One thousand applications. Ten cheques. Which ten?
Lanternfield, an accelerator and seed fund, meets in April 2026 to choose its spring round: eighty applications are waiting, the round has $2.5 million and room for at most ten investments, and four years of records are there to learn from. The committee's chair asks the analyst five questions: what did four years of decisions get us, are we backing the right signals, which markets are worth being in, are we fair to every founder, and which ten of the eighty should we back?
Worth Backing? answers all five from the data and ends with a scorecard, ten picks and a one-page investment memo. On the way it teaches SQL and Python from their first lines: SQL on DuckDB, a database engine that runs inside Python and reads files directly, and Python with pandas for tables and matplotlib for charts. Each idea arrives when one of the committee's questions needs it, shown with the result it printed, the mistake beginners make with it, and the same move in the other language, in a spreadsheet or in a BI tool.
Inside:
- How accelerators and seed funds work, why most startups fail while a few return the fund, and what revenue per customer, churn, burn and runway say about a young company
- SQL for the shortlist: filters, totals and rates, joins across the plan, the pitch and the outcome, and window functions for ranks, shares and growth
- Cleaning with a record: placeholders, duplicates, impossible values, a unit slip in the revenue and names that miss a join, fixed in views with a cleaning log and never in the raw files
- Python for the analysis: DataFrames, distributions and outliers, fair comparisons between groups, cohorts, churn and runway month by month
- From chart to decision: the chart the question needs, a market map and a one-page dashboard, a scorecard tested for how much its ranking depends on its weights, and the memo
- Ten checks before an insight reaches the committee, and ten ways a startup chart misleads
The examples use six tables about Lanternfield and the 1,000 startups that applied to it from January 2022 to March 2026: business plans, pitch decks, funding outcomes, revenue snapshots, the monthly metrics of the funded startups and a market analysis. Lanternfield, its applicants and their numbers are fictional, and the data contains 30 faults put there on purpose for you to find. Every block of code marked as run was executed in the order printed, and what is printed under it is what it returned.
The quickest way to run the examples is the LucidPlexus Workbench, a notebook in your browser that runs SQL and Python with the six tables already loaded: nothing to install. The same code runs in JupyterLab on your own computer with Python 3.11 or later, DuckDB, pandas 3 and matplotlib. Data: lucidplexus.com/community/dataset.php?slug=digitalfinance
For analysts and students who want the whole workflow, from a vague question to a decision; for people in or near venture capital, accelerators, consulting, corporate strategy or business intelligence; and for founders who want to see their own numbers the way an investor sees them. No programming is needed to start: you need to have used a spreadsheet and to be comfortable with percentages. Nothing in the book is investment advice.
A LucidPlexus Field Book. First edition, 2026.