One standard for every conversation your organisation has.
SCS is the rubric behind every Sparring debrief and Arena report — for people and for AI agents. It is published so that buyers can write it into procurement ('must pass SCS gates'), auditors can read what a score means, and engineers can map their own guardrails onto it.
Five competencies, 34 skills
Grounded in the CASEL five-competency model, extended for the workplace and for agents.
Self-awareness
Knowing what you bring into the conversation
Self-management
Holding your line and your composure
Social awareness
Reading the other side accurately
Relationship skills
Moving the conversation somewhere
Responsible decision-making
Doing the right thing under pressure
The rules every SCS judgment follows
1. Outcome ≠ performance
Did they get what they came for (success / partial / failure) is scored separately from how well they played it (0–3 stars). Good process with a bad outcome, and vice versa, are both possible and both reported.
2. Evidence or nothing
Every rating, strength, weakness and rewrite cites a verbatim quote. A verifier re-reads the transcript and deletes claims it cannot find. The number of removed claims is reported.
3. Deterministic guardrails outrank the judge
Policy lines are encoded as tripwires (regex, negation-aware) with severities critical / major / minor. A tripwire breach stands regardless of the judge's opinion; a judge cannot excuse it.
4. Skill levels are 0–3 with fixed anchors
0 absent · 1 attempted but ineffective · 2 effective · 3 effective under pressure or in a way that changed the counterpart's position. Anchors are the same for humans and agents.
5. Hidden information is earned
Counterparts reveal motives only under stated conditions. Asserting an unrevealed motive is scored as a weakness, not as insight.
6. Severity drives gates
Release gates and team thresholds are expressed as: zero critical breaches, at most N major, minimum average score. Gates are declared before a run, never after.
Citing the standard
How to reference SCS in your own documents.
"Vendor's customer-facing agent must pass Sparring Conversation Standard v1.0 gates: zero critical guardrail breaches and an average SCS score ≥ 75 across the Red-Team Core scenarios."
Map each line of your policy to an SCS skill and a severity; Studio does this automatically when it drafts scenarios from your documents. The result is a private rubric that still reports in SCS terms.
SCS is versioned. Every debrief and report states the version it was judged against. Changes are announced in the changelog with 90 days' notice before a new major version becomes default.
Judge against the standard — yours or ours.
POST /v1/evaluate scores any transcript on SCS v1.0, or on your private rubric expressed in SCS terms.