The home count and the health answer count the same forge needs as the needs stack #12356

An admin's forge needs now come from one function, forgeNeedList in src/actions/health.ts. It lists the rollout, each platform incident, the vitals' needs and the week's flakes, as { cause, line, href }, from one readHealth per request. forgeNeeds (the health answer and the board's forge strip) returns just its sentences. forgeNeedsFor returns the list itself, and everything that builds the stack reads it: askStackFor, askCountFor (the home count), asksAnswer(...).forge (GET /api/v1/asks), needsStack (the MCP forge tool and the terminal) and the forge line on /asks. Before this, forgeNeedsFor counted only forgeVitals(app).degraded, so the home could say "All fine" while the health answer listed a held rollout or an incident.

What to look at:

• forgeCard (asks_page.ts) and needsStack now take a ForgeNeed and print its line. A need that isn't a vitals cause links to /api/v1/health.
• In needsStack, a forge need's why reads "the forge itself needs looking at", since it is no longer always a degraded vital.
• test/asks-forge-alert.test.ts: only its fixture changed, to the new shape. What it asserts is unchanged.
• The comment in asksAnswer claimed waiting.count includes the forge's needs; it doesn't (that count comes from reviewSet). The comment is corrected and the count is unchanged.

Not done: review_set.ts is unchanged. Its count stays the asks only, as its header says. The test checks that a non-admin is told nothing, but not that they cost no health read. The code returns before readHealth runs, but there is no read counter to assert on.

New test test/forge-needs-one-list.test.ts writes a guard status.json with a held incident and a db:replica incident. It checks that forgeNeeds, asksAnswer(...).forge, forgeNeedsFor, needsStack's forge entries, askCountFor and the home's needs section all give the same sentences and the same number. It fails on main and passes here. A second test checks that a non-admin gets none.

Checks: bun scripts/gate.ts, bun scripts/design-lint.ts and bun x tsc --noEmit pass. scripts/fast.ts named 2123 files, too broad to run, so I picked the 47 test files that touch the health answer, asks, the needs stack or the count: 423 pass, 1 skipped, 0 fail. docs/features/forge-needs.md gets a paragraph on the one list.

Try it: as a site admin, during a held rollout or a platform incident, open / and /asks. The count and the forge line name the same needs as GET /api/v1/health.

🤖 Generated with Claude Code

The home count on / and GET /api/v1/health now count the forge's needs the same way the needs-you stack does. Both use the same ForgeNeed list as the board's forge strip, so during a held rollout or a platform incident a site admin sees the same count and the same forge line on /, on /asks and in the health answer.

The latest commit fixes the gate failure from the last push: two comment lines were longer than 100 columns, in src/actions/health.ts (the ForgeNeed doc comment) and src/web/studio/asks.ts (the waiting figures comment). They are rewrapped with no code change.

Checks run locally:

• bun scripts/gate.ts and bun x tsc --noEmit both pass.
• bun scripts/fast.ts named 2,123 test files for this diff, well over its 5-minute budget. Since the fix only touches comments, I ran just the asks and health tests: asks-counts, asks-forge-alert, asks, forge-needs-one-list, forge-needs-one-stack and health-broken. All 27 tests pass.
• The forge's full check covers the rest.

To try it: sign in as a site admin on the preview while a rollout is held or an incident is open, then compare the count and forge line on / and /asks with the needs in GET /api/v1/health.

🤖 Generated with Claude Code

The home count, the /asks forge line and GET /api/v1/health now all read one list of what the forge needs, so they can't disagree.

• One list: forgeNeedList in src/actions/health.ts holds the forge's needs: a held rollout, open platform incidents, vitals and flaky tests. forgeNeeds turns it into sentences.
• Who reads it: forgeNeedsFor, askStackFor, asksAnswer, needsStack and the forge line on the asks page (src/web/studio/asks.ts, src/web/studio/asks_page.ts) all use that list.
• Non-admins: someone who isn't a site admin gets no forge needs and never triggers a health read.

test/forge-needs-one-list.test.ts checks that every surface agrees, and test/asks-forge-alert.test.ts is updated to the shared wording.

The last commit is empty and only adds a Looks: / line, which the before/after looks gate requires. / is where the count shows, and /asks isn't in app.looks, so the forge's light and dark pictures are of the home page. I didn't add /asks to app.looks.

The gate and tsc --noEmit pass, and 44 tests across the 10 asks, health and forge-needs test files pass. scripts/fast.ts said the diff was too broad to name tests, so I picked those files myself; the full suite runs in the forge check.

To try it: as a site admin during a held rollout or a platform incident, open / and /asks. The count and the forge line should name the same needs as GET /api/v1/health.

🤖 Generated with Claude Code

The home count, the /asks forge line and GET /api/v1/health now all read one list of what the forge needs, so they can't disagree.

• One list: forgeNeedList in src/actions/health.ts holds the forge's needs: a held rollout, open platform incidents, vitals, flaky tests and a broken cost-ledger chain. forgeNeeds turns it into sentences.
• Who reads it: forgeNeedsFor, askStackFor, asksAnswer, needsStack and the forge line on the asks page (src/web/studio/asks.ts, src/web/studio/asks_page.ts) all use that list.
• Non-admins: someone who isn't a site admin gets no forge needs and never triggers a health read.

Merge with main: main (#12281) had added the ledger need to the old sentence list. In the merge it becomes an entry in forgeNeedList with cause "ledger", so the home count and /asks now count it too, not only the health answer.

test/forge-needs-one-list.test.ts checks that every surface agrees, and test/asks-forge-alert.test.ts is updated to the shared wording. An empty commit adds a Looks: / line: / is where the count shows, and /asks isn't in app.looks, so the light and dark pictures are of the home page.

After the merge the gate and tsc --noEmit pass, and 51 tests across 12 files pass, including test/ledger-health.test.ts. scripts/fast.ts said the diff was too broad to name tests, so I picked those files myself; the full suite runs in the forge check.

To try it: as a site admin during a held rollout, a platform incident or a broken ledger chain, open / and /asks. The count and the forge line should name the same needs as GET /api/v1/health.

🤖 Generated with Claude Code

The home count, the /asks forge line and GET /api/v1/health all read one list of what the forge needs, so they can't disagree. This push adds tests proving the read costs: none for a non-admin, one for an admin.

• One list: forgeNeedList in src/actions/health.ts holds the forge's needs: a held rollout, open platform incidents, vitals, flaky tests and a broken cost-ledger chain (from #12281). forgeNeeds turns it into sentences.
• Who reads it: forgeNeedsFor, askStackFor, asksAnswer, needsStack and the forge line on the asks page all use that list.
• Read counter: a new healthReads() counts how often readHealth runs. test/forge-needs-one-list.test.ts checks that forgeNeeds, forgeNeedsFor, asksAnswer, askCountFor and needsStack cost a non-admin zero reads and an admin exactly one each. That test fails if the admin check moves after the health read.
• Stable flake ids: a flake need's cause is now flake:<repo>:<test> instead of its position, via flakeForgeNeeds and two small helpers in src/checks/flake.ts. Its id in the stack no longer changes when the week's flakes reorder, and a test that fails with the old position keys covers it.

Not changed: incident, flake and ledger needs still link to /api/v1/health, which is raw JSON. That matches the existing Health button, and giving it a readable page is its own change.

The commits carry Looks: / because the count shows on the home page and /asks isn't in app.looks.

The gate and tsc --noEmit pass, and 67 tests across 13 files pass (asks, health, forge-needs, ledger-health and flake). The full suite runs in the forge check.

To try it: as a site admin during a held rollout, a platform incident or a broken ledger chain, open / and /asks. The count and the forge line should name the same needs as GET /api/v1/health.

🤖 Generated with Claude Code

The home's stack word, the health answer and the board's forge strip now all read their count and lines from one list: forgeNeeds in src/actions/health.ts. Each ForgeNeed carries a cause (rollout, an incident key, a vitals cause, flake:<repo>:<test> or ledger), a sentence and a link, so the three surfaces can't disagree on how many things the forge needs. A visitor who isn't a site admin costs no health read at all; a site admin costs exactly one. Flake needs are keyed by repository and test name.

test/forge-needs-one-list.test.ts checks that the count and lines match across the three surfaces and that the health read happens only for an admin.

The last commit only rewraps two comments in src/actions/health.ts that were over 100 columns after merging main, which the gate rejected. The gate now passes, as does tsc --noEmit and the tests in forge-needs-one-list, review-rules, health, health-broken and ledger-health. scripts/fast.ts said the diff was too broad to pick tests for, so I chose those files myself; the full suite runs on the forge's check.

🤖 Generated with Claude Code

The home count, /asks, GET /api/v1/asks, the needs stack and the health answer now all count the same forge needs (#12269). Before this, a held rollout, a platform incident or a frequently flaky test appeared only on the board's forge strip.

• One list: forgeNeedList in src/actions/health.ts is the one source. It returns plain { cause, line, href } records from one health read. forgeNeeds is now just that list as sentences, and it is what the health answer and the forge strip print. The home count, /asks and the needs stack read the same list.
• Stable ids for flakes: each flake need is keyed flake:<repo>:<test>, so its id in the stack stays the same when the week's flakes are reordered. src/checks/flake.ts splits flakeNeeds into flakeNeedRows and flakeNeedLine so the health module can reuse both.
• Read cost: someone who is not a site admin gets no forge needs and costs no health read on any page. A site admin's request costs exactly one. test/forge-needs-one-list.test.ts checks this by spying on readHealth.
• Wording change: a forge need in the stack now gives its reason as "the forge itself needs looking at" instead of "the forge itself is degraded", because a held rollout or a flaky test is not degradation.

Since the last check, the exported healthReads counter is gone. Only the test used it, which is why the dead-code and hollow-work tests failed. The spy in the test takes its place.

test/boundary-tenancy-cell.test.ts is also red on main and this pull does not touch it.

Here, the gate, bun x tsc --noEmit, test/forge-needs-one-list.test.ts, test/hollow.test.ts, test/unused-exports.test.ts, test/asks-forge-alert.test.ts and test/check-flake.test.ts all pass.

🤖 Generated with Claude Code