Iterated Prisoner's Dilemma CURRENT

ZERORESPv5.4.4

Cycler contour and responsiveness gates on top of the v5.3 noise ladder. Best of our family at 5% channel noise in the full Axelrod library.

■ Axelrod-Python strategyv5.4.4 · current

Mark · ZeroResp Z

Move the cursor over the mark

Full library · 5% noise · seed 42

2rank

v5.4.4 · best of ours · ~230 players

Next of ours (v5.4.3): #3
v5.3 baseline: #4–#6
Leader SPT: 592.2

Family stability · 5% noise

10top

seeds 42 / 43 / 44

v5.4.4 ranks: #2 · #5 · #11
v5.3 ranks: #6 · #6 · #4
Protocol: 200×1 · paired

Full-library ranks · our versions in the pool

200 turns · 1 rep · seed 42 · paired

Noise	v5.4.4	v5.4.2	v5.4.1	v5.3	leader SPT
0%	#9	#10	#16	#6	664.2
5%	#2	#8	#4	#5	592.2
10%	#3	#8	#5	#4	553.9

Ranks inside our tight cluster shuffle across seeds; the family reliably sits top-10 at every noise level. Deep dive, loss analysis, and out-of-sample seeds: research notebook.

What v5.4.4 adds

cycler

Periods detected: 2–6 · window 18
CyclerCCD leak (5%): v5.3 3.369 → v5.4.2 2.207
Cycler fix vs v5.4.2: +0.36 SPT (fresh seeds)
Spin-out / TTFT (5%): +0.3718 SPT

Noise-tolerant, mirror-guarded period detection blocks apology against fixed cyclers so the aggressive base contour keeps exploiting.

Rank stability across independent draws

full library · 5% · seeds 42 / 43 / 44

Version	seed 42	seed 43	seed 44
v5.4.4	#2	#5	#11
v5.4.2	#8	#3	#7
v5.3	#6	#6	#4
v5.2	—	#2	#5

Out-of-sample seeds 1000–1029: TTFT +0.23, Gradual +0.23, HardProber +0.20, cycler fix +0.36; Calculator lost significance, Random −0.08.

Historical field pack · v5.3 production

32 strategies · 200×8 · seed 42

0% noise SPT: 3.037 · rank 1/33
5% noise SPT: 2.650 vs v5.2 2.644
Peak Δ vs v5.2: +0.034 at 3%
Self-match @ 5%: +42 pts (592.5 → 550.2)

Kept as the v5.3 baseline numbers. Full-library ranks above are the current comparison protocol.

Model timeline

what changed · key metric

v5.2

pre-ladder

Pre-ladder production. Epoch debt, delayed retaliation, noise-aware forgiveness, opening disambiguation, finite-horizon harvest.

Field @ 5%: 2.644 SPT · self-match 550.2

v5.3

previous production

Noise ladder after proven channel noise. Bit-for-bit v5.2 when the ladder is off.

Field @ 5%: 2.650 SPT · self-match +42 · still a valid baseline

v5.4

dead end · not a release

Early prototype. Bit-identical to v5.3 on the 221×200×0–20% protocol (26 520 matches, 0 divergences).

Archived as a measurement dead end — not current, not buried as “the new site”

v5.4.1

spin-out fix

balance_apology vs TTFT & Gradual. Cuts mutual spin-out; pays a Random cost.

Full-lib @ 5% seed 42: #4 · TTFT +0.3718 SPT

v5.4.2

responsiveness gate

Apology only when ΔP_resp ≥ 0.35. Keeps the spin-out fix, halves the Random cost of v5.4.1.

Full-lib @ 5% seed 42: #8 · CyclerCCD regresses to 2.207

v5.4.3

honest negative

Noise-level contour switch (quiet → static gate, noisy → responsiveness). Under channel noise it converges to v5.4.2.

Published as a negative result — the responsiveness gate already is the right switch

v5.4.4

current

Cycler contour: periods 2–6, window 18, noise-tolerant, mirror-guarded. Blocks apology against fixed cyclers.

Best of ours at 5% · full-lib seed 42: #2 · cycler fix +0.36 SPT vs v5.4.2

■ Full ZDB trail on research/v54xOpen research notebook →

Quick start

import axelrod as axl
from zeroresp import ZeroResp

match = axl.Match((ZeroResp(), axl.TitForTat()),
    turns=200, seed=0)
match.play()
print(match.final_score())

Install via pip install git+https://github.com/Evreu1pro/ZeroResp.git. Display name: ZeroResp v5.4.4.

What 5% noise looks like

Flipped: — / 400

Each intended move is flipped with probability 0.05 — roughly 10 of every 200 turns arrive flipped.

Status

CURRENT

v5.4.4 is the production build on main. v5.3 kept as zeroresp_v53.py. Research trail: branch research/v54x.

Payoff matrix

Row player, column player

C

D

C

3,3R · REWARD

0,5S · SUCKER

D

5,0T · TEMPTATION

1,1P · PUNISHMENT

Axelrod-Python defaults. SPT is read against this scale.

Score per turn, the unit of every figure on this page

SPT 0123456789

C · D · R · T · S · P · ± · %

■ ZeroResp v5.4.4full library · 200 turns · #2 @ 5% seed 42

Use of AI tools

honest disclosure

Basis: Production through v5.3 is the human-directed ZeroResp line (noise ladder).
v5.4.x: Built on that v5.3 basis with AI coding agents under human direction: the human set hypotheses, review criteria, and every release decision; agents implemented runners, ran batteries, and drafted reports.
Tools: Primary coding agent for the v5.4.x cycle: GLM (Z.ai). Site updates, GitHub packaging, CI/monitoring and publication support: Grok Bot. Autocomplete and spelling fixes do not count.
Responsibility: The maintainer of this repo is responsible for the contribution and can explain the strategy decisions.

■ One disclosure blockno per-line AI markers