The Failure Oracle
Vertex AI Agent Builder · Gemini 3 · MongoDB Atlas

Catch startup failure 3–6 months early.

Enter your 11 startup metrics. A 3-agent AI pipeline matches them against 100 documented failure patterns — then shows you the warning signs, a deterministic escape plan, and exactly what the companies that survived did.

100
documented failure patterns
90%
fail for known, preventable reasons
3-agent
adversarial AI pipeline
0–100
transparent Oracle Score

Watch it call a collapse

We ran WeWork's Q4 2019 numbers through the Oracle — three months before the IPO imploded. Here's the verdict.

100 Oracle Score
WeWork · Q4 2019
Analyzing…
  • Burn multiple 54× — threshold 4×
  • Runway 7 months — floor 12
  • LTV:CAC 0.5× — needs 3×
Outcome 3 months later: $47B → $2.9B, IPO withdrawn.

From metrics to forecast in seconds

Three AI agents run in sequence — and every step streams live to your screen as it happens.

1

Investigator

Embeds your metrics with Voyage AI and searches all 100 patterns using MongoDB Vector Search + BM25, fused with Reciprocal Rank Fusion.

2

Challenger

A second, independent Gemini agent actively tries to prove the match wrong — catching false positives a single AI would miss.

3

Reporter

Assembles your Oracle Score, a deterministic escape plan, and the failure cascade showing exactly what breaks next.

What every analysis gives you

Oracle Score

A transparent 0–100 health score — click to audit every penalty and bonus with your numbers.

Pattern Match

Which of 100 documented failure patterns your metrics resemble, with a confidence score.

Escape Plan

The exact metric changes to drop below the danger threshold. Pure algebra, not AI advice.

Failure Cascade

What fails next, in how many days, at what probability — traversed with MongoDB $graphLookup.

Warning Signals

Early indicators already in your data, with how many days ago each first became detectable.

Survival Playbook

The numbered steps used by the companies that actually survived this exact pattern.

Proven on failures everyone knows

Run the famous collapses and watch the Oracle call them — with the same numbers the headlines missed.

WeWork Quibi Theranos Homejoy Jawbone Vine

Built on the full MongoDB Atlas + Google Cloud stack

Not a demo integration — every feature runs in the actual pipeline. The agents are built with Google's Agent Development Kit (the framework of Vertex AI Agent Builder); MongoDB is the partner, via its MCP server.

Google ADK (Agent Builder)Gemini 3 FlashVertex AI Atlas Vector SearchAtlas Search · BM25 + RRFMongoDB MCP $graphLookup$bucket · $facetChange Streams ACID TransactionsVoyage AI EmbeddingsGoogle Cloud Run Cloud Scheduler
The Failure Oracle
The Failure Oracle
150M+ startups worldwide YC · Techstars · Bootstrapped founders 90% fail for known, preventable reasons 100 documented failure patterns 3–6 months early warning window

Diagnostic Dashboard

3-Agent Adversarial Pipeline
3-Agent Adversarial AI Pipeline
1
Enter 11 metrics
MRR, churn, burn, NPS, LTV:CAC, runway — the signals that predict failure 3–6 months early
2
Investigator Agent finds the match
Voyage AI embedding → Atlas Vector Search + BM25 RRF → Gemini Flash parallel scoring → top pattern selected
3
Challenger Agent disputes or confirms
Second independent Gemini 3 Flash instance — actively seeks counter-evidence. >10pp divergence triggers DISPUTE and catches false positives.
4
Reporter synthesises + Oracle outputs
Oracle Score (transparent formula) · Escape Plan (deterministic algebra) · Cascade Graph ($graphLookup) · Survival Playbook
First time? Click the Quibi card below — watch the Oracle catch a $1.75B failure live. The Challenger verdict and Δpp confidence gap appear in the Agent Log. Want to see a DISPUTE? Try HighVelocity AI in the demos.
MongoDB Atlas — active in every analysis:
Atlas Vector Search Atlas Search · BM25 + RRF MongoDB MCP · 24 tools $graphLookup Cascade Change Streams · Self-Improving $bucket · $facet · $setWindowFields ACID Transactions

Your Current Metrics

Enter your startup's numbers below. Or let Oracle extract them automatically from any text.

Company & Capital
Revenue & Unit Economics
Load a real scenario — see exactly what the Oracle would have said:
LIVE VERIFIED

Proven on Real Failures

These are actual Oracle outputs — we ran the system against each company's representative pre-collapse metric profile. Click any card to reproduce the result yourself.

Pattern similarity scores shown above are live Oracle outputs — run by us before submission against each company's representative pre-collapse profile (illustrative figures chosen to capture the real failure pattern, not audited financials). Click any card to reproduce them on the live system. The Quibi result is projected from its early trajectory. Pattern similarity reflects narrative match against the failure library, not a statistical probability of failure.

Metrics Glossary — what do these numbers mean? Show ▾

Google ADK · Vertex AI Agent Builder Gemini 3 Flash MongoDB Voyage AI Atlas Vector Search Atlas Search · BM25 MongoDB MCP · 24 tools $graphLookup Change Streams $bucket · $facet ACID Transactions Cloud Run Cloud Scheduler

 ·  Apache 2.0  ·  GitHub  ·  Live on Cloud Run