New Strategy Submission: "Schwoba sen gnitz" (A 1985-inspired Reciprocal Hybrid)

[Gerowasch]

Hello Axelrod Team,

I would like to submit "Schwoba sen gnitz", a highly robust, adaptive strategy to the repository. It is based on a successful empirical design I developed with TurboPascal during the German c't tournament era back in the late 80's, now fully ported to clean Python.

Author: Georg 'HackyHackberger' Schmidt

Behavioral Record:
In my private offline simulations against common adaptive baselines, this strategy successfully contains and outperforms standard pattern-recognizing models, neural networks, and stochastic gamblers:

• It successfully disrupts standard Artificial Neural Networks by intentionally breaking clean predictive patterns.
• It constraints Stochastic Risk Models through its strict, asynchronous timing.
• It completely isolates and starves permanent defectors via its unique quarantine mechanic.

Core Mechanics:

Organic Self-Play:
It starts 100% cooperatively. If it meets a clone of itself, both stay peaceful forever (Action.C), maximizing tournament efficiency without relying on fragile, noise-prone "secret handshakes".
The Reactive Phase Shift:
The moment the opponent defects once, the internal tracking mode wakes up permanently. From this point on, it introduces a low unprovoked defection rate of 10.2% to systematically test boundaries and exploit softer strategies.
De-escalation:
To prevent destructive, infinite echo-loops with other reciprocal strategies (like standard TFT), it forgives and dampens hostility with a 89.7% probability.
Quarantine Lock:
If an opponent continuously defects for 10 rounds, it identifies it as an unredeemable defector (Grim or AllD) and locks down into permanent defection to minimize further damage.

I look forward to seeing "Schwoba sen gnitz" in the official tournament framework!

[Gerowasch]

Hello Axelrod-Team

I wanted to provide a quick update regarding the tournament tests. I ran a full Round Robin tournament with all available strategies in my fork to see how it performs under realistic conditions.

Unfortunately, the standard 1,000-turn run with one execution ran into the hard 6-hour GitHub Actions execution timeout without completing.

After I reduced it to 200 turns, the same happened. Again, it ran into the execution timeout without completing. A small test with only very few strategies finished without any problems, so I can exclude that it happens because of my strategy.

Since further reducing the turn count wouldn't make sense from a game-theoretic perspective (as learning algorithms require a baseline data horizon to function properly), I will rely on the theoretical proof for now.

I look forward to seeing how your frameworks handle this test.