Give a Python knowledge base's Slack sync a raw layer that stores every message and reply exactly as Slack returns it and builds identical month documents from it, refetching a thread's replies only when it changed.
- Version
- Category
- Engineering
- Model configurations
- 2
- Attempts
- 17
- Public trials
- 178 passed
- Median runtime
- 13m 56s
Instruction
Slack is the slowest part of the n0 sync, most of its three minutes. Every run lists each channel's history, and whenever anything in a month changes, the month is rebuilt by refetching the replies of every thread in it. So one new message in a busy month costs dozens of serial `conversations.replies` calls at Tier 3 pacing. Let's do the raw reconciler layer, at 999 page size: - Keep the Slack messages and thread replies in the project's store exactly as Slack returns them: every message and reply, bot posts included, with all of their fields. Nothing gets decided at fetch time, so when we later change what a month document shows (we do want bot posts in there eventually) it is a re-render, not a refetch of the workspace. - Build the month documents from what is stored, so that a thread's replies are fetched again only when its reply count or latest reply has changed. Listing each channel's history on every run stays: that listing is where those signals come from. - Month documents must come out byte-for-byte as they do today. - Months that Slack's retention no longer shows must keep working the way they do now. - Page Slack's history and replies listings at 999. ## Environment PostgreSQL on localhost:5432 and the local database endpoint the app's driver uses are running when you start, and the dev env files point at the dev database.
Public trials
Results by model
| Model · harness | Mean reward | Range | ± SE | Attempts |
|---|---|---|---|---|
| Claude Opus 5.5Grok Build | 100.0% | 100% – 100% | — | 1 |
| Frontier model ACursor CLI | 43.8% | 0% – 100% | 12.8 pp | 16 |