Skip to content

Obelus-Labs-LLC/STRESS

Repository files navigation

STRESS — System Threat Resilience & Extreme Stress Suite (v0.2)

STRESS is a reliability benchmarking framework designed to evaluate how computational workloads behave when foundational operating assumptions are violated by environmental and systemic constraints. Unlike terrestrial benchmarks — which typically assume continuous power, stable connectivity, and rare environmental disruption — STRESS focuses on resilience and behavioral stability under persistent stress, rather than performance optimization, throughput, or cost efficiency.

Status

  • Specification: Frozen (v0.2)
  • Reference Implementation: Rust (stress-ref/) + Python (stress/)
  • Compliance: Binary

Quick Start

pip install -e .
stress-benchmark --workload W1-A --profile SP-1 --seed 42 --runs 3 --out-dir ./results

Or from Python:

from stress.runner import run_benchmark

run_benchmark(
    out_dir="./results", workload_id="W1-A", workload_version="0.2",
    stress_profile_id="SP-1", stress_parameters={"SP-1": {"rate": 0.001}},
    execution_environment={"runtime": "python"}, master_seed=42, n_runs=3,
    gds_levels=[0.1, 0.2, 0.3],
)

Full Specification

Repository Structure

Path Description
docs/specification-v0.2.md Normative specification
docs/STRESS_v0_Reference_Workloads.md W1-A, W2-A, W3-A workload definitions
docs/STRESS_v0_Reference_Stress_Profiles.md SP-0 through SP-5 profiles
docs/metric-independence.md GDS/ARR independence proof
docs/validation-methodology.md SRI correlation validation protocol
docs/glossary.md Term definitions
docs/historical/ Archived OCRB v0.1 specification
stress-ref/ Rust reference implementation (canonical, 34 tests)
stress/ Python reference implementation
tests/ Python test suite
examples/hello_benchmark.py Minimal "hello benchmark" walkthrough
examples/run_w1a_sp1.py W1-A with SP-1 radiation profile
examples/run_w3a_sp1.py W3-A distributed workload example
schema/run-record.schema.json Per-run JSON report schema
schema/aggregate.schema.json Aggregate statistics schema
schema/manifest.schema.json Benchmark manifest schema
STRESS_v0_Implementation_Guide.md Implementation reference
STRESS_FRAMEWORK_RESEARCH.md Background research and design rationale
CONTRIBUTING.md Contribution guidelines

What This Repo Is NOT

  • Not a performance benchmark
  • Not an optimization framework
  • Not adaptive or learning-based

Running a Benchmark

Rust (recommended)

cd stress-ref
cargo run -- --workload W1-A --profile SP-1 --seed 42 --runs 10 \
  --gds-levels 0.1,0.2,0.3 --isolation-duration 60.0 --c-total 5

Python CLI

pip install -e .
stress-benchmark --workload W1-A --profile SP-1 --seed 42 --runs 10 --out-dir report

SRI Aggregation

SRI uses the geometric mean of five behavioral proxies (GDS, ARR, IST, REC, CFR), each normalized to [0,1]:

SRI = (GDS × ARR × IST × REC × CFR)^(1/5) × 100

Geometric mean ensures zero resilience in any single dimension drives SRI toward 0 — a system cannot hide catastrophic failure behind strong scores elsewhere. GDS also reports monotonicity and smoothness metadata to distinguish graceful degradation from cliff-drops.

Domain-Specific SRI

STRESS supports domain-specific weighting profiles for SRI computation (weighted geometric mean):

Profile IST Weight CFR Weight Use Case
Equal (default) 0.20 0.20 General-purpose comparison
Satellite/LEO 0.35 0.10 Isolation survival matters most
Data Center 0.10 0.35 Cascade containment matters most
Tactical Edge 0.20 0.10 Recovery and degradation balanced

Version History

Version Date Description
v0.2 2026-03 STRESS — Current specification with SRI [0,100] scale
v0.1 (Archived) OCRB — Original specification with ORI [0,1] scale

Migration from OCRB v0.1

STRESS v0.2 supersedes OCRB (Orbital Compute Readiness Benchmark) v0.1.

OCRB v0.1 STRESS v0.2 Conversion
ORI [0, 1] SRI [0, 100] SRI = ORI * 100
0.85 threshold 85 threshold Direct mapping
Stress Regimes Stress Profiles Renamed
SR-1 to SR-5 SP-1 to SP-5 Renamed

Archives


Maintained by Obelus Labs, LLC


If this framework helped your research, consider giving it a star — it helps others discover it.

About

A benchmarking framework for evaluating computational resilience under environmental constraint violation.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors