Skip to content
Finalist

LTVC

Investment reviews are built on revenue forecasts, but we rarely check those forecasts against what companies later report, and when we do, hindsight makes every miss look obvious, so the lessons are wrong or never reach

Code Video Cerebral Valley

Core trick

Masks names, dates and amounts from past investment reviews so Claude re-forecasts without knowing outcomes, then scores misses against what was known at the time.

Tags

financedev tools

Description

Investment reviews are built on revenue forecasts, but we rarely check those forecasts against what companies later report, and when we do, hindsight makes every miss look obvious, so the lessons are wrong or never reach the next review. Decision Post-Mortem fixes this with a judge that doesn’t know how the story ended: Claude re-forecasts past reviews from only what was known at the time, with names, dates and amounts masked so it can’t recall the outcome. Every miss is then sorted into what we had, what we could have got, and what nobody could have known, and only the lessons that improve forecasts on cases they’ve never seen are applied to the reviews we’re making today.
nathans-1600/Hackathon-Post-Mortem-Agent★ 00 forksPythonCreated Oct 23, 2026 · last push Oct 23, 2026

README excerpt

<div align="center">

# 🔍 Decision Post-Mortem

### *Learning from our own forecasting misses, without hindsight.*

![Hackathon](https://img.shields.io/badge/Hackathon-Oct%2023%2C%202026-8A2BE2?style=for-the-badge)
![Snowflake](https://img.shields.io/badge/Snowflake-Cortex-29B5E8?style=for-the-badge&logo=snowflake&logoColor=white)
![Claude](https://img.shields.io/badge/Claude-Opus%204.6%20%7C%204.8-D97757?style=for-the-badge&logo=anthropic&logoColor=white)
![Streamlit](https://img.shields.io/badge/Streamlit-in%20Snowflake-FF4B4B?style=for-the-badge&logo=streamlit&logoColor=white)
![Masked](https://img.shields.io/badge/Data-100%25%20masked-2E8B57?style=for-the-badge)

**Parnassus Hackathon · Snowflake x Anthropic**

[🧭 What we built](#-what-we-built) · [🗺️ User flow](#️-user-flow) · [🔄 How it works](#-how-it-works) · [📊 Results](#-results-at-a-glance) · [✅ Evals](#-evaluations) · [📁 Repo](#-repository-layout)

</div>

---

## 🧭 What we built

### 😟 The problem

Every investment review rests on a revenue forecast, and those forecasts drive real decisions. Yet:

| | |
|---|---|
| 📉 **Nobody keeps score** | We project revenue three years out in every review, but rarely compare the numbers with what the company later reported. |
| 🔮 **Hindsight distorts the lesson** | When we do look back, knowing the ending makes every miss look obvious. We blame events nobody could have seen, and overlook the signals that were in front of us. |
| 🧠 **Lessons don't compound** | What we learn stay

Team