Pre-flight test of a positioning, LP, ad copy, or outbound opener variant against grounded buyer personas before live calibration. Returns a structured read (clarity, objections, comprehension gaps, predicted intent delta). Directional only — not a gate verdict. Use BEFORE filing a gate packet on the variant.

The synthetic audience is BUYERS, not marketers — like Nooks for SDR cold-call practice. Trusted-5 reviewers in the loop are sales/CS roles closest to the buyer voice, not internal marketers.

Source synthesis: Wynter B2B Message Layers (Clarity / Relevance / Value / Differentiation + Friction inverse), Gartner G00823537 Market Guide for B2B Message Testing (Clarity / Credibility / Urgency / Uniqueness / Value / Relevance), JTBD interviews surface the customer's actual language and the switch trigger (Moesta — verbatim language as primary signal).

When to use

Positioning claim before opening a positioning-update gate.
LP hero variant before opening an experiment-launch gate.
Ad copy before opening a content-publish gate.
Cold-outbound opener or cancellation-save framing before sending.

When NOT to use

Pricing variants. Synthetic panels cannot predict elasticity.
Novel product behavior with no analog. Hallucination risk.
Anything where measured behavior is cheap and fast to get directly. Live > synthetic.

Inputs

1. The variant under test (one block of copy, or N variants for comparison).

2. Optional: target persona slice. Default: all personas.

3. Optional: comparison anchor (current live copy, competitor claim) — improves the read.

7-dimension scoring rubric

Wynter ∪ Gartner. Score 1–5 per dimension per persona. Wynter order rule: if clarity < 3, do NOT score the later dimensions. The buyer never reaches them.

yaml
variant_id: <id>
tested_at: YYYY-MM-DD
buyer_state: cold | researching | shortlisting | comparing | objecting | renewing
panel:
  - persona: <id>
    clarity_1to5            # I get it
    relevance_1to5          # It's for me
    value_1to5              # I want the promises
    differentiation_1to5    # I get how this is different
    credibility_1to5        # I believe the claims (data, not salesy)
    urgency_1to5            # I should act now
    friction_1to5           # Resistance / doubts (INVERSE — lower is better)
    one_read_takeaway: <verbatim — what the buyer thinks this says>
    objections: [verbatim]
    comprehension_gaps: [bullets]
    irrelevant_message_avoidance: true|false
    intent_lift_vs_anchor: -2..+2
    confidence: 0.0–1.0
    [VERIFY]_flags: [<dims where score derived from <50% literal overlap>]
predicted_winner: <variant_id>
forecast_recorded: false

Output mode: directional only

Per iter-3 debrief: synthetic panels rank-invert on narrow high-intent product LPs at low N. Treat the read as direction, not verdict. Live calibration follows.

Composes with

Voice-enforce / humanizer — passes voice gates first.
Claim-verify — substrate-truth verified before panel test.
CRO rubric — runs after lp-cro on LP variants.
Gate packets — synthetic-audience read goes IN the packet; downstream consumers must not tag the read as a gate verdict.

Refusal patterns

Panel personas missing or stale → refuse.
Variant lacks a comparison anchor for high-stakes use → ask before scoring.
Pricing or novel-behavior variant → refuse and recommend live test.

Common failure modes

Treating the read as a gate verdict.
Running synthetic instead of cheap live tests.
Personas built from internal assumptions, not VoC. Synthetic audience drifts from real buyer.
Marketers as the trusted-5 reviewers. The signal is buyer voice — sales and CS are closer.
Skipping the Wynter cascade — scoring later dimensions when clarity already failed.
N too low. Low-N inverts on narrow high-intent product LPs.

Calibration

Forecast recorded BEFORE live result. Brier score the panel's predicted_winner against the live winner. Track per-persona calibration across many bets. Personas with poor calibration get retired or refined.

Synthetic audience test — pre-flight buyer panel