Difficulty
Your best score
N/A
May the ORION be with you - story-driven, deceptively simple, deeply strange.
In the year the stars stopped beating on that side of the sky, the Orbital Station Aethelgard began rewriting its memory. Not by mistake, but out of caution: a nearby binary black hole began to "sing" - gravitational pulses that frayed the edges of time and distorted the language of sensors.
You are named the Last Custodian - not because other Custodians no longer exist, but because you are the only one who still responds to an internal call. The Weaver (a conceptual entity, not a being) weaves coordinates and laws; he does not fail at mathematics. But when matter is in the fibrillation of space, imitators appear: the Echoes.
The Echo does not invent; it copies imperfectly. Sometimes it mirrors a phrase, other times it changes a frequency, and when nothing remains of the original, it whispers pseudo-texts that seem coherent: they sound human, but they do not carry the Primordial Charge - that invisible semantic thread that only the Weaver respects.
At the edge of observation, you have two registers:
Mission: Decide which recordings from the Test are Canonical (truly created by the Weaver) and which are Echoes (imitations).
Narrative Note
The Weaver does not fail at mathematics. This means that wherever an invariant numerical property exists in Train, it is usually a clue to separate Canon from Echo. But the Echo is cunning: it learns patterns, distorts them, or hinges signals through temporal mirroring. Take nothing at face value - even "simplicity" can be a trap.
Each row contains:
id_hex: hex identifier (e.g.: 5F3A1)sensor_x: raw X coordinate (affected by rotation)sensor_y: raw Y coordinate (affected by rotation)flux_energy: event energy (float)resonance_str: the whisper/text of the message (string, may contain archaic characters, symbols, noise)id_hex: identifierflux_energy: energyresonance_str: whisper/textsensor_x and sensor_y are missing (telemetry lost)Build a model that, for each row in obsidian_test.csv, produces a binary label:
Evaluation is performed using the Matthews Correlation Coefficient (MCC).
MCC is calculated as follows:
MCC = (TP * TN - FP * FN) / sqrt((TP+FP) * (TP+FN) * (TN+FP) * (TN+FN))
where:
Scoring (Total 100p):
The submission.csv file must contain one line for each row in the test set:
subtaskID,datapointID,answer1,1A2B3,11,4C5D6,0where:
subtaskID - always 1datapointID - the id_hex value from the test setanswer - the binary prediction (0 or 1)