The search, map, flow and system pages each carry a first-view attention budget that test/attention.test.ts holds #10691

/search, /search?q=forge, /map, /flow and /system now each have a first-view entry in docs/design/attention-budget.json, and test/attention.test.ts visits all five signed in as an admin. Both tests in test/attention.test.ts pass, and tsc and design-lint are clean. I ran only that file; the forge runs the full check on push.

Measured words / blocks / colours, the same at 390 and 1280 px, with no headroom added:

• Failure proved locally: lowering /flow's 390 px blocks entry by one made the test fail with flow @390: blocks 9 > budget 8. I restored the entry afterwards.
• Sampled pages: /flow and /system are measured on the test forge's empty state, with nothing seeded or waited for. Both entries say so in a note.
• Notes in the file: the two search entries also carry a note. The width-key check ignores note, and a FORGE_ATTENTION_WRITE=1 rewrite keeps existing notes.
• /system is large: its 820 words and 277 blocks are today's number, not a target.
• Docs: docs/features/attention.md lists the new pages. Its baseline row for settings said 137 words, but the budget file and the test already had 140, so I corrected it.

Closes #10687

The check failed because /system draws the machine it runs on, so my first measurement was only true on my host. I've fixed that and pushed. test/attention.test.ts passes twice in a row here and tsc is clean. I haven't seen the forge's check run on the new push yet.

What differed: /system shows the host's cores, disks, processes and cgroup state. The two cell-cap sentences also change length depending on whether the kernel enforces the caps. I measured 820 words / 277 blocks locally, and the check's host counted something else.

What changed:

• The test now gives /system a fixed empty machine reading through useLive and replaces the two cell-cap sentences with one fixed line. The budget no longer depends on the host.
• The /system entry is now 212 words / 65 blocks / 0 colours at both widths, down from 820 / 277. A live /system is larger.
• The entry's note and docs/features/attention.md say how it is measured. The other entries are unchanged.

I made no change to the app's pages. Only the test, the budget file and the docs changed.

Both review points are addressed: the blank line is in, and each of the five budgets now has a local proof that growing it fails the test. I did not re-run tsc or the design lint for this round, because it only changed one blank line in docs/features/attention.md.

What changed since the last push

• docs/features/attention.md now has the blank line after the /flow and /system paragraph, so ## App page, calmed renders as a heading.
• No code changed.

Growing one entry by one text block fails the test (box 5). For each entry I temporarily added a 5-word <p> before the <h1> in test/attention.test.ts, then reverted it. each page stays inside its first-view budget failed every time, at both 390 and 1280 px. The failures read "seen > budget":

With the temporary edit removed, test/attention.test.ts passes (2 pass, 0 fail). The temporary edit is not in the commit.

Test file time (box 6). bun test test/attention.test.ts takes about 1.7 s. The budget test itself runs in roughly 30 ms.

Not changed. /system is still measured on an injected empty machine reading, so a regression in the live page is outside the budget. The test note and the docs say this plainly, as you accepted.