The terminal's Usage screen opens the fleet graph with `f`: your repositories' spend and runs per hour over 24 h and per day over 14 d #8059

f on the terminal's Usage screen now opens a fleet graph with a sparkline per repository, as /fleet/graph shows on the web. bun x tsc --noEmit, bun scripts/gate.ts, bun scripts/design-lint.ts and the new test plus the two existing Usage tests pass. I didn't run the full suite; the forge's check does that.

• Graph: src/tui/screens/fleetgraph.ts reads readFleetGraph once when the screen opens. It draws a Runs block and a Spend block. Each row is the repository name, a sparkline scaled to that row's own peak, and the row's total (N runs, or dollars). Past FLEET_GRAPH_MAX_SERIES the tail shows as "N others". w toggles 24 h and 14 d, and an empty window reads "Nothing spent in this window."
• Opening it: f is bound in src/tui/screens/usage.ts and listed in that screen's keys.
• Visibility: the action's rules decide, so someone who may not see the spend gets an empty result.
• Narrow terminals: names are truncated and the graph keeps at least 10 columns. If the window has more buckets than the graph has room for, adjacent buckets are summed into one column.
• Tests: test/tui-fleetgraph.test.ts seeds ledger lines in two repositories and checks both names and their totals against readFleetGraph. It also checks that w switches the window, that a second user sees none of the first user's spend, and that a 44-column terminal still draws sparklines.

Choices to review:

• No f on a repository's Usage: the graph is forge-wide, so f is hidden and does nothing there. The "empty ledger" sentence is not tested through a fresh empty database. The test reaches it through a user who may see no spend, and the code path is the same.
• Admin-only entry: plain Usage is admin-only, and admins can write every repository. So on this screen the "may not see" case can only be reached from a repository's Usage, where f is hidden. The visibility rule is therefore tested against readFleetGraph directly.

To try it, open Usage from the palette (Ctrl+K, then "Usage"), press f, then w.

Closes #8048

f on the terminal's Usage screen now opens a fleet graph with a sparkline per repository, as /fleet/graph shows on the web. bun x tsc --noEmit, bun scripts/gate.ts, bun scripts/design-lint.ts and the new test plus the two existing Usage tests pass. I didn't run the full suite; the forge's check does that.

• Graph: src/tui/screens/fleetgraph.ts reads readFleetGraph once when the screen opens. It draws a Runs block and a Spend block. Each row is the repository name, a sparkline scaled to that row's own peak, and the row's total (N runs, or dollars). Past FLEET_GRAPH_MAX_SERIES the tail shows as "N others". w toggles 24 h and 14 d, and an empty window reads "Nothing spent in this window."
• Opening it: f is bound in src/tui/screens/usage.ts and listed in that screen's keys.
• Visibility: the action's rules decide, so someone who may not see the spend gets an empty result.
• Narrow terminals: names are truncated and the graph keeps at least 10 columns. If the window has more buckets than the graph has room for, adjacent buckets are summed into one column.
• Tests: test/tui-fleetgraph.test.ts seeds ledger lines in two repositories and checks both names and their totals against readFleetGraph. It also checks that w switches the window, that a second user sees none of the first user's spend, and that a 44-column terminal still draws sparklines.

Choices to review:

• No f on a repository's Usage: the graph is forge-wide, so f is hidden and does nothing there. The "empty ledger" sentence is not tested through a fresh empty database. The test reaches it through a user who may see no spend, and the code path is the same.
• Admin-only entry: plain Usage is admin-only, and admins can write every repository. So on this screen the "may not see" case can only be reached from a repository's Usage, where f is hidden. The visibility rule is therefore tested against readFleetGraph directly.

To try it, open Usage from the palette (Ctrl+K, then "Usage"), press f, then w.

f on the terminal's Usage screen opens a fleet graph: one sparkline row per repository for runs and for spend, over 24 h by hour or 14 d by day, with w toggling. This push addresses all four review comments. test/tui-fleetgraph.test.ts (7 tests), bun x tsc --noEmit, bun scripts/gate.ts and bun scripts/design-lint.ts pass. I did not run the full suite; the forge's check does that.

• Vacuous visibility test: I dropped the bob assertions, since bob can't reach Usage. The test now checks readFleetGraph for bob directly: no repositories and empty spend totals. A new test reaches "Nothing spent in this window." through a fresh forge with an empty ledger.
• Narrow terminal: the 44-column test now asserts every line is at most 44 columns, sparkline rows are drawn, and a long repository name is cut with …. Names are also capped at 32 columns, and the totals column is right-aligned.
• Short terminal: rows past the body height are replaced by a final "+N more rows (make the terminal taller)" line, and a test covers it.
• f on a repository's Usage: a new test opens "Usage of a repository", presses f, and checks the screen stays on Usage with no fleet graph. The README now documents f, w, the "N others" folding and the overflow line.
• Extra tests: I also added one for "N others" past eight series.
• Singular run: the total now reads "1 run" instead of "1 runs".

I did not add scrolling: the overflow line says how many rows were cut, which keeps the screen small.

To try it, open Usage from the palette (Ctrl+K, then "Usage"), press f, then w.

I made no code change for the three tui-notes failures, because I couldn't reproduce them. I merged current main into the branch and pushed it, so the forge's check will re-run on that.

• Reproduction: test/tui-notes.test.ts passes 4 of 4 on two runs here. After the merge it also passes together with test/tui-fleetgraph.test.ts (11 of 11).
• Likely cause: I think these were timing failures under load. They never touch Usage or the fleet graph, and the check log showed shard 12 killed after 19 s, which suggests the machine was busy.
• Untested: I can't say for sure it was load, and I haven't run the full suite here. If the check fails the same way again, the failure is real and I'll look at it properly.
• Merge: the merge was clean, bun x tsc --noEmit is clean, and the pull's own changes are the same as before.