How accurate is the deal evaluation?

83.2% Top-3 accuracy (primary or secondary cause matched within the top 3) and 72.1% primary-cause accuracy (primary cause matched within the top 3), validated on n=4,345 editorial autopsies.

What data goes into a deal evaluation?

15 structural dimensions: sector, country, stage, total funding, moat type and strength, hype cycle, business model, B2B/B2C orientation, founding year, founder archetype, primary fatal mistake, collapse velocity, survival months, and peak valuation.

How does startup failure cause matching work?

Two signals are blended — historical base rates for the profile and structural similarity to documented failures — then reranked by recency to surface the most likely primary and secondary causes, reaching 83.2% Top-3 accuracy.

What is the survival curve in the risk report?

A Kaplan-Meier survival curve computed from the matched corpus. It shows the probability that a structurally similar startup survives to 12, 24, and 36 months, based on historical data from the 4,345 autopsy dataset.

// failure intelligence · how it works

The only system that matches a deal to documented failure patterns — with verified accuracy.

Matched against 4,345 documented autopsies, editorially curated. 83.2% Top-3 match accuracy (72.1% primary cause only) · the largest curated startup-failure dataset. It surfaces structural risks worth investigating — not a prediction of any one company's fate. CB Insights and PitchBook don't publish cause-match accuracy because they map outcomes, not structural precedent.

Evaluate a deal →Request access for funds

4,345

VERIFIED AUTOPSIES

editorially reviewed · 15 structured dimensions

83.2%

TOP-3 ACCURACY

primary or secondary cause · n=4,345

72.1%

TOP-3 PRIMARY CAUSE

primary cause only

$1.4T

CAPITAL DESTROYED

estimated · documented failures

// data quality

What "editorially reviewed" means.

Every case in the database passes a structured review before it is published. The criteria are the same regardless of sector, geography or funding size.

Documented shutdown

The company must have demonstrably ceased operations, been liquidated, declared bankruptcy, or been acquired under distress. Pivots and rebrands are excluded.

Attributable cause

The primary cause of failure must be identifiable from public documentation — founder post-mortems, regulatory filings, press coverage, court records, or investor statements. If the cause cannot be substantiated, it is left unassigned.

Reconstructible timeline

Key milestones — founding, peak, first warning signals, shutdown — must be reconstructible from at least two independent sources. Survival months are computed from these dates, not estimated.

Human field assignment

Analytical fields — collapse style, hype cycle, moat type, fatal mistake, archetype — are assigned by a human reviewer, not generated automatically. Each field is a considered judgment, not a classification label from a model.

Cases that do not meet all four criteria are either excluded entirely or published with a "unverified" flag. The verified / unverified split is reflected in the precision metrics above — backtesting is always run on verified cases only.

// precision metrics

Model accuracy evaluation.

5-fold stratified cross-validation on the full editorial corpus. Every verified case is evaluated exactly once. n=4,345, measured 2026-08-19. Backtesting runs on verified editorial cases only.

Metric	Accuracy	n	CI
Top-3 (primary or secondary cause)	83.2%	4,345	±1.3pp
Top-3 (primary cause only)	72.1%	4,345	±1.3pp

Version	Top-3	Dataset
V2	60.0%	BACKTESTING_RESULTS.v2_top3_cause_accuracy
V5	64.0%	BACKTESTING_RESULTS.v5_top3_cause
V6	69.9%	BACKTESTING_RESULTS.v6_top3_cause
V6 calibrada	71.2%	BACKTESTING_RESULTS.v6_calibrated_top3_cause
V6 (activa)	72.1% / 83.2%	scripts/backtest-nb-v8-ensemble.mjs

* Last measured 2026-08-19: 72.1% primary cause only / 83.2% primary or secondary. 5-fold stratified cross-validation, n=4,345, NB ensemble α=0.8. Each row cites the constant or script its figure comes from; versions without verifiable backing were removed from this table.

// model pipeline

Five steps in the risk analysis.

Each analysis runs the same deterministic pipeline. Every score is traceable to concrete signals.

Exact parameters, bonus functions and bayesian adjustment values are proprietary and not published. The pipeline is auditable — every score is traceable to concrete signals.

// model signals

The six signals that most distinguish failures.

Measured weights from the active model. A profile accepts up to 15 input fields; the model condenses them into 11 weighted scoring signals.

Founder archetype23.3%

Sector14.5%

Country / Region12.3%

Moat strength10.6%

Moat type9.3%

Hype cycle7.1%

Collapse velocity6.5%

Business model5.8%

Total funding5.4%

Founding year3.7%

B2B / B2C1.5%

Weights shown are the live model values (2026-08-19). Internal scoring internals beyond these weights are proprietary.

// reading the results

What the numbers actually mean.

The output is not a grade. Each number has a precise technical meaning that determines how to use it.

The score is not a failure probability

A score of 72/100 means the startup shares 72% of structural patterns with companies that failed in this pool, not that it has a 72% chance of failing. The model identifies resemblance, not destiny.

Structural similarity

Top-3 is more reliable than Top-1

The model identifies the primary failure cause in the Top-3 in ${MEASURED.top3Primary}% of analyses (${MEASURED.top3MultiLabel}% when secondary cause also counts). Collapse causes overlap — if the cause you suspect appears in the Top-3, the signal is strong.

n=4,345

How to read the matches

Score ≥ 65: strong match, review the autopsy in detail. Score 40–64: moderate, relevant patterns but structural differences exist. Score 30–39: weak, pool is small, treat with caution. Closer score = more weight on that collapse timeline.

Min threshold: 30

// common questions

Frequently asked questions.

The questions we get most from investors and analysts before they start using the platform.

We don't sell guarantees — we give you the tools to verify it yourself. Every case has a source you can check: a founder post-mortem, a regulatory filing, press coverage, court records. The entire editorial process is documented on this page. We publish the full backtesting history, regressions included — not just the versions that looked good. If something doesn't add up, you'll find it.

// auditable · verifiable sources

In 72.1% of analyses, the correct primary failure cause appears among the first three results. In 83.2% of cases, it's there when secondary causes count too. We validated this on 4,345 verified cases — every single one evaluated exactly once. We publish the full version history, including the releases where precision dropped. That's not marketing copy; it's the actual number.

72.1% primary · 83.2% Top-3 · n=4,345

Every case goes through four checks before entering the database: the shutdown must be documented, the cause must be traceable to public sources, the key milestones must be reconstructible from at least two independent records, and all analytical fields are assigned by a human reviewer — not auto-generated. No scrapers, no unverified aggregators. Backtesting always runs on verified editorial cases only.

4,345 verified cases · editorial review

The model weights structural similarity, not exact match. A company doesn't have to be identical to a case in the database for the pattern to be meaningful — a similar moat, a comparable founder profile, a matching market cycle all carry weight. If there aren't enough cases in your exact sector, the algorithm automatically expands to adjacent sectors and flags it explicitly. A score below 40 in a small pool is a weak signal, and the output tells you that.

Auto-expansion · minimum threshold: 30

Building this requires three things at once: thousands of structured, verified cases (years of editorial work), a rigorous backtesting framework, and the willingness to publish the numbers even when they show a regression. Most data providers don't invest in the editorial infrastructure. Those who have data don't build the model. We did both, and we publish the results.

Largest curated dataset · published precision

Two different standards. Top-1 means the model's first guess matches the actual primary failure cause exactly. Top-3 means the right answer is somewhere in the first three. We lead with Top-3 because startup failures are rarely one-thing — unit economics problems, market timing issues, and execution failures tend to overlap. In practice, analysts work with the full cause distribution, not a single slot. For context: a random baseline across 15+ cause categories would get under 10% Top-1 and under 25% Top-3.

Top-1 = exact first hit · Top-3 = correct cause in top 3

// GET STARTED

Does any startup in your portfolio follow these patterns?

1 free risk match, no card — in 30 seconds. The full platform (IC memos, deal scouting, portfolio signals) includes a 7-day trial.

Evaluate a deal free →Book a demo or compare plans