mirror of
https://github.com/saymrwulf/autoresearch-quantum.git
synced 2026-05-14 20:37:51 +00:00
saymrwulf
f9b8f3457f
Karpathy-style autoresearch engine for encoded magic-state preparation on the [[4,2,2]] quantum error-detecting code using Qiskit Aer simulation. Five-rung progressive search: baseline -> stability -> transfer -> factory -> Rosenfeld. Smart challenger generation (neighbor walk + random combo + lesson-guided). Machine-readable lesson feedback with per-dimension effects, interaction detection, and cross-rung propagation. Factory throughput scoring. Resumable execution. 21 tests, all passing.
71 lines
1.8 KiB
YAML
71 lines
1.8 KiB
YAML
rung: 5
|
|
name: "Rosenfeld Direction"
|
|
description: "Identify which heuristics matter for cultivation/distillation workflows. Narrowest search space; only proven dimensions survive."
|
|
objective: "Determine which preparation, verification, and transpilation choices are load-bearing under factory-realistic conditions, as a precursor to distillation pipeline integration."
|
|
|
|
bootstrap_incumbent:
|
|
seed_style: ry_rz
|
|
encoder_style: cx_chain
|
|
verification: z_only
|
|
postselection: z_only
|
|
ancilla_strategy: reused_single
|
|
optimization_level: 3
|
|
layout_method: dense
|
|
routing_method: basic
|
|
approximation_degree: 1.0
|
|
target_backend: fake_brisbane
|
|
noise_backend: fake_brisbane
|
|
shots: 512
|
|
repeats: 3
|
|
notes: "Rosenfeld-direction incumbent — propagated from rung 4 winner."
|
|
|
|
search_space:
|
|
max_challengers_per_step: 4
|
|
dimensions:
|
|
verification: [z_only, both]
|
|
postselection: [z_only, all_measured]
|
|
ancilla_strategy: [reused_single, dedicated_pair]
|
|
optimization_level: [2, 3]
|
|
shots: [384, 512, 768]
|
|
|
|
tier_policy:
|
|
cheap_margin: 0.0005
|
|
confirmation_margin: 0.0
|
|
cheap_shots: 512
|
|
expensive_shots: 2048
|
|
cheap_repeats: 3
|
|
expensive_repeats: 2
|
|
promote_top_k: 1
|
|
enable_hardware: false
|
|
confirm_incumbent_on_hardware: true
|
|
hardware_budget: 1
|
|
|
|
score:
|
|
name: factory_throughput
|
|
base_cost: 1.0
|
|
cheap_quality:
|
|
noisy_fidelity: 0.10
|
|
logical_witness: 0.35
|
|
codespace_rate: 0.20
|
|
stability_score: 0.25
|
|
spectator_alignment: 0.10
|
|
expensive_quality:
|
|
logical_witness: 0.40
|
|
codespace_rate: 0.20
|
|
stability_score: 0.30
|
|
spectator_alignment: 0.10
|
|
cost_weights:
|
|
two_qubit_count: 0.12
|
|
depth: 0.02
|
|
shot_count: 0.00050
|
|
runtime_estimate: 0.04
|
|
queue_cost_proxy: 0.60
|
|
|
|
step_budget: 4
|
|
patience: 3
|
|
|
|
hardware:
|
|
backend_name:
|
|
channel:
|
|
instance:
|
|
token_env_var: QISKIT_IBM_TOKEN
|