How the CFR Score Actually Works

Last updated: October 2026

Investors ask us this constantly, so here's the honest, specific answer: which parts of Ficsers are deterministic math you can audit, and which parts use AI — and exactly what the AI is and isn't allowed to touch.

The short version

The score itself — the number, the grade, the pass/fail on each metric — is 100% hardcoded arithmetic. It runs the same way every time, on every submission, with no model in the loop. AI is used in exactly two places, both clearly separated from scoring: live market-benchmark lookups, and the written advisor narrative. Neither one can change your score.

Step by step

Hardcoded 1. Metric scoring

Each of the 5 scored dimensions — LTV:CAC, CAC payback, net revenue retention (or the archetype equivalent), runway, and month-over-month growth — is run through a fixed formula against a benchmark threshold for your industry archetype and target round. The archetype engine (ARCH) swaps the labels and input fields per industry — e.g. "NRR" becomes "Repeat Purchase Rate" for e-commerce — but the scoring math underneath is the same deterministic function for every company. Two founders who enter identical numbers get identical scores, always.

score = clamp(0, 100, f(metric_value, benchmark_threshold)) overall = sum(5 metric scores) grade = overall >= 80 ? "Strong Candidate" : overall >= 61 ? "Fundable" : ...lower tiers, all fixed cutoffs

AI 2. Live market context

Before you see any score, one call goes out to Claude (Anthropic's model) with live web search, asking for current funding climate, typical round sizes, and active investors for your country + industry + target round. This can nudge the benchmark thresholds used in step 1 (e.g. a hot market shifts the expected growth bar) — but it cannot touch the formula or hand-pick your grade. Its output is a small, typed JSON object the scoring code validates before using it; if the call fails or times out, scoring proceeds on static fallback benchmarks instead of breaking.

Hardcoded 3. The reveal

Your score stays hidden — no colors, no pass/fail, no running number — while you're still entering data. We made this change deliberately: letting people watch the score shift live as they type invites people to tune their inputs until it turns green, which defeats the point of an honest readiness check. The score is computed in the background the whole time either way; what's gated is only the on-screen reveal, which happens once, after you submit, as a single animated calculation.

AI 4. The advisor brief

Once your score exists, a second, separate call asks a language model to write the 5-section advisor narrative (round diagnosis, market context, top 3 priorities, what to do this week, timeline to raise) about your already-final score and metrics. This step only writes prose — it has no path back into the numbers. If this call fails, you still get your full numeric score and report; you just don't get the narrative section that run.

Hardcoded 5. Verification

Every CFR Brief is signed with an HMAC signature over its ID, company name, score and date. Anyone — a VC, an acquirer, a co-founder — can paste your report link into our public verify page and get a cryptographic confirmation that the score shown matches what we actually generated, with no database lookup required and no way for a founder to hand-edit a PDF into a better grade.

What this means for investors reviewing deal flow

Questions for diligence

If you're evaluating Ficsers and want to see the scoring code, the benchmark tables, or a specific submission's full audit trail, email us and we'll walk through it directly.

hello@ficsers.com