Skip to content

README.md

Latest commit

 

History

History
44 lines (29 loc) · 1.71 KB

README.md

File metadata and controls

44 lines (29 loc) · 1.71 KB
Raw

Dual-rate Metric Depth Observer R1

Frozen protocol: DUAL_RATE_METRIC_DEPTH_OBSERVER_R1_PROTOCOL_2026-08-03.md

Run the focused tests:

E:\codex-tools\bin\blindassist-python.cmd -m unittest discover `
  -s scripts/research/dual_rate_metric_depth_observer_r1 `
  -p "test_*.py" -v

Run the hash-bound consumed Development replay once to a new ignored output:

E:\codex-tools\bin\blindassist-python.cmd `
  scripts/research/dual_rate_metric_depth_observer_r1/evaluate_r1.py `
  --output artifacts.local/evidence/hftf/dual-rate-metric-depth-observer-r1/result.json `
  --trace-output artifacts.local/evidence/hftf/dual-rate-metric-depth-observer-r1/trace.json

The evaluator refuses to overwrite either output. A/B/C are read from their frozen A0 reports. D is a causal async replay with a fixed robust affine fit and source-age UNKNOWN policy. The phone section is a scheduling/resource audit only because the measured phone DA asset is not the same metric checkpoint used by the PC quality arm.

状态：development

稳定 Interface

公开入口、输入不变量和失败模式以本目录脚本帮助和专项协议为准；跨域调用不得依赖私有 Implementation。

输出

只写入 artifacts.local/ 下的明确证据目录；不写仓库根目录或正式 App 资产。

安全边界

本模块不产生默认 App、生产、安全或 unseen confirmation authority；结果按当前协议声明的 Development/diagnostic 角色使用。

停止条件

最小判别实验完成、输入权威缺失、预算耗尽或重复失败时停止当前 evidence version，并保持最小 failure scope。