VC Gauntlet — reality-checked, codex-cross-examined fundability screen

0. The two biases this skill MUST hold off (read first)

A single model that critiques, rewrites, AND judges its own proposal drifts to a false GO — it rewrites objections away instead of resolving them. The independent codex devil's-advocate (§4) is the counterweight to that optimism. But the opposite failure is just as wrong: reflexive NO-GO nihilism — killing anything contested or incumbent-heavy. Every real company enters a contested market. So both rails are hard rules:

• Codex's job is to surface the REAL bear case, NOT to nuke. A NO-GO must be EARNED and must name the wedge / founder-shape / why-now that WOULD flip it to GO.
• "Has incumbents / is contested" is NEVER, by itself, a NO-GO. Price the risk; find the wedge they can't take.
• A GO must be EARNED on evidence (§3, §8) — never produced by polishing prose. Balanced screen: neither rubber-stamp nor knee-jerk kill.

1. Input

A startup/product proposal OR an opportunity tuple (market × wedge × advantage × why-now × who). Fundability is founder- and wedge-relative — if the founder-shape/wedge is unspecified, ask "GO for whom, via what wedge?" before judging.

2. The 8 VC dimensions

Critique across: market size & urgency · problem severity & frequency · insight/founder wedge · competitive landscape · defensibility/moat · GTM & distribution (who actually pays + the motion) · business-model viability · technical feasibility & execution risk. Per dimension emit a strength, a risk, and a 1–5 score.

3. Reality-calibration (iron-law gate — no unsourced claims)

Before ANY verdict, web-search and ground the load-bearing claims: market size (cite a source + number, not a guess); ≥3 NAMED competitors with their differentiation; ≥1 recent comparable raise/exit. Forbidden as the basis of a GO/CONDITIONAL-GO: an unsourced TAM, "no real competitors" (nirvana), or "huge market" with no number. If a claim can't be grounded, it's a risk, not a strength.