[olliethedev]

Summary

- Related-item reads can stop at adapter-default scan limits and fetch each item separately.
+ Complete batched scans, unique totals, stable timestamp/ID ordering, and page-only content reads.
- CMS reads always return the editor envelope and raw JSON.
+ Opt-in content views and literal top-level field selections on existing reads and picker hooks.

Preserve default responses and authorization. Full, content, and projected variants have separate cache identities, equivalent field sets share an identity, and mutations refresh affected variants. Existing SSR/SSG paths retain serialized timestamps and matching hydration envelopes.

Closes #290
Closes #291
Closes #292
Closes #293
Closes #294

Evidence

• Before: A default-limited adapter regression returned zero of 125 matching source records after unrelated relation rows; different relation page sizes shared a cache entry.
  After: Direct and inverse reads return all successive unique pages with deterministic ordering, while full payload retrieval stays limited to requested page IDs.

• Before: Content views and field selections retained editor data or shared full-response caches.
  After: Operation/HTTP/getter/resource tests verify exact reduced envelopes; literal punctuation, Unicode and declared-dot names; invalid selections; optional/null/nested values; authorization and race checks; projected picker search/preloads; mutation freshness; and prefetch/type parity.

• Final head: build, typecheck, lint, Knip and docs production build pass. 1,191 Stack + 198 CLI unit tests, 115 CMS/relation/SSG browser tests across Next.js, React Router and TanStack (SSG on Next.js), 7 release-gate tests, and 15 product-proof tests pass.

• Regenerated registry output is clean. A fresh packed registry consumer installs all nine plugin items and builds successfully. Built-export type probes verify default/content/static-field and dynamic-field results.

• Standards and spec reviews have no remaining material findings. Adapter coverage uses memory plus a default-limited/counting fixture; named production database adapters were not run.

• Media concurrency regression waits for committed user A and suspends only speculative user B. It verifies A refreshes and B stays uncached, removing a scheduler-dependent cache-eviction assertion. Twenty repeated Media-suite runs pass, and an injected render-phase ownership fault still makes the test fail; the CI repair is confined to this test.

• Full CI/CD verification passes on 6156e2f9: all five workflows, all expected jobs, and both Vercel deployments. The repository release gate passed after two matching evidence collections.

Merge Danger

Door: two-way

No schema migration or new dependency. Reduced output is opt-in; existing calls retain their envelopes and access checks.

Blast Radius: CMS

Relation retrieval, read typing, query cache identities and prefetch hydration. Output projection reduces response payloads; stored JSON may still be fetched and parsed, and relation ID/metadata scans grow with cardinality.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
better-stack-docs	Ready	Preview	Oct 28, 2026 8:04pm UTC
better-stack-playground	Ready	Preview	Oct 28, 2026 8:04pm UTC

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-28T19:22:53.541439Z	b85aaa7	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.