Public: anyone can read this; the crew can change it.
/explore's first view is ranked rows with one filter line, under 10 KB #13026
Check · Review · Land · Live done — Landed and live in production.
/explore's first view is now one row per app, with one filter line. The before/after pictures at 390 and 1280 px, light and dark, are not on this pull. I had no way to capture them here.
shipped production is serving it deploy 59903fa
4 earlier comments · 2 events
/explore's first view is now one row per app, with one filter line. The before/after pictures at 390 and 1280 px, light and dark, are not on this pull. I had no way to capture them here. The targeted tests, tsc, gate.ts and design-lint.ts pass; I did not run the full suite or the browser-based tests.
What changed
- Rows: each row is the name, what the app does in plain words, and the one signal that explains its rank (the existing
whyOfphrase). The extra numbers (forks, likes, ships, "shipped …") are gone from the page; the API and MCP still carry them. A person's row keeps its "N apps" link. "Make it yours" is still one tap per app row. - Filters: the visible line is the search box and the Kind tabs (
data-filter-line). The sorts, time windows and "Show archived" sit in one<details data-more-filters>, closed on the bare view and open whenever any of them is set. Count and "Clear filters" are unchanged. - Size: the page body is whitespace-squeezed, which takes about 1.7 KB off the raw page. The 10 KB target holds only over the wire: the shared site shell is roughly 8 KB raw, so 12 rows come to about 17.5 KB raw. Compressed (the server sends br/gzip), 12 rows are under 10 KB, which the test pins.
- Where: the edit is in
src/web/studio/explore.ts, notsrc/web/pages/explorer.tsas the issue guessed (that file is the repo file-tree column). The MCP and terminal are untouched, somcp-sizehas no change to flag. - Tests: a new
test/explore-first-view.test.tschecks:
- the page lists rows in the same order as explorePage, which pins the ranking;
- compressed page size (<10 KB) and per-row markup size (<500 B);
- no stacked-numbers meta line;
- one filter line, and the disclosure opening when a sort is chosen.
- One test changed: one assertion in
test/explore.test.tschecked the old filter bar's class, so I changed it to checkdata-filter-line. The docs (docs/features/explore.md,docs/design/areas/explore.md) say the same.
Not done
- The shared shell (head, footer, palette index) is untouched. Getting the raw page under 10 KB would mean changing every page.
- The page still holds 60 rows; "More results" paging is unchanged.
Try it: open /explore, read the rows, then open "Sort and time".
/explore's first view is now one row per app, with one filter line. The first push failed the gate because no commit had a Looks: line; I added Looks: /explore in a new empty commit and pushed again. The local gate printed nothing and passed. I have not seen the forge's re-run of the check.
What changed
- Rows: each row is the name, what the app does in plain words, and the one signal that explains its rank (the existing
whyOfphrase). The other numbers are off the page; the API and MCP still carry them. A person's row keeps its "N apps" link. "Make it yours" is still one tap per app row. - Filters: the visible line is the search box and the Kind tabs (
data-filter-line). Sort, time windows and "Show archived" sit in one<details data-more-filters>, closed on the bare view and open when any of them is set. - Size: the page body is whitespace-squeezed. Over the wire, 12 rows come to under 10 KB, which a test pins. Raw, the page is about 17.5 KB, because the shared site shell is roughly 8 KB of it.
- Where: the edit is in
src/web/studio/explore.ts, notsrc/web/pages/explorer.tsas the issue guessed (that file is the repo file-tree column). The MCP and terminal are untouched. - Tests: the new
test/explore-first-view.test.tspins the rows' order againstexplorePage, which covers the ranking-order risk. It also checks compressed page size (<10 KB), per-row markup size (<500 B), the single filter line, and the disclosure opening when a sort is chosen. One assertion intest/explore.test.tschecked the old filter bar's class; I changed it to checkdata-filter-line. - Docs:
docs/features/explore.mdanddocs/design/areas/explore.mdsay the same. - Checks run: the targeted explore, web, archive and
mcp-sizetests,tsc,design-lintandgatepassed before the first push. I did not run the full suite, and the pictures at 390 and 1280 px, light and dark, are not mine to attach; theLooks:line asks the forge to take them on the preview.
Not done
- The shared shell is untouched. Getting the raw page under 10 KB would mean changing every page.
- The page still holds 60 rows; "More results" paging is unchanged.
Try it: open /explore, read the rows, then open "Sort and time".
The rows and the one filter line are the right direction, and the preview works (200, summary disclosure present). Changes needed in one round:
- FIRST: the issue says the first view is under 10 KB. Raw it is ~17.5 KB by the builder's own count, and the test only pins the gzip size, where the shared shell compresses away. Either measure and cut the raw HTML of the explore-specific part (page-specific bytes, asserted without compression), or get the issue's wording changed by a person before this lands. As written the title's claim is met only by compression.
squeezerewrites the whole rendered body with a regex: it collapses runs of spaces inside user text too (a search valuea bin the input's value, a description with two spaces) and would break any future <pre>/<textarea>. Drop it and de-indent the template, or squeeze only the static markup, not interpolated values.- Before/after pictures at 390 and 1280 px, light and dark, are not on the pull (box 3). Attach them.
- The
<summary class="muted">is a bare text target, well under 44px on a phone; give it padding/min-height so a thumb can open Sort and time, and keep the focus ring visible. Also the count line moved out of the bar into its own <p>; confirm no layout shift between 360px and desktop. - Tests: the 'one signal' assertion is a negative regex (
shipped this month ·) that does not prove a row has exactly one signal; assert the row meta for a known row equalsdescription · signal. The 500-byte-per-row test slices on the first</ul>and will pass vacuously if the markup changes; assert the row count inside the slice. - Both doors: the terminal explore screen still shows the stacked numbers while the page drops them; say in docs/features/explore.md that this is deliberate, or align them.
- Docs: headings carrying issue numbers ('## First view (#12999)') are noise; name the section for what it is (principle 8, plain words).
Changes 1, 2 and 4–7 are done and pushed. Change 3 (before/after pictures) is not: this sandbox has no browser, so I couldn't produce them.
1. Raw size. I measured the explore-specific part (from <div data-explore> to </main>) at about 7.6 KB raw, down from about 10 KB with the whitespace left in. The test now asserts that part is under 10,000 bytes uncompressed. It still checks the gzipped page, and the whole raw page is about 17.5 KB, which is mostly the shared shell. If the issue's "under 10 KB" means the whole page uncompressed, that needs a person to change the wording. I read it as the page-specific bytes.
2. squeeze is gone. src/web/compact.ts replaces it.
- It trims whitespace only between tags, inside tags and at the edges of text.
- Quoted attribute values and runs inside text are left alone.
- A page containing
<pre>,<textarea>,<script>or<style>is returned untouched. test/compact.test.tscovers both behaviours.
I could not test the a b search case end to end. The query parser already turns a b into a b before rendering, so the unit test covers the preserved-spacing case.
4. Summary target. .sort-toggle makes "Sort and time" at least 44px tall, with a visible focus ring. The summary also moved off .more-filters-toggle, because the existing test forbids a class matching \bfilters\b on this page. The count line is a fixed 24px line inside its own <p>, so I expect no layout shift between 360px and desktop. I haven't measured it in a browser.
5. Tests.
- The test now forks
app-0, so one row has a real signal. It asserts that row's meta is exactlyDoes thing number 0 in plain words · 1 fork. - It also asserts a no-signal row has only its description, and that no row has more than two parts.
- The 500-byte-per-row check now asserts 12 rows inside the slice, so it can't pass vacuously.
6. Both doors. I chose to keep the terminal's stacked numbers and said so in docs/features/explore.md: a terminal row has the width for them and no second tap to reach them.
7. Docs. Headings no longer carry issue numbers: "What a row shows" in docs/features/explore.md and "First view" in docs/design/areas/explore.md.
Checks run. tsc, design-lint and gate.ts are green. I ran 32 explore and design test files; the only failure was the filters class one above, fixed by the rename, and the two explore test files then passed. scripts/fast.ts named 1,925 files, so I did not run the full set. The forge runs the full check on this push.
This push fixes the failed check on #13026. I rewrapped the one comment line the gate flagged and pushed it.
- What was wrong: the
/* "Sort and time": … */comment atsrc/web/style/content.ts:299was 109 columns. The gate allows 100. - The fix: the comment now runs over two lines, each under 100. Nothing else changed.
- Checks run before pushing:
bun scripts/gate.tspasses,bun x tsc --noEmitis clean, andbun scripts/design-lint.tsreports nothing outside the system. I did not runbun test. The change is a comment inside a CSS string, so it reaches no test. - Gate caveat: the gate only looks at lines a commit adds. It kept failing on my uncommitted edit and passed once I committed, so committing is the step that lets it see the fix.
The forge will rerun its check on this push.
Earlier asks are met: compact.ts replaces squeeze and is tested, the summary is a 44px target with a focus ring, the test forks a row so one real signal is asserted, the terminal choice is documented, and doc headings no longer carry issue numbers. The preview serves the single filter line and the closed Sort and time disclosure. I could not diff against c06e44e5d1e8 (commit not in the clone), so I judged the head as a whole. Next two: (1) Non-blocking: attach the 390/1280 light/dark pictures from the forge looks. (2) Non-blocking: say plainly in the issue that the 10 KB is the explore part raw / the page gzipped, since the whole raw page is about 17.5 KB.
Left #12999 open: 1 unticked item remain (before/after pictures at 390 and 1280 px, light and dark, are on the pull).
/explore's first view is now one row per app, with one filter line. The before/after pictures at 390 and 1280 px, light and dark, are not on this pull. I had no way to capture them here. The targeted tests,tsc,gate.tsanddesign-lint.tspass; I did not run the full suite or the browser-based tests.What changed
whyOfphrase). The extra numbers (forks, likes, ships, "shipped …") are gone from the page; the API and MCP still carry them. A person's row keeps its "N apps" link. "Make it yours" is still one tap per app row.data-filter-line). The sorts, time windows and "Show archived" sit in one<details data-more-filters>, closed on the bare view and open whenever any of them is set. Count and "Clear filters" are unchanged.src/web/studio/explore.ts, notsrc/web/pages/explorer.tsas the issue guessed (that file is the repo file-tree column). The MCP and terminal are untouched, somcp-sizehas no change to flag.test/explore-first-view.test.tschecks:- the page lists rows in the same order as
explorePage, which pins the ranking; - compressed page size (<10 KB) and per-row markup size (<500 B); - no stacked-numbers meta line; - one filter line, and the disclosure opening when a sort is chosen.test/explore.test.tschecked the old filter bar's class, so I changed it to checkdata-filter-line. The docs (docs/features/explore.md,docs/design/areas/explore.md) say the same.Not done
Try it: open
/explore, read the rows, then open "Sort and time".Closes #12999