Author durable scenarios
Draft, version, and publish complete practice situations with hidden ground truth and rubric criteria.
Consultative practice infrastructure
Practice the conversation.Score the evidence.
A provider-neutral practice system where an external agent plays the buyer, preserves hidden context, and writes an evidence-backed assessment against an immutable transcript.
Private preview. Product behavior shown here reflects the current application contract.
BUYER
We have already tried two tools. The team stopped using both after a month.
SALESPERSON
What was happening in the workflow when adoption started to fall away?
CURIOUS FOLLOW-UP18 / 20
Evidence: salesperson turn 6 · asked about process before positioning.
Versioned scenariosImmutable session snapshotsAlternating turnsEvidence-backed score
What it owns
AI plays the role. The infrastructure keeps the rules.
The server preserves scenario truth, transcript order, hidden-context discipline, and scoring evidence without embedding a model provider.
Freeze each session
Start from an immutable participant-safe scenario snapshot and enforce alternating buyer and salesperson turns.
Assess from evidence
Validate cited transcript evidence and calculate the deterministic weighted score after qualitative judgment is written.
Deliberate product boundary
Keep AI behavior replaceable. Keep assessment mechanics stable.
Codex, Claude Code, or another agent can own the dialogue and judgment. Sales Trainer owns the session facts that should not change with the model.
Owned here
- Scenario versions
- Participant-safe projection
- Session transcript invariants
- Evidence and weighted scoring
Owned elsewhere
- A model client
- Prompt execution
- Buyer dialogue generation
- Qualitative judgment
The operating loop
One explicit path from intent to durable fact.
Prepare
An agent authors a scenario and publishes the stable version.
Practice
The agent receives hidden buyer context while the participant sees only the safe projection.
Alternate
Buyer and salesperson turns build one ordered, immutable session record.
Assess
The agent cites transcript evidence; the server validates and computes the weighted result.
$ sales-trainer sessions context SESSION_ID {"agentContext":{"scenarioSnapshot":{"definition":{"hidden":{"painPoints":[...]}}}}}
Private preview
Use the system without surrendering the judgment.
Codex, Claude Code, or another agent can own the dialogue and judgment. Sales Trainer owns the session facts that should not change with the model.