A pure read estimates when an approved pull lands, from its place in the queue and the recent landing pace #6561

Adds src/flow/landing_eta.ts, a pure read for #5930 that estimates how long an approved pull will wait before it lands. Closes #6548.

• landingEta({ ahead, landedLastHour, medianGateMinutes }) returns minutes, worked out as (ahead + 1) × per-landing time:

- When something landed in the last hour, the per-landing time is 60 / landedLastHour. The real pace wins over the gate time because it already counts batching, red gates and idle gaps. - When nothing landed, it falls back to medianGateMinutes (gates run one after another). - With neither, it returns null. - It reads no clock. Negative, NaN or infinite counts are read as zero, so the result is never a negative wait.

• etaSays(minutes) turns the number into words:

- Under an hour it rounds to the nearest 5 min, never less than 5 ("lands in about 40 min"). - From 57.5 min on it rounds to the nearest hour ("lands in about 3 h"). The issue wrote this one as "about 3 h"; I kept the "lands in" prefix so both forms read the same. - null gives "no estimate: nothing landed in the last hour".

• test/flow-landing-eta.test.ts checks results against hand calculations: first in line (60/6 = 10 min), deep in line (12 × 60/4 = 180 min), a stalled queue that falls back to gates (3 × 14 = 42 min), a stalled queue with no gate time (null), nonsense inputs, the rounding edges (42 → 40, 42.5 → 45, 57.5 → 1 h, 209 → 3 h, 210 → 4 h), and both functions together.

The issue limited this to the two files, so the read isn't connected to anything yet: no pull page, terminal or MCP output shows it, and it doesn't read the queue position or landing count from the database. That belongs to the next child of #5930.

bun x tsc --noEmit is clean and the new test passes (9 tests). I didn't run the rest of the suite; the forge's check does that on push.

🤖 Generated with Claude Code

Closes #6548

Adds src/flow/landing_eta.ts, a pure read for #5930 that estimates how long an approved pull will wait before it lands. Closes #6548.

• landingEta({ ahead, landedLastHour, medianGateMinutes }) returns minutes, worked out as (ahead + 1) × per-landing time:

- When something landed in the last hour, the per-landing time is 60 / landedLastHour. The real pace wins over the gate time because it already counts batching, red gates and idle gaps. - When nothing landed, it falls back to medianGateMinutes (gates run one after another). - With neither, it returns null. - It reads no clock. Negative, NaN or infinite counts are read as zero, so the result is never a negative wait.

• etaSays(minutes) turns the number into words:

- Under an hour it rounds to the nearest 5 min, never less than 5 ("lands in about 40 min"). - From 57.5 min on it rounds to the nearest hour ("lands in about 3 h"). The issue wrote this one as "about 3 h"; I kept the "lands in" prefix so both forms read the same. - null gives "no estimate: nothing landed in the last hour".

• test/flow-landing-eta.test.ts checks results against hand calculations: first in line (60/6 = 10 min), deep in line (12 × 60/4 = 180 min), a stalled queue that falls back to gates (3 × 14 = 42 min), a stalled queue with no gate time (null), nonsense inputs, the rounding edges (42 → 40, 42.5 → 45, 57.5 → 1 h, 209 → 3 h, 210 → 4 h), and both functions together.

The issue limited this to the two files, so the read isn't connected to anything yet: no pull page, terminal or MCP output shows it, and it doesn't read the queue position or landing count from the database. That belongs to the next child of #5930.

bun x tsc --noEmit is clean and the new test passes (9 tests). I didn't run the rest of the suite; the forge's check does that on push.

🤖 Generated with Claude Code