perf(orchestration): avoid full history loads in runtime paths

[tarik02]

What Changed

• Added bounded operational projections that load only the latest turn item.
• Added targeted reads for runs, checkpoint scopes, run and node items, and item existence checks.
• Switched recovery, provider control, checkpointing, queue handling, subagent finalization, and thread management paths to bounded or targeted reads.
• Kept full-history reads for snapshots, legacy imports, forks, merges, and provider context handoffs.
• Validated historical turn items sequentially in batches of 256.

Why

Routine server operations loaded and schema-decoded every historical turn item. Memory and CPU usage therefore grew with the full thread history, even when a call only needed current runtime state.

This reduces server-side loading without changing client behavior, thread snapshots, pagination, or visible history.

Checklist

• This PR is small and focused
• I explained what changed and why

Note

Avoid full history loads in orchestration runtime paths by introducing operational projections

• Adds getOperationalProjection (latest turn item only) and getRecoveryProjection (pending/running/waiting items only) to ProjectionStoreV2, alongside targeted fetch methods getRun, getCheckpointScope, getRunTurnItems, getNodeTurnItems, and getPresentTurnItemIds.
• Replaces most getThreadProjection calls across orchestration services (Orchestrator.ts, ThreadManagementService.ts, ProviderTurnStartService.ts, etc.) with the appropriate scoped fetch; full history is only loaded when explicitly required (e.g. provider resume fallback, legacy import).
• ProjectionMaintenance verification now uses validateThreadProjection to decode turn item payloads in batches without loading full projections.
• Message ordering is preserved when turn items are not fully loaded by seeding ordinals from persisted message_ordinal rows.
• Behavioral Change: all orchestration decision-making paths (launch, interruption, continuation, title regeneration, checkpoint capture/rollback, session detach) now operate on bounded projections rather than full thread history.

^{Macroscope summarized 41fcc34.}

---

Note

Medium Risk
Touches core orchestration dispatch and projection loading across many services; incorrect bounded reads could change runtime behavior on long threads, though full-history paths are preserved for handoffs and imports.

Overview
Orchestration V2 no longer hydrates every historical turn item on most server paths. ProjectionStoreV2 adds bounded projections (getOperationalProjection loads only the latest turn item; getRecoveryProjection loads pending/running/waiting items) and targeted reads (getRun, getCheckpointScope, getRunTurnItems, getNodeTurnItems, getPresentTurnItemIds).

The orchestrator, thread management, provider turn control/start, checkpoint capture/rollback, runtime requests, provider session manager, and continuation/recovery services now call those APIs instead of getThreadProjection. Full history remains for snapshots, provider handoffs, v1 import / provider-switch dispatch, and client-facing getThreadProjection.

Projection maintenance startup verification uses validateThreadProjection, which decodes turn items in batches of 256 instead of loading full projections. Tests cover bounded loads and a larger many-items scenario (301 items).

^{Reviewed by Cursor Bugbot for commit 41fcc34. Bugbot is set up for automated code reviews on this repo. Configure here.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 760aeea8-0a12-4151-ae47-e4ceb4f9a21c

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR changes how projection data is loaded throughout the orchestration system, switching from full history loads to bounded/targeted queries. While designed as a performance optimization, this fundamentally changes what data is available at runtime in core orchestration paths, warranting careful human review.

^{You can customize Macroscope's approvability policy. Learn more.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want fixes drafted automatically? Bugbot Autofix can create code changes for findings. A team admin can enable Autofix in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit bc39163. Configure here.}