Milestone 25: deterministic run effort estimates and size-oriented queue filters #46

Implement Milestone 24 for ChatCodex: add deterministic run effort estimates and size-oriented queue filters so ChatGPT can explicitly classify runs by expected execution size without introducing backend autonomy.

Summary

The control plane can now represent blockers, readiness, priority, due dates, ownership, and blocker impact, but it still lacks a compact way to express how large or small a run is.

ChatGPT should be able to say things like:

• mark this as a small / medium / large run
• clear or replace an effort estimate explicitly
• inspect effort metadata in run.get, run.refresh, and runs.list
• filter the queue to small quick wins or larger projects
• sort runs deterministically by effort bucket where helpful

This milestone is about deterministic planning metadata and queue visibility, not automation.

In scope

1. Deterministic effort metadata

Add compact structured effort metadata to runs.

Conservative first version:

• enum-style bucket only, such as small | medium | large
• optional operator note / rationale is out of scope unless already trivial
• clearable by explicit action
• sensible null/default state for existing runs

2. Dedicated explicit update operation

Add a tightly scoped metadata operation such as:

• set_run_effort
• run.set_effort

This operation should:

• set, replace, or clear effort metadata only
• not execute work
• not replan, refresh, reopen, archive, snooze, reprioritize, or assign ownership automatically
• append an audit entry

3. Authoritative inspection support

Expose effort metadata in:

• run.get
• run.refresh where appropriate
• runs.list

RunSummary should carry concise effort fields if practical.

4. Deterministic list behavior

Extend run listing in a tightly scoped way.

At minimum support:

• filtering by effort bucket
• optional deterministic sort by effort bucket
• compatibility with existing filters and queue views

5. Audit trail integration

Append a deterministic audit entry such as:

• run_effort_set

If helpful, include concise metadata such as previous and next effort value.

6. TypeScript MCP gateway updates

Keep TypeScript thin:

• add schemas
• validate inputs
• call the daemon
• map responses

Do not move effort logic into TypeScript.

7. SQLite persistence updates

Persist effort metadata with safe, idempotent migration support.

New databases must work immediately.
Older databases must migrate safely and deterministically.

8. Tests and CI

Add milestone-scoped tests for:

• setting effort
• clearing effort
• rejecting invalid effort values
• persistence roundtrip of effort metadata
• audit trail entry creation
• list filtering by effort
• deterministic ordering behavior
• exact MCP tool registry and daemon method registry
• no hidden-agent regression

Out of scope

Do not implement:

• automatic effort inference
• time tracking
• velocity calculations
• reminders
• notifications
• background wakeups
• autonomous reprioritization
• backend LLM usage

Acceptance criteria

• The only LLM in the stack is ChatGPT
• no model/provider SDKs added
• no hidden backend agent loop
• no coarse autonomous public tools
• effort metadata is explicit, deterministic, and inspectable
• list filters remain narrow, composable, and mergeable