Authorized: maintainer 2026-08-23, live PM chat — options ①②③ approved verbatim as 「1+2+3」. Filed by the skills seat (session 5213b871-5164-5bc3-8874-28b336bbcd40), self-triaged under the skills-lane exception. This card is option ③.
Facts (measured + official)
- Measured:
list_issues with perPage=100 full fields = 107 points for a 34-card lane. GraphQL point cost ≈ requested nodes / 100 — perPage is the lever, call count is not. - Official (docs re-verified 2026-08-23): GraphQL secondary limits exist BESIDE the hourly pool — max 2,000 points/minute on the GraphQL endpoint, ≤100 concurrent requests, and the guidance to pause ~1 second between mutative requests (mutations count 5× for the secondary calculation). A burst of same-second label writes can trip the minute-level wall while the hourly pool is still green.
Change
references/platform-readings.md API-quota section, two additions paid by compression: (a) one row stating the nodes/100 cost rule with the measured 107-point example and the instruction to size perPage to the expected population, not to 100 by habit; (b) one row stating the minute-level secondary limits and the ~1s mutation pacing rule for batch label/comment sweeps (the dual-carrier clears and batch retriage sweeps are the callers that bunch writes).- The since-window incremental inventory (already landed for the triage seat) is the same principle at the query level — the row may point at it rather than restate it.
Tier: references-only ⇒ opus build + contract-review-tier acceptance.
Ceiling: platform-readings 134 headroom 0 — payment by in-place compression; coordinate with option ① card (same section of the same file — fold-or-serial answer required at claim time; the five-gate family fold test likely PASSES for ①+③ since both are quota-section rewrites of the same defect family, same file, both ruled in one stroke).
Serial constraint
Queues BEHIND the #11086 restructuring PRs (same hot-file chain). Related: option ① card (REST-first reads) and option ② card (dev usage shaping), filed in the same stroke.
Generated by Claude Code
Authorized: maintainer 2026-08-23, live PM chat — options ①②③ approved verbatim as 「1+2+3」. Filed by the skills seat (session
5213b871-5164-5bc3-8874-28b336bbcd40), self-triaged under the skills-lane exception. This card is option ③.Facts (measured + official)
list_issueswith perPage=100 full fields = 107 points for a 34-card lane. GraphQL point cost ≈ requested nodes / 100 — perPage is the lever, call count is not.Change
references/platform-readings.mdAPI-quota section, two additions paid by compression: (a) one row stating the nodes/100 cost rule with the measured 107-point example and the instruction to size perPage to the expected population, not to 100 by habit; (b) one row stating the minute-level secondary limits and the ~1s mutation pacing rule for batch label/comment sweeps (the dual-carrier clears and batch retriage sweeps are the callers that bunch writes).Tier: references-only ⇒ opus build + contract-review-tier acceptance.
Ceiling: platform-readings 134 headroom 0 — payment by in-place compression; coordinate with option ① card (same section of the same file — fold-or-serial answer required at claim time; the five-gate family fold test likely PASSES for ①+③ since both are quota-section rewrites of the same defect family, same file, both ruled in one stroke).
Serial constraint
Queues BEHIND the #11086 restructuring PRs (same hot-file chain). Related: option ① card (REST-first reads) and option ② card (dev usage shaping), filed in the same stroke.
Generated by Claude Code