Productized audit · 48-hour turnaround · 3 slots per week

Bell Tuning Rapid Audit

Five Bell Tuning sensors run against your retrieval / agent pipeline. PDF report with bell curves, flagged pathologies, and a prioritized fix list. Walkthrough call. Seven days of Q&A. Fixed price. No scope negotiation.

$2,500 USD · flat · one-time
Book an audit What's included
Two slots open this week

What you get

Deliverables, not deliberations. Everything listed below ships within 48 hours of data handoff.

You receive

  • All 5 Bell Tuning sensors run against your data: context-inspector, retrieval-auditor, tool-call-grader, predictor-corrector, audit-report-generator
  • 8-12 page PDF report with bell curves of your per-query alignment distribution
  • Flagged pathologies by kind, with severity scores and concrete examples from your data
  • Prioritized fix list (3-5 items), ranked by effort and impact
  • 30-minute Zoom walkthrough of findings
  • 7 days of Slack / email Q&A after delivery

You provide

  • Retriever config, or read-access to your pipeline
  • 10-20 sample queries representative of production traffic
  • Small corpus sample (1k-10k chunks)
  • 30-minute kickoff call to confirm scope
  • Any existing eval set or ground-truth labels (optional, useful)

Pathologies the audit flags

None of these move precision@K cleanly. All are visible in the shape of your retrieval distribution before they surface as user complaints.

Score miscalibration
Rank order stays right while score values drift as the corpus grows. Precision@K is blind to it.
Rank inversion
Top-K contains all relevant docs but in reverse order. LLM reads top-down, hallucinates.
Redundancy attacks
Near-duplicate sources evict distinct ones from top-K. Answer is confidently incomplete.
Bimodal retrieval
Two content clusters both score high. Wrong cluster wins top-K. Confidently wrong answer.
Contamination drift
Off-topic chunks slowly accumulate in top-K. Mean alignment drops until complaints arrive.
Tool fixation / agent loops
Per-tool-call relevance distribution collapses. Loop detectable several turns before production breaks.

48-hour timeline

Clock starts at data handoff, not at purchase. Kickoff call same day.

1
Kickoff call
Same day (30 min)
2
Data handoff
Hours 0-4
3
Sensor analysis
Hours 4-36
4
Report delivered
Hour 48
5
Walkthrough
Day 3 (30 min)

Who this is for

Good fit

  • Team already running RAG or multi-agent in production
  • Users reporting "plausible but wrong" answers
  • Eval suite still passing, but quality feels off
  • Corpus or agent complexity has grown since launch
  • Need a defensible, numbers-backed read before a bigger decision (re-architecture, re-embedding, rewrite)

Not the right fit

  • Still in prototype phase, no production traffic
  • Looking for implementation labor (this is diagnosis, not remediation)
  • Need a custom research project or bespoke framework
  • Under $10k annual AI budget (this is $2,500; the remediation work that follows costs more)

Book an audit

$2,500 flat. 48-hour turnaround. Two slots open this week.

Pay now Discuss fit first

Not sure if your setup fits? Email first; we'll confirm fit before you pay.
Full framework: Bell Tuning manifesto + whitepapers + open-source sensors

How this relates to the open-source sensors

The five Bell Tuning sensors are MIT-licensed and installable via npx contrarianai-context-inspector --install-mcp. They are free. If you want to run them yourself, you do not need this audit.

The audit exists for teams that want the analysis done for them in 48 hours with a defensible report attached. You're paying for the turnaround, the pathology interpretation, and the prioritized fix list - not the tools themselves.