Re-run the Pathfinder audit — Phase 0 logged empty FAQ, thin troubleshooting, no pricing docs and nothing followed up

The Phase 0 Pathfinder audit in the Build Plan (internal) returned GO on a 90% hit rate across a 20-question probe set (18 excellent, 2 partial, 0 misses). It also recorded three content gaps that were never tracked:

Gaps: FAQ empty, troubleshooting thin, no pricing docs.

Four months later nothing has revisited them.

Why this belongs to response quality

These are retrieval gaps, and retrieval gaps are where fabrication comes from. When Pathfinder returns nothing useful, the generator still produces an answer — the groundedness gate (#143) then either withholds it or lets it through depending on whether it happened to name a CopilotKit identifier. Fixing the source content is the only intervention that produces a correct answer rather than a suppressed one.

"No pricing docs" is the sharpest of the three: pricing questions come from exactly the commercial-surface audience we most want a good answer for, and the bot currently has nothing to ground one in.

What to do

• Re-run the audit against the current index — 20 real questions, measure hit rate, and compare to the 90% baseline. Four months of docs changes make the old number stale.
• Confirm whether the three gaps still exist before acting on a four-month-old finding.
• For any that remain: decide whether it is a Pathfinder configuration fix (add sources, tune thresholds) or missing source content someone has to write. The Build Plan's own decision rule.

Also worth folding in: the Phase 0B prototype noted "confidence calibration needs tuning" and that was never closed out either. #122 (near-constant HIGH heuristic confidence) is likely the same finding surfacing again from the other end.

Start by reading the Build Plan's Phase 0 and 0B findings, then run Pathfinder against 20 real questions and compare the hit rate with the 90% baseline. Check whether the FAQ, troubleshooting, pricing, and confidence-calibration gaps remain, using #143 and #122 as context. Done means each remaining gap has a decided source-content or Pathfinder-configuration follow-up.