What the Machine Sees
A few weeks ago, our system surfaced a pattern nobody had asked it to find: startups killed by competition take an average of 4.2 years to die. Long enough for every founder to believe they're winning. That insight emerged from 188 postmortems cross-referenced against 13,499 companies across 48 sectors — from AI and deep tech to agritech and space. No analyst requested it. The knowledge graph surfaced it on its own, because that's what compounding intelligence does. This is what we're publishing, starting Monday.
The Problem We Couldn't Unsee
When you evaluate hundreds of startups, you start seeing ghosts. A pattern in Company #14 that reminds you of something from Company #87 — two years ago, different sector, similar death. But you can't prove it. You can't even retrieve it. It's somewhere in your memory, diluted by a thousand pitches since.
That's the gap we spent a year building a machine to close. We built Manthan Intelligence to do what human analysts can't: cross-reference every company against every other company, every postmortem against every pitch, every sector pattern against every deal — instantly, continuously, and with total recall. Multiple independent analytical lenses evaluate each company from angles most teams never consider. A calibration loop scores every assessment against what actually happened — and feeds the misses back into the system so the same mistake doesn't repeat. The architecture behind this is not something we can explain in a single article. But we can show you what it finds.
The knowledge graph now holds 13,499 companies, 1,747 mapped investors, 11,486 relationships, and 188 dissected startup postmortems. Every entity enriched. Every analysis scored against what actually happened months or years later. The scoring methodology alone — how we weight confidence, penalise hindsight bias, and separate genuine foresight from lucky guesses — took longer to build than the knowledge graph itself. And every new company that enters the system makes the pattern recognition sharper, because intelligence compounds.
Why We're Publishing This
Intelligence that sits in a closed system doesn't compound as fast as intelligence that's tested in public. Every Charaka Note we publish is a hypothesis — backed by data, open to scrutiny, and scored against what happens next. When we're wrong, we'll say so. When the data contradicts conventional wisdom, we'll show you why. The patterns are too useful to keep locked in a database, and too important to publish without showing the work.
The name comes from Charaka — the ancient Indian physician who wrote the first systematic medical text. He didn't theorise about disease from first principles. He observed thousands of patients, classified what he saw, and wrote down what actually worked. Twenty-five centuries later, we're doing the same thing with startups. Observing at scale. Classifying what kills and what survives. And publishing the patterns so others can use them.
What You'll Get
One data-backed insight every weekday. 500-800 words. Free. No hand-waving, no "top 10 tips for founders." Every claim backed by the knowledge graph — specific numbers, specific patterns, specific implications. If we can't point to the data, we don't publish it.
Five pillars, one per weekday:
| Day | Pillar | What You Get |
|---|---|---|
| Monday | Pattern Intelligence | Cross-company patterns with predictive power. The insights that connect Company A's failure to Company B's risk — across sectors, stages, and geographies. |
| Tuesday | Sector Deep Dive | Sector-level analysis from real company evaluations — AI, fintech, deep tech, healthcare, logistics, space, and 42 others. What the data says and why it disagrees with consensus. |
| Wednesday | Death Diagnosis | What kills startups — drawn from the postmortem corpus. Named failure modes. Time-to-death data. The patterns that companies walk into right now without knowing. |
| Thursday | Inside the Machine | How an AI-native analytical system actually works. What broke this week. What we recalibrated. Radical transparency about the methodology — because trust is earned by showing the wiring, not hiding it. |
| Friday | Signal Detection | Early signals from our daily research agents. Regulatory shifts, competitive moves, emerging patterns — before they hit the mainstream news cycle. |
Intelligence That Compounds
Here's what makes this fundamentally different from every other startup newsletter: the system gets better with time, and it can prove it.
Every company analysed teaches the system something about the sector that company operates in. Every postmortem sharpens the failure pattern library. Every prediction scored against reality calibrates the next prediction. The 100th Charaka Note will draw on patterns the 1st note couldn't see — because by then, the knowledge graph will have absorbed another thousand data points, another fifty postmortems, another hundred cross-references that didn't exist today.
Most analysis gets stale. A report published in January is outdated by March. The knowledge graph doesn't have that problem. It's additive. The insight we publish on Monday is built on a foundation that includes everything we published before it, everything we've analysed since, and everything the autonomous research agents discovered overnight. There are layers of this system we haven't described yet — how sectors talk to each other inside the graph, how a postmortem in logistics sharpens a prediction in fintech, how the calibration loop knows when it's getting overconfident. That's not a newsletter model. That's compounding intelligence — and it's the reason we believe this gets more valuable the longer you read it, not less.
Who Should Read This
If you're a founder and you've ever wondered whether the risk that keeps you up at night is actually the one most likely to kill your company — this is for you. We've classified the real causes of startup death, but we've also mapped what the survivors did differently: which operational patterns correlate with successful follow-on rounds, which growth strategies actually hold up at scale, and which sector-specific playbooks produce compounding advantage rather than a race to the bottom. The Charaka Notes archive is designed to be mined — the deeper you go, the more patterns connect to your specific situation. Start with the failure modes that apply to your sector. Stay for the growth intelligence that only emerges when you can read across thousands of companies, dozens of sectors, and hundreds of outcomes simultaneously — the kind of pattern recognition that no human team can hold in their heads, no matter how experienced.
If you evaluate startups — as an investor, advisor, accelerator, or board member — and you've ever suspected that conventional diligence checklists miss the patterns that actually matter, we'll show you where diligence time is wasted and where it's underweighted — patterns that only surface when you can cross-reference every deal against every other deal, across every sector, at a scale no analyst team can replicate. Specific enough to change your next memo.
If you're building with AI and you want to see what happens when you let a knowledge graph, a calibration loop, and a multi-lens analytical system run at scale — not as a demo but as a production system held accountable to reality — Thursday's column will show you the wiring. Every week. Including the failures.
We will publish what the machine sees. We will show our methodology. We will admit when we're wrong and publish what the system learned from the miss. And we will let the intelligence compound in public — so you can decide for yourself whether a system that has analysed 13,499 companies across 48 sectors has something useful to say about yours.