ZeroResp

Adaptive strategy for the Iterated Prisoner's Dilemma
Version 2.2 · Compatible with Axelrod-Python

Upstream submission: Axelrod-Python/Axelrod#1496 — ZeroResp + ZeroResp v2 (rev 2.2)

ZeroResp is a long-memory state-machine strategy designed for tournament evaluation: delayed adaptive retaliation, epoch-based debt accounting, noise-tolerant forgiveness, deadlock recovery, and cautious finite-horizon end-game logic.

---

Highlights

Property	Value
Class name	ZeroResp
Memory	Infinite (memory_depth: inf)
Stochastic	Yes (short retaliation delay 1–2; end-game probe)
Uses match length	Yes, when finite (match_attributes["length"])
Source inspection / state manipulation	No

---

Pre-submission benchmarks

Results from an independent PD sandbox (Top-40 pool: classic jailers, Axelrod elites, and family variants). Full tables: benchmarks/RESULTS.md.

Higher League (Stage 1)

• Pool: 40 strategies
• Match length: 400 rounds · reps: 2 · seeds: 7

Rank (avg)	Strategy	Avg score / round	Coop %
#2	ZeroResp v2.2	2.760	79.3%
#1	Omega TFT	2.761	79.8%
#3	ZeroResp v2.1	2.758	79.8%
#4	Generous TFT	2.721	87.0%

Best single seed (42): #1 / 40 (avg 2.760, win rate 79.5%).

Deep validation (summary)

Block	Setting	Outcome for ZeroResp v2.2
H2H titans	vs Omega TFT, 1000 & 5000 turns	Near-ideal mutual C (~2.999 avg/r); tiny edge via end-game
Endurance	Top-40 × 1501 × 5 seeds	Stable #2–#3 (avg rank 2.40); no memory collapse
Noise	0% / 1% / 3% flip noise	Strong at 0%; degrades under sustained 1–3% noise
Survival league	Eliminate bottom-5 until Top-5	#1 every round; final avg/r ≈ 3.000, coop ≈ 99.5%

Strengths: elite placement (#1–#3), clean H2H with Omega TFT, survival champion in cooperator markets.
Limitation: channel noise 1–3% hurts more than Omega TFT; not claimed as noise-optimal.

---

Quick start

pip install axelrod

import axelrod as axl
from zeroresp import ZeroResp

match = axl.Match((ZeroResp(), axl.TitForTat()), turns=200, seed=0)
match.play()
print(match.final_score())
# → e.g. (600, 600) under mutual cooperation

Unit tests

python -m unittest tests.test_zeroresp -v

---

Algorithm (v2.2)

1. Dynamic epochs (base_epoch=25) — systemic counters reset only when debt and the retaliation queue are empty.
2. Short adaptive buffer — retaliate after 1–2 turns (delay 1 under hostility / late pressure).
3. Noise-aware forgiveness — echo-forgive after own strike; limited one-shot forgive after long clean peace.
4. Deadlock break — detect CD↔DC alternation; offer cooperation to exit spirals.
5. Red line — permanent D after 2–3 systemic abuse events (threshold tightens under hostility).
6. Anti-raider — ≥2 late-game opponent defects → immediate red line.
7. Smart harvest / grim probe — only with known finite length and short remaining horizon; cautious vs never-defectors.

---

Repository layout

.
├── zeroresp.py                 # Canonical strategy module
├── tests/test_zeroresp.py      # Standalone unit tests
├── axelrod/
│   ├── strategies/zeroresp.py  # Drop-in for Axelrod-Python
│   └── tests/strategies/...
├── benchmarks/
│   ├── RESULTS.md              # Commission-ready summary
│   └── data/                   # Aggregate rankings & executive summary
├── REGISTRATION.md             # How to register in Axelrod
├── PR_BODY.md                  # Pull-request template
├── LICENSE
└── README.md

---

Axelrod-Python submission status

Item	Link / status
Upstream PR	#1496 (open, CI re-running after rev 2.2)
Superseded PR	#1495 closed (merged into #1496)
Contributing guide	Adding a strategy

Local drop-in copies live under axelrod/ for offline inspection. Registration snippets: REGISTRATION.md.

---

Classifier

Key	Value
memory_depth	inf
stochastic	True
long_run_time	False
inspects_source	False
manipulates_source	False
manipulates_state	False

---

Parameter

Name	Default	Meaning
base_epoch	25	Clean epoch length before systemic reset

---

Author

Evreu1pro · github.com/Evreu1pro

Prepared as a professional submission package for independent evaluation and possible inclusion in public strategy lists / Axelrod-Python.

License

MIT — see LICENSE.