fix(search): bound retrieval and progressively read document evidence

[icecrasher321]

Summary

• Give live search SQL one shared deadline across connection acquisition, ranking, fallback, and hydration. Cancel expensive statements in PostgreSQL and expose incomplete retrieval explicitly; no usable evidence returns a retryable error.
• Return verbatim search previews and let Assistant read more by chunk and character position, with bounded pages, current authorization, secret projection, and citations. Remove the unreleased ordinal-offset reader contract and use the same precise continuations on the organization document page.
• Extend existing diagnostics to separate pool wait, SQL execution, document reads, and original versus returned text size. Extend the existing realistic harness rather than adding a separate reporting system.

Companion Assistant contract and telemetry change: https://github.com/simstudioai/mothership/pull/495. Deploy these changes together.

Type of Change

• Bug fix

Testing

• 1,085 search, authorization, provenance, reader, and route tests; includes protected query projection and stale continuation regressions.
• Real PostgreSQL/pgvector harness with 45,000 synthetic chunks: server cancellation, exhausted pool, partial and fully timed-out retrieval, permission revocation, concurrent search, and organization scope.
• Live local Go Assistant + Sim + configured model: search, targeted document read, correct facts, and citation. Synthetic search payload fell from 54 KB to 28 KB with the same 15 results.
• Full workspace typecheck, lint, all 46 audits, API validation, block-registry audit, docs manifest, and generated contract checks.

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Sep 11, 2026 11:03pm UTC

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge, with no accepted new findings or outstanding previous findings.

Summary

• Distinguishes complete, partial, and retryable timed-out retrieval.
• Cancels expensive PostgreSQL work and records pool, SQL, hydration, and document-read diagnostics.
• Returns projected, bounded search previews with citations and precise document continuations.
• Adds authorization, provenance, stale-position, cancellation, and realistic integration coverage.

Diagram

sequenceDiagram
    participant A as Assistant
    participant S as Search tool
    participant R as Retrieval
    participant D as Document reader
    participant DB as PostgreSQL

    A->>S: search_workspace(query)
    S->>S: Project protected query
    S->>R: Search with shared deadline
    R->>DB: Vector and keyword retrieval
    DB-->>R: Ranked authorized rows
    R-->>S: Complete or partial retrieval
    S-->>A: Bounded previews + chunk indexes + citations
    A->>D: read_document(chunk index, offset)
    D->>DB: Load currently authorized enabled chunks
    D-->>A: Bounded verbatim page + precise next position

Loading

_{Reviews (2) · Last reviewed commit: "fix(search): protect query output and va..."}

[icecrasher321]

@greptile

[icecrasher321]

@cubic-dev-ai review this PR

[cubic-dev-ai]

@cubic-dev-ai review this PR

@icecrasher321 I have started the AI code review. It will take a few minutes to complete.