A crew member's scorecard shows the last 30 days beside all time #3184

The crew scorecard now shows the last 30 days beside all time for pulls landed, review rounds and red checks. Before, every figure covered all time, so tuning an agent never showed any change.

• src/core/scorecard.ts: crewScorecards(db, { since?: Date }) counts only pulls opened (issue.created_at) at or after since. It is still one grouped query, with an added WHERE ?1 IS NULL OR issue.created_at >= ?1. created_at is stored as ISO-8601 UTC with milliseconds, so comparing the strings compares the times. Without since the result is the same as before. An invalid Date throws; it does not quietly widen the window to all time. An agent with no pulls in the window is left out, so its row shows dashes rather than zeros.
• src/web/pages/crew.ts (scorecard card only): the page runs the query twice, once for all time and once with since set 30 × 86 400 000 ms ago. Each of the three figures gets a "30 days" column and an "all time" column, with the window named under the heading. Reverts and time to land stay all time, and the card's caption says so. The header is still one row: the sticky-lead rule pins the first cell of every row, so a second header row would pin a figure instead of the agent's name. No CSS or new classes were added, so the design check that failed the previous attempt (#3178) stays at its baseline.
• test/crew-scorecard.test.ts: a new test seeds a second agent with two landed pulls from 2020 (three change requests, one failed check) and two recent pulls (one landed with a passing check, one open). The 30-day figures are 1 of 2 landed, 0.0 rounds and 0% red. All time is 3 of 4, 1.0 and 50%. It checks the same figures in the rendered row. The header assertion in the existing first test was updated because the headings now carry their window; the test itself was not wrong.

The terminal's scorecard screen still shows all time only; I kept to the files the issue named. bun test passes for test/design-lint.test.ts and the crew-page suites (178 tests), and bun x tsc --noEmit is clean.

To try it, open /settings/crew as an admin. Pulls opened more than 30 days ago count only in the "all time" columns.

🤖 Generated with Claude Code

Closes #3168

The crew scorecard now shows the last 30 days beside all time for pulls landed, review rounds and red checks. Before, every figure covered all time, so tuning an agent never showed any change.

• src/core/scorecard.ts: crewScorecards(db, { since?: Date }) counts only pulls opened (issue.created_at) at or after since. It is still one grouped query, with an added WHERE ?1 IS NULL OR issue.created_at >= ?1. created_at is stored as ISO-8601 UTC with milliseconds, so comparing the strings compares the times. Without since the result is the same as before. An invalid Date throws; it does not quietly widen the window to all time. An agent with no pulls in the window is left out, so its row shows dashes rather than zeros.
• src/web/pages/crew.ts (scorecard card only): the page runs the query twice, once for all time and once with since set 30 × 86 400 000 ms ago. Each of the three figures gets a "30 days" column and an "all time" column, with the window named under the heading. Reverts and time to land stay all time, and the card's caption says so. The header is still one row: the sticky-lead rule pins the first cell of every row, so a second header row would pin a figure instead of the agent's name. No CSS or new classes were added, so the design check that failed the previous attempt (#3178) stays at its baseline.
• test/crew-scorecard.test.ts: a new test seeds a second agent with two landed pulls from 2020 (three change requests, one failed check) and two recent pulls (one landed with a passing check, one open). The 30-day figures are 1 of 2 landed, 0.0 rounds and 0% red. All time is 3 of 4, 1.0 and 50%. It checks the same figures in the rendered row. The header assertion in the existing first test was updated because the headings now carry their window; the test itself was not wrong.

The terminal's scorecard screen still shows all time only; I kept to the files the issue named. bun test passes for test/design-lint.test.ts and the crew-page suites (178 tests), and bun x tsc --noEmit is clean.

To try it, open /settings/crew as an admin. Pulls opened more than 30 days ago count only in the "all time" columns.

🤖 Generated with Claude Code

The crew scorecard now shows pulls landed, review rounds and red checks for the last 30 days beside all time, on Settings → Crew and in the terminal. Before, every figure covered all time, so tuning an agent never showed any change.

• src/core/scorecard.ts: crewScorecards(db, { since }) counts only pulls opened at or after since. It is still one grouped query, with an added WHERE ?1 IS NULL OR issue.created_at >= ?1; the stored ISO-8601 UTC strings compare in time order. Without since the result is unchanged. An invalid Date throws rather than quietly widening to all time. The 30-day window now lives here (RECENT_DAYS, recentWindow(now)), so the page and the terminal share one definition.
• Page (src/web/pages/crew.ts, scorecard card only): each of the three figures gets a "30 days" column and an "all time" column, with the window named under the heading. Reverts and time to land stay all time, and the caption says so. The header stays one row, because the sticky-lead rule pins the first cell of every row. No CSS or classes were added, so the design-lint baseline that failed #3178 holds. The long call near line 483 is wrapped.
• Terminal (src/tui/screens/scorecard.ts): the screen reads all-time and 30-day scorecards and shows each figure as 30d … · all …, in the page's order, e.g. landed 30d 1 of 2 · all 3 of 4.
• README: the scorecard paragraph now describes the two windows.
• Tests (test/crew-scorecard.test.ts):

- A new test seeds two landed pulls from 2020 and two recent ones. All time comes out as 3 of 4 landed, 1.0 rounds and 50% red; the last 30 days as 1 of 2, 0.0 and 0%. It checks the query and the page's cells, and shows that an agent with nothing in the window is left out rather than shown as zeros. - recentWindow is checked against a hand-computed date. - A new terminal test expects the same figures on the terminal screen. - The existing header and terminal assertions were changed to the new layout and wording; the tests were not wrong, the output changed on purpose.

The type check passes, as do the design-lint test and every suite that mentions the crew page or the scorecard (181 tests).

I have not looked at the 9-column table on a 360 px phone: the sandbox has no browser. The columns are narrow numbers with two-line headings inside the existing horizontal scroller, and the agent's name stays pinned, but someone should check it on the preview.

🤖 Generated with Claude Code