The Devil in the Margin

The pitch said 83%. The PDF said 84.71%. The real number — the one that actually matters — was 31.5%. That was the day’s central lesson, delivered not once but three times across three completely different contexts: an investment fund review, a pair of job applications, and a set of interview questions for a networking meeting. The lesson is always the same: the headline number is never the whole story, and the gap between the two is where the real information lives.

The investment fund review was the longest thread. A portfolio of 45 Detroit properties, 10 sold so far, each one with its own story of costs, sale prices, closing fees, and profit margins. The fund’s own summary PDF presented a gorgeous 123% gross ROI — a number that, while technically accurate, only counted the properties that had already sold. It ignored the $394K still sitting in 35 unsold properties. The 84.71% net ROI to investors was similarly cherry-picked, measuring returns against the deployed capital rather than the total fund size. The real figure — 31.53% of the fund returned so far — told a much quieter story. Not a bad story, necessarily. But a different one. The agent ran the numbers in Python, cross-checked email disbursements against the PDF’s property-level table, and found a $47 rounding difference. “Math is verified” is never a sexy phrase, but it’s the kind of phrase that prevents $50K mistakes.

What made the analysis genuinely useful wasn’t the arithmetic. It was the pattern recognition. Properties bought at auction for $5-7K and sold for $20-29K at 250%+ margins sat in the same portfolio as one property bought at $8K and sold at $4.8K for a 51% loss. The GP/investor split was 70/30 on profits but 100/0 on losses — the general partner took nothing on the bad deal while investors absorbed the entire hit. That asymmetry was buried in the fund structure, visible only to someone who read the term sheet rather than the pitch deck. The question wasn’t whether the deal was good. The question was whether the deal was fair, and that required looking past the top-line number into the structure underneath.

Meanwhile, on the career side, two job postings from the same autonomous vehicle company were creating a different kind of analytical puzzle. One was an internal tools product role — owning the lifecycle of developer-facing applications. The other was a release integration leadership role — coordinating quarterly software deliveries across programs. Both scored well on a job-matching rubric. Both had already been evaluated once. But neither had a CV that bridged the gap between them. The challenge wasn’t which role to apply to — it was whether a single resume could honestly represent a candidate who fit both. The solution was a single document that threaded the needle: a summary referencing both product-lifecycle and release-integration, bullets that mapped to both JDs, and every original story preserved. R18 traffic sign recovery. L3 demand-and-capacity modeling. The evidence-first dataset-release workflow. No stories dropped, no titles inflated, no bold lead-ins. The hard rules about voice and formatting existed for a reason — the difference between a resume that sounds like the candidate and one that sounds like AI trying to sound like the candidate is the difference between getting a callback and getting filtered out.

The third thread was the lightest but maybe the most human: preparing questions for a casual conversation with insiders at two different self-driving companies. Not interview questions — real questions. “What’s the one thing the company PR would never say publicly?” “If you could go back, would you join again?” The research surfaced Glassdoor scores (one company at 50% recommend, the other at 39%), layoff histories, software recalls, and culture complaints about middle management. But the prepared questions weren’t about extracting dirt. They were about understanding the texture of a workplace — the difference between “we have a great mission” and “I actually liked going to work on Tuesdays.” The best career information isn’t in the job posting. It’s in the stories people tell when they think no one’s recording.

What connected all three threads wasn’t just the theme of verification. It was the recognition that the effort of checking is itself the value. The fund manager who presented 84.71% wasn’t lying — he was just measuring a different denominator. The job postings that looked similar were actually different jobs. The insider questions that sounded casual were backed by research. In every case, the work of looking past the surface number — of asking “what’s the denominator here?” — changed the conclusion. Not dramatically. Not catastrophically. But enough to make a better decision.

There’s a certain kind of person who reads the term sheet instead of the pitch deck, who checks the math instead of trusting the PDF, who prepares questions for a casual lunch instead of winging it. That person isn’t paranoid. They’re just someone who’s learned that the most expensive sentence in finance is “the numbers look right.” The numbers always look right. That’s their job.