_standard.py tests/test_shared_chrome.py tests/test_cover_page.py tests/test_agent_docs.py → 229 passed; theme.css untouched, the five copies still share one checksum.

@claude

claude Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Claude finished @AlecROndo's task in 12s —— View job


Review of PR #96 after the fifth commit (a4476b6)

I read the diff of a4476b6 only. I did not run uv run pytest or npm test, and I did not open the site in a browser.

The commit answers both findings from my last pass. The conditional check now runs over the whole comment-stripped text, so a conditional on the previous line ({open && ( then <main) is caught. The pattern's probes are now a parametrized test, test_the_conditional_main_pattern_reads_jsx_conditionals. The docstring and the assertion message match the new behaviour.

Findings

  1. : before <main can still false-positive. Severity: nit. Confidence: low.

    • A TypeScript return-type annotation, a case x: <main… or an object literal like { a: <main/> } would match. None occurs in these shells today.
    • The failure is loud and names the matched text, so it's cheap to fix when it happens.
  2. The probes lack a case for a ? or : inside the text between tags. Severity: nit. Confidence: low.

    • For example, a <p>why?</p> on the line above a <main, with nothing between them. The pattern needs \s*\(?\s* between the operator and <main, so it does not match. It is correct today, and one extra probe would hold it.
  3. Comment stripping can still drop real code. Severity: nit. Confidence: low.

    • This is unchanged from last round. The failure direction is a count of zero, which fails loudly.

Docs and pedagogy

  • The CONDITIONAL_MAIN comment, the test docstring and the assertion message agree with each other and with the code.
  • No solver or game-theory math is touched. No fixtures or figures are hand-edited.
  • This commit touches only tests/test_ui_standard.py, so the five theme.css copies are unchanged.

Not verified

  • Test results, because CI runs no tests and I did not run the suite.
  • Behaviour in Chrome. The 4× Animations panel check, the phone check and the throttled-connection check remain open on the PR's own checklist.

No finding blocks the merge. Every earlier finding is addressed or answered with a stated reason.

@AlecROndo

Copy link
Copy Markdown
Owner Author

Fifth-pass findings dispositioned at a4476b6, with no further commit — each is a nit on the tripwire test with no wrong claim, no failing test and no behaviour change in theme.css behind it:

  1. : before <main can false-positive on a return type, case x: or { a: <main/> } — not changed. : <main is a ternary's else branch, a real conditional render; dropping it would let {x ? null : <main>} through. None of the three shapes precedes <main in any shell, and if one ever does the test fails loudly and quotes the match, which is the right outcome for a tripwire.
  2. No probe for a ? or : in text between tags — not changed. As the finding says, the pattern allows only whitespace and ( between the operator and <main, so <p>why?</p> on the line above cannot match; the pattern is correct today, and the ten pinned probes already hold both directions (a ? in an attribute and in a URL on the legitimate side).
  3. Comment stripping could drop real code — unchanged from the previous pass, same reason: the stripped text is only these eleven shells, none carries a /* in a string, and the failure direction is a count of zero, which fails loudly.

Across all five passes: 1–5 of the first review fixed (a7298bd); 1, 3, 4 of the second fixed and 2 kept with a reason (48d7923); 1 of the third fixed (838853a); 2 and 3 of the fourth fixed (a4476b6); every remaining item is a nit or informational with a stated reason. The phone, throttled-connection and 4x Animations checks remain the PR's own open boxes for its author.

uv run pytest tests/test_ui_standard.py tests/test_shared_chrome.py tests/test_cover_page.py tests/test_agent_docs.py → 229 passed at a4476b6; the five theme.css copies share one checksum.

…ick on localhost; the test, the standard and the allowlist reason say 700
@claude

claude Bot commented Nov 10, 2026 •

Copy link
Copy Markdown

Claude finished @AlecROndo's task in 19s —— View job


Review of PR #96 after the sixth commit (8f17cd7)

I read the diff of 8f17cd7 only. I did not run uv run pytest or npm test, and I did not open the site in a browser.

The commit changes the arrival from 500 ms to 700 ms. It touches the arrive rule and its reduced-motion restatement in the theme, the ARRIVAL_MS constant and the allowlist reason in tests/test_ui_standard.py, and three places in web/DESIGN.md. I grepped for leftover 500 ms or 500ms in web/DESIGN.md, the test file, the theme.css copies and web/cover. None remain.

Findings

  1. The PR title and description still say 500 ms. Severity: nit. Confidence: high.

    • The title reads "…500 ms under a still top bar". The summary and the test plan also say 500 ms, including "arrive, 500 ms" and the Chrome check numbers.
    • They now disagree with the code. Update them before merge.
  2. The 700 ms arrival sits beside the 700 ms .reveal that the standard cites as the anti-example. Severity: informational. Confidence: low.

    • The DESIGN.md row says the old homepage sections rose in on scroll at 700 ms, and the new row also has 700 ms. The row says the fade is acceptable and the rise was not because the rise moved things on scroll. That is a different reason from the duration, so nothing is wrong.
    • A reader skimming the row may think the duration is what changed. No edit is needed.
  3. The 300 ms rule has a second arrival-sized exception. Severity: informational. Confidence: medium.

    • Raising the arrival to 700 ms is still covered by the main, footer allowlist entry. The test pins ARRIVAL_MS, so a future edit to the duration fails loudly. That is the right shape.
    • The "feel" was judged on localhost, as the commit subject says. The phone and throttled-connection checks on the PR's checklist matter more at 700 ms, because on a slow mount the empty paper is shown before the fade begins.

Docs and pedagogy

  • The theme.css comment, the DESIGN.md rows and the test constant agree on 700 ms.
  • The five theme.css copies share one blob SHA (537a721), so they are byte-identical.
  • No solver or game-theory math is touched, and no committed fixtures or figures are hand-edited.

Not verified

  • Test results, because CI runs no tests and I did not run the suite.
  • Behaviour in Chrome, and the open phone, throttled-connection and 4× Animations panel checks on the PR's own checklist.

No finding blocks the merge. Finding 1 is the only one that needs an action.

@AlecROndo
AlecROndo merged commit 7b09204 into main Nov 10, 2026
3 checks passed
@AlecROndo
AlecROndo deleted the AlecROndo/anthropic-fade-in-effect branch November 10, 2026 09:19
@AlecROndo AlecROndo changed the title feat(web): the page fades in once on arrival — main and footer, opacity only, 500 ms under a still top bar feat(web): the page fades in once on arrival — main and footer, opacity only, 700 ms under a still top bar Nov 10, 2026
@AlecROndo

Copy link
Copy Markdown
Owner Author

Sixth-pass finding 1 — fixed: the PR title and body now say 700 ms, and the Chrome numbers are marked as measured at the earlier 500. Findings 2 and 3 are informational; no change. The phone and throttled-connection check stays the PR's open box.

This branch was successfully deployed

1 active deployment
Preview — 8f17cd70 Deployed Nov 10, 2026 by vercel[bot]
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