Skip to content

FAQ

Jump to bottom
Your Name edited this page Jul 23, 2026 · 2 revisions

❓ Frequently Asked Questions (FAQ)

📌 General

💬 Q: Can I use this without an AI agent?

A: Yes. The JSONL log is portable, transparent, and queryable via plain text or scripts. You can log decisions manually or integrate the MCP server with any system capable of stdio execution or JSON-RPC.

💬 Q: What if I don't have all the information when recording?

A: That's completely fine. Record what you know at decision time:

  • problem and chosen_solution are mandatory.
  • rejected_alternatives, predictions, and domain are optional.
  • Predictions can be added later using record-prediction.

💬 Q: How long are decisions kept?

A: Indefinitely. The log is an append-only JSONL file without deletion routines or retention limits.

⚙️ Technical & Storage

💬 Q: Does this server require an external database?

A: No. Storage is a single append-only JSONL file (~/.local/share/mcp-decisions/decisions_log.json). Zero server setup, zero schema migrations, zero third-party drivers.

💬 Q: Can I modify or edit a decision after it's recorded?

A: No. The log is immutable by design. To supersede a decision, record a new decision setting status="SUPERSEDED" and cross-reference the previous decision ID in adr_ref.

💬 Q: Can multiple agents or CLI instances write to the same log?

A: Yes. Append-only writes to line-delimited JSON files are concurrency-safe under standard OS filesystem POSIX append semantics.

🎯 Predictions & Outcome Validation

💬 Q: How should accuracy scores (0–100) be scored?

Compare observed empirical values against the initial prediction:

  • 90–100 (SUCCESS ✅): Prediction was accurate.
  • 50–89 (PARTIAL_SUCCESS ⚠️): Outcome was close, but drifted slightly.
  • 0–49 (FAILED ❌): Outcome missed expectations significantly.

💬 Q: What is the Outcome Gate Pattern?

Outcome Gate is an in-band reminder embedded inside MCP tool response payloads. It lists open predictions from the current session that still lack recorded outcomes, ensuring feedback loops are completed before sessions conclude.

Made for AI agents. Built for teams. Learn from every decision.

Clone this wiki locally