01
Overview
The current methodology is deliberately narrow. The parser is deterministic, the scoring formula is public, and there are no knobs to turn. This is a design choice — keeping scores comparable across accounts is more important right now than letting each user customize their setup.
02
Parser rules
The parser works on the raw text of each captured response plus its citation list. It never calls an LLM at scoring time, which keeps scores reproducible.
- Casing
- Case-insensitive brand matching.
- Boundaries
- Word-boundary matches only; no substring traps.
- Citations
- Citation URLs are normalized to root domains.
- Language
- English responses currently; other locales are planned.
03
Scoring model
The scoring formula is the one shown in Scores & Metrics. There is no A/B on weights and no learned component. Any change to the formula would bump a version number and be documented in the changelog.
04
Current constraints
Several things are intentionally not configurable right now. These are tracked for later releases.
- Prompt editing
- Prompts are system-generated only.
- Model selection
- ChatGPT and Gemini only, always both.
- Scheduling
- Runs are manual-trigger; no cron yet.
- Webhooks
- No outbound webhooks right now.