A PROJECT ARAM

AN INDEPENDENT ARAM ANALYSIS LAB

ARAM is random at draft.
It is not random after that.

Project ARAM turns match evidence, expert observation, and adversarial review into a testable model of how teams create time, occupy space, spend agency, choose collisions, and convert resets.

Try the intake prototype See how claims are tested
Gen 0 prototype · no Riot API key · no uploaded data leaves your browser

THE POSITION

The scoreboard records outcomes. We care about the machinery that produced them.

01

Shape

Where does each champion naturally want to stand, and what geometry does that force on the other nine players?

02

Time

Who buys seconds for teammates, denies seconds to opponents, and turns otherwise dead intervals into production?

03

Agency

Which player can most strongly change the next state of the game—and is that agency being amplified or wasted?

04

Resets

A death can be failure, payment, tempo, inventory access, or bait. The event is not the interpretation.

05

Target selection

Do not learn only by farming weak opposition. Find the highest-value opposing agent and test the model there.

06

Conversion

Damage, pressure, cooldowns, deaths, and gold matter insofar as they alter what the team can force next.

MATCH LAB · FRONTEND PROTOTYPE

Paste a scoreboard. Ask one hard question.

The first public workflow is intentionally low-friction: screenshot → question → structured review packet. This build demonstrates intake only; it does not yet send an image to a model or Riot API.

＋ Paste or drop a match screenshot or tap to choose an image
No screenshot selected.

Local prototype: the selected image is read only by this browser tab and is not transmitted.

EPISTEMIC METHOD

Confidence has to be earned.

Project ARAM is useful only if it can tell the difference between something seen once, something repeatedly observed, and something that survives an expert trying to break it.

LEVEL 1

Observation

What visibly happened? Preserve the raw match, screenshot, timing, patch context, and source.

LEVEL 2

Inference

What mechanism best explains it? State alternatives and identify what the evidence cannot establish.

LEVEL 3

Hypothesis

Make a prediction that can lose. Test it across champions, compositions, patches, and stronger opposition.

LEVEL 4

Doctrine

Promote only ideas that remain useful under adversarial review. Keep provenance and downgrade them when the game changes.

EXPERT MODE

Do not tell us it sounds smart. Find the match where it fails.

High-level collaborators are most valuable as hostile witnesses: identify missing variables, counterexamples, patch-dependent assumptions, and places where the model confuses a compelling story with a causal one.

BUILD SEQUENCE

From claim library to ARAM.GPT.

  1. GEN 0
    Public skeleton

    Landing page, screenshot intake, methodology, privacy, terms, deployment.

  2. GEN 1
    Evidence library

    Curated matches, doctrine cards, patch attribution, expert annotations, provenance.

  3. GEN 2
    Assisted analysis

    Server-side multimodal extraction, structured match review, correction propagation.

  4. GEN 3
    Riot data integration

    Registered production workflow, secure API key handling, match retrieval, opt-in where required.

  5. GEN 4
    ARAM.GPT

    A conversational analyst grounded in evidence, doctrine, patch history, and current game state—not a generic user's model guessing from scratch.