The Contribution I Was Sure Would Work, and Killed | by Alan Scott Encinas | Jul, 2026 | MediumSitemapOpen in appSign up<br>Sign in

Medium Logo

Get app<br>Write

Search

Sign up<br>Sign in

The Contribution I Was Sure Would Work, and Killed

Alan Scott Encinas

3 min read·<br>3 hours ago

Listen

Share

Every origin story has the genius idea scrawled on a napkin. Mine had one too. I sketched it, I believed in it, I built it carefully, and then the data walked up and quietly tore the napkin in half.<br>This is the sixth entry in a log from inside the Hyperspectral Object Tracking Challenge 2026, where I’m tracking a single object through video shot in light the eye was never built to see. The goal is a podium and a paper. This entry is about the contribution I was most certain about, and the afternoon it died.<br>The idea that sounded obviously correct<br>My first real contribution, the one I expected to be the centerpiece, was a smarter way to choose which information the tracker pays attention to. The logic was airtight on the napkin. Be selective. Find the signal that best separates the target from its background, feed the model that, and drop the rest. It is the kind of idea that sounds so obviously correct that nobody bothers to check it.<br>That should have been my first warning. The ideas that feel unarguable are the ones most worth arguing with.<br>What the data said<br>I built it properly, written test-first, then measured it head to head against the cruder approach already sitting in the pipeline. The incumbent does roughly the same job in a blunter, less principled way, the kind of thing you would be slightly embarrassed to put in a paper.<br>My elegant version lost. Not by a hair. It was beaten clearly by the blunt incumbent, and when I stacked more cleverness on top to rescue it, the gap only widened.<br>The reason, once I dug in, was humbling in a way I keep having to relearn. My principled method quietly assumed it had enough clean data to estimate what mattered. In the hardest cases it simply does not. The estimate goes noisy at exactly the moment the stakes are highest, and the crude method, precisely because it was not trying to be clever, shrugged off that noise. Elegance was the liability. Bluntness was the feature.<br>Why I’m telling you a story where I lose<br>I did the thing this whole log exists to do. I published the loss. The failed contribution is written up as a documented negative result, not deleted, so the next person with the same airtight-on-paper idea can find my receipt and spend their week on something better.<br>Complexity has to earn its place against the simple baseline, every single time. The fact that your idea is more sophisticated is not evidence that it is better. It is only evidence that it is more sophisticated, which is a different thing wearing the same suit.<br>There is a quieter gift buried in a failure like this. A confident idea dying tells you something true about the shape of the problem that a success never would have. I can’t say more about what it told me, because that part is still my edge and this log runs on a deliberate delay. But that dead end was the most useful hour of the week.<br>Where I’m standing right now<br>A small change I’m keeping to myself for now nudged the board up to around 0.547, a hair under the podium line. The elegant idea that was supposed to carry me there is in the graveyard. The thing that actually moved the number was almost embarrassingly simple. That is, reliably, how it seems to go.<br>I’m going to set this competition down for a little while. Not because it is over, it is very much not over, but because I have six other competitions running and one of them has a story that has aged far enough to be safe to tell. This log was always going to jump between them. That is the whole reason I can write it truthfully while the races are still live.<br>More in this series This is part of an ongoing builder’s log written from inside live competitions. You’re reading where I was, not where I am.

Originally published at https://alanscottencinas.com on July 3, 2026.

Machine Learning

Kaggle

Computer Vision

Data Science

Negative Result

Written by Alan Scott Encinas<br>13 followers<br>·18 following

AI Engineer & Systems Architect. I turn complex ideas into working systems: cognitive AI, autonomous systems, robotics, defense.

Help

Status

About

Careers

Press

Blog

Store

Privacy

Rules

Terms

Text to speech