Editorial: Project OROBOROUS
It looks like a complex classification problem at first glance, but it actually tests your ability to explore data visually and logically. We are given a dataset with a mix of authentic data points (Canon) and fake ones (Echoes).
However, the training set has no labels, and the test set is missing the spatial coordinates (sensor_x and sensor_y). We can't train a standard model like a Random Forest because we don't have a target variable to train on, and even if we did, the test set is missing the most important numerical columns.
To solve this, we have to stop thinking about algorithms and start looking at the raw data geometry.
Finding True Signal in Noise
Since we don't have labels, we have to assume that the authentic data follows a strict underlying mathematical rule, while the fake data is just scattered noise. We have spatial coordinates () and flux_energy.
A standard trick in data science when dealing with 2D coordinates is to check their magnitude or distance from the origin. If we calculate and plot it against flux_energy, a massive clue reveals itself.
Plotting the 10,000 training points reveals a dense band around , with other points scattered away from it.
[ VISUAL: R² vs Energy Plot ] R² (x² + y²) ^ | x x | x / ⟵ Likely Canon (narrow band) | x / | x / x | x / | x / x ⟵ Likely Echoes (scattered points) | / x | x / x +----------------------------------> EnergyThe ratio of flux_energy to clusters near 13.37. Looking at the residual from this line gives us a way to select likely Canon points from the training data.
Bridging Gap to Test Set
We now have a group of likely Canon points in the training set. But we have a major problem: the test set doesn't have sensor_x and sensor_y. We can't plot to see which test points fall on the line.
We need to find a secondary pattern. We take the subset of authentic points we just isolated and look at their remaining feature: the resonance_str text.
If we count the number of characters (the string length) for every authentic point, a striking pattern emerges:
- Length: 3
- Length: 7
- Length: 13
- Length: 19
- Length: 31
Near the line, most strings have a prime length. Many of the scattered points have composite lengths (8, 10, 15, 21, etc.).
This is the hidden invariant. Since the test data contains the resonance_str column, we don't need the spatial coordinates anymore. We can completely ignore the missing and and just use this string length rule.
Python Implementation
The prime-length rule reaches an MCC of 0.9979 on the 3,000 test rows. It gets three rows wrong and earns the full 100 points.
import pandas as pd# 1. Load test datatest = pd.read_csv('obsidian_test.csv')# 2. Define a fast prime checking functiondef is_prime(n): if n < 2: return False if n in (2, 3): return True if n % 2 == 0 or n % 3 == 0: return False i = 5 while i * i <= n: if n % i == 0 or n % (i + 2) == 0: return False i += 6 return True# 3. Apply the rule discovered from the geometric plot# If the length of the string is prime -> 1 (Authentic)# Else -> 0 (Fake)test['answer'] = test['resonance_str'].apply(lambda x: 1 if is_prime(len(str(x))) else 0)# 4. Format the final submissionsubmission = pd.DataFrame({ 'subtaskID': 1, 'datapointID': test['id_hex'], 'answer': test['answer']})submission.to_csv('submission.csv', index=False)Wrapping Up
This problem demonstrates that Exploratory Data Analysis (EDA) is often more powerful than complex modeling. By plotting spatial properties, isolating a geometric invariant, and finding its correlation with text metadata, we bypassed missing data entirely. Before throwing data into a neural network, graphing it and looking for simple mathematical patterns is usually the smartest first step.