Every sparring round ends with a scorecard. This page explains exactly how that card is built — the rubric, the two grading layers, the scope rules, and the edge cases — from the platform's own grading implementation, so a coach can check the grade instead of trusting it.
Described from the Spar member platform's grading implementation and build records (personas and grading build reports, November 2026). Where grading behavior is a rule in code, this page says so.
Layer one
Each round declares its scope before it starts: which of the system's seven steps are in the run — Entry, Warm-Up, Probing & Layering, Presentation, Pre-Close, Close, and Button-Up — and whether the objection framework is graded. Steps that happened before the round's starting point are out of scope: a round that begins at the closing table can never mark Entry as missed. A drill opts out of steps explicitly; it never silently opts in.
An AI model grades the rep's side of the conversation against that scope: steps followed, steps missed, and which of the documented twelve closes were used. The model's output is then checked against the canonical lists in code — a step or close name that is not in the documented system is discarded before it can reach a card. The win line for a round is 80; winning a round's title fight cold, at 85 or better, credits the round outright — except the Round 10 capstone, which is always taken cold and never credited by test-out.
Layer two
Independently of the model, a deterministic scan in the platform's code checks the rep's own words against the documented language rules of the system:
Never: sign. Say instead: approve. Never: contract. Say instead: work order / agreement. Never: price, cost. Say instead: investment / overall investment. Never: finance, financing, monthly payments. Say instead: monthly investment plan. Never: buy / buy today. Say instead: move forward today.
The scan is authoritative — it needs no judgement call, so it is not left to one. Its results merge with the model grade on the card: steps followed and missed (in scope only, in canonical order), closes used, and every language violation, shown to the rep word for word.
The hard cases
One check overrides the score. If the homeowner invokes a missing decision maker and the rep keeps pitching instead of holding the no-one-leggers line and rescheduling, the round fails the doctrine check: the overall score is capped at 40 — below the win line, so the round can never bank a win — and the card cites the doctrine on its face: once the missing decision maker is invoked, the pitch is over; you do not run it, you reschedule it.
And when the grading call itself fails, the card says so. The language scan's results still stand — they are real and counted — the step score is left empty, and no score is invented to fill the gap. The card tells the rep to run it back. A grading system that fabricates a score when it cannot grade would make every other score on the platform worthless, so this one does not.
Sources: Spar member platform grading implementation (grading module, personas module, curriculum module), described November 2026; Personas Build Report, November 2, 2026; Curriculum Design, November 1, 2026.
Round 1 is free, and it is graded the way this page describes — scope, rubric, language scan, doctrine checks. The full camp is $29/month. Spar before you knock.