DecktraceAutomated capability lab for card-game agents

Decktrace: OPTCG · fixed Imu mirror

From play
to proof.

Decktrace turns one checked-in One Piece Card Game matchup into a deterministic self-play lab. A challenger qualifies only when the rules contract is correct and balanced simulator games clear uncertainty, validity, and archive checks.

Play the latest champion Rules-Grounded Refresh status Review historical records
The game itself

Build a board. Defend the Leader. Land the final hit.

Two seats

One Piece Card Game is a two-player trading-card game. Players deploy Characters, attach DON!! to raise power, attack the opposing Leader, and answer attacks with counters, blockers, and effects. This project holds both players to the same checked-in Imu leader and deck so policy changes can be measured cleanly.

Players
2
Current profile
Imu mirror
Result
Win / loss
Action families
8
Information
Public board, resources, Leader and Life counts; private hands and hidden cards stay outside the opponent view
Agent boundary
A masked shared-action policy chooses only actions legal in the current player-visible state
Evaluation unit
Two games on one opening, with challenger and incumbent swapping seats

Game boundary: the capability claim is intentionally limited to this checked-in mirror matchup, not arbitrary leaders or decks.

Current simulator champion

Archive-safe champion

Controlled paired evidence cleared promotion and archive gates.

Subject: champion generation-019-candidate-03 · Opponent: generation-018-candidate-02 · 1,536 games

Paired score
53.2%
Observed point share in the controlled matchup.
Adjusted lower bound
50.3%
Conservative floor after uncertainty.
Paired openings
768
Common openings tested from both seats.
Invalid actions
0
Promotion requires zero.
Automatic promotion gateAdjusted lower bound > 50%

Promotion also checks invalid actions and comparisons with archived champions. Native qualification remains a separate lane.

  • Archive safety gatePass
  • Native qualificationNot run
Observed behavior
Critical misses
0 forced lethal · 1 immediate survival
Counts; opportunity totals are not recorded.
Event plays
100% effectful · 0% trash-only
5,410 total event plays
Counter spend
95.4% avoided hit · 4.6% still hit
27,648 resolved spend windows
DON attachment actions
9.5% to rested · 9.6% already attacked
15,330 attachment actions

Evidence is scoped to the checked-in Imu mirror and simulator runtime. Play the latest champion · Open the full campaign.

Champion evidence

Measured improvement stays attached to its game boundary.

Decktrace keeps training, comparison, and runtime evidence connected to the exact game profile and checkpoint that produced it.

Champion selectionAutomatedGame outcomes decide
Decision path

The operator retains a champion from measured game outcomes.

Training, comparison, and checkpoint publication stay automated; no human game, expert label, or manual promotion vote is required.

Training
Self-play
without labels
Comparison
Paired openings
and seats
Play
Packaged
browser actor
Evaluation boundaryExplicitSimulator scope stays visible
What the result means

A simulator champion is not automatically a universal game champion.

The evidence applies to the checked-in Imu mirror. Native execution, other decks, and broad live-game strength require their own evidence.

Rules
Fixed Imu
mirror
Runtime
Python
simulator
Coverage
Scoped
claim
Continue exploringAvailableTraining and engineering context
What is established

The latest champion powers simulator Play.

The verified campaign produced the archive-safe checkpoint that powers Play. Native qualification remains a separate, withheld boundary.

Preserved

Automated campaign execution

Execution
Deterministic headless games, masked legal actions, resumable training, and source-bound checkpoints
Selection
Common openings, paired seats, fresh lower bounds, zero-tolerance invalid actions, and archive opponents
Current result
The latest champion cleared paired-seat promotion and archive confirmation in the checked-in simulator and passed the Play-adoption comparison

See how this fits the Decktrace platform tour.

Technical evidence

Native reference remains a separate lane

A fresh six-game smoke executed the Generation 19 browser checkpoint against the recorded native runtime and finished with both native inventories matching the exact training deck. This is not per-game deck, card-rule, or Python/native parity evidence. OP13-097 may legally pay its optional 5 DON and then resolve no K.O.

Inspectable artifact
Complete source-bound native execution path
Purpose
Checkpoint execution and observation smoke, not the default training backend
Not established
Exhaustive rules parity, arbitrary-deck strength, or a user-qualified OPTCG champion
Engineering case studyPlatform Atlas