dumen 0.7.3__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- dumen-0.7.3/.github/ISSUE_TEMPLATE/bug_report.yml +48 -0
- dumen-0.7.3/.github/ISSUE_TEMPLATE/config.yml +8 -0
- dumen-0.7.3/.github/ISSUE_TEMPLATE/feature_request.yml +30 -0
- dumen-0.7.3/.github/PULL_REQUEST_TEMPLATE.md +20 -0
- dumen-0.7.3/.github/assets/avatar.png +0 -0
- dumen-0.7.3/.github/assets/favicon-128.png +0 -0
- dumen-0.7.3/.github/assets/favicon-16.png +0 -0
- dumen-0.7.3/.github/assets/favicon-180.png +0 -0
- dumen-0.7.3/.github/assets/favicon-32.png +0 -0
- dumen-0.7.3/.github/assets/favicon-48.png +0 -0
- dumen-0.7.3/.github/assets/favicon-512.png +0 -0
- dumen-0.7.3/.github/assets/favicon-64.png +0 -0
- dumen-0.7.3/.github/assets/favicon.ico +0 -0
- dumen-0.7.3/.github/assets/og.png +0 -0
- dumen-0.7.3/.github/workflows/ci.yml +37 -0
- dumen-0.7.3/.gitignore +14 -0
- dumen-0.7.3/CHANGELOG.md +119 -0
- dumen-0.7.3/CITATION.cff +21 -0
- dumen-0.7.3/CODE_OF_CONDUCT.md +28 -0
- dumen-0.7.3/CONTRIBUTING.md +48 -0
- dumen-0.7.3/LICENSE +201 -0
- dumen-0.7.3/PKG-INFO +274 -0
- dumen-0.7.3/README.md +235 -0
- dumen-0.7.3/README_TR.md +212 -0
- dumen-0.7.3/SECURITY.md +40 -0
- dumen-0.7.3/brand/LOGO_PROMPT.md +97 -0
- dumen-0.7.3/brand/REGEN_B.md +34 -0
- dumen-0.7.3/brand/SEcim_OLCUMLERI.md +19 -0
- dumen-0.7.3/brand/candidates/cell_A.png +0 -0
- dumen-0.7.3/brand/candidates/cell_B.png +0 -0
- dumen-0.7.3/brand/candidates/cell_C.png +0 -0
- dumen-0.7.3/brand/candidates/cell_D.png +0 -0
- dumen-0.7.3/brand/candidates/cell_E.png +0 -0
- dumen-0.7.3/brand/candidates/cell_a1.png +0 -0
- dumen-0.7.3/brand/candidates/cell_a2.png +0 -0
- dumen-0.7.3/brand/candidates/cell_a3.png +0 -0
- dumen-0.7.3/brand/candidates/cell_a4.png +0 -0
- dumen-0.7.3/brand/candidates/cell_a5.png +0 -0
- dumen-0.7.3/brand/candidates/cell_/303/2741.png +0 -0
- dumen-0.7.3/brand/candidates/cell_/303/2742.png +0 -0
- dumen-0.7.3/brand/candidates/cell_/303/2743.png +0 -0
- dumen-0.7.3/brand/candidates/cell_/303/2744.png +0 -0
- dumen-0.7.3/brand/candidates/cell_/303/2745.png +0 -0
- dumen-0.7.3/brand/candidates/cell_/303/2746.png +0 -0
- dumen-0.7.3/brand/candidates/cell_/303/2747.png +0 -0
- dumen-0.7.3/brand/dumen-mark.svg +48 -0
- dumen-0.7.3/brand/logo.png +0 -0
- dumen-0.7.3/brand/logo_B_hi.png +0 -0
- dumen-0.7.3/dumen/__init__.py +96 -0
- dumen-0.7.3/dumen/benchmarks/__init__.py +28 -0
- dumen-0.7.3/dumen/benchmarks/agentharm_loader.py +77 -0
- dumen-0.7.3/dumen/benchmarks/capability_gate.py +152 -0
- dumen-0.7.3/dumen/benchmarks/gateway_selfredteam.py +148 -0
- dumen-0.7.3/dumen/benchmarks/harmbench_loader.py +71 -0
- dumen-0.7.3/dumen/benchmarks/jailbreakbench_loader.py +250 -0
- dumen-0.7.3/dumen/benchmarks/judge_calibration.py +171 -0
- dumen-0.7.3/dumen/benchmarks/sae_quality.py +163 -0
- dumen-0.7.3/dumen/benchmarks/seeds.py +480 -0
- dumen-0.7.3/dumen/benchmarks/steering_efficacy.py +95 -0
- dumen-0.7.3/dumen/benchmarks/steering_overhead.py +149 -0
- dumen-0.7.3/dumen/cli.py +525 -0
- dumen-0.7.3/dumen/core/__init__.py +38 -0
- dumen-0.7.3/dumen/core/hooks.py +111 -0
- dumen-0.7.3/dumen/core/kv_drift.py +99 -0
- dumen-0.7.3/dumen/core/miner.py +355 -0
- dumen-0.7.3/dumen/core/ov_circuits.py +129 -0
- dumen-0.7.3/dumen/core/quantization.py +111 -0
- dumen-0.7.3/dumen/core/sae_engine.py +141 -0
- dumen-0.7.3/dumen/core/serialization.py +142 -0
- dumen-0.7.3/dumen/core/steering.py +307 -0
- dumen-0.7.3/dumen/core/transcoder.py +192 -0
- dumen-0.7.3/dumen/core/types.py +160 -0
- dumen-0.7.3/dumen/gateway/__init__.py +16 -0
- dumen-0.7.3/dumen/gateway/filters.py +152 -0
- dumen-0.7.3/dumen/gateway/proxy.py +212 -0
- dumen-0.7.3/dumen/gateway/validator.py +180 -0
- dumen-0.7.3/dumen/redteam/__init__.py +21 -0
- dumen-0.7.3/dumen/redteam/api_runner.py +111 -0
- dumen-0.7.3/dumen/redteam/hrl_engine.py +211 -0
- dumen-0.7.3/dumen/redteam/inspect_adapter.py +147 -0
- dumen-0.7.3/dumen/redteam/judge.py +252 -0
- dumen-0.7.3/dumen/reports/__init__.py +40 -0
- dumen-0.7.3/dumen/reports/annex_xi.py +509 -0
- dumen-0.7.3/dumen/reports/cop_commitments.py +167 -0
- dumen-0.7.3/dumen/reports/eu_ai_act.py +119 -0
- dumen-0.7.3/dumen/reports/evidence_chain.py +224 -0
- dumen-0.7.3/dumen/reports/incident_report.py +116 -0
- dumen-0.7.3/dumen/reports/scorecard.py +139 -0
- dumen-0.7.3/examples/README.md +16 -0
- dumen-0.7.3/examples/audits/Qwen2.5-0.5B-Instruct_audit.json +56 -0
- dumen-0.7.3/examples/audits/Qwen2.5-0.5B-Instruct_run.log +51 -0
- dumen-0.7.3/examples/audits/README.md +58 -0
- dumen-0.7.3/examples/audits/gateway_selfredteam_qwen2.5-3b.json +160 -0
- dumen-0.7.3/examples/audits/llama3.2-3b_ollama_jbb40.json +44 -0
- dumen-0.7.3/examples/audits/llama3.2-3b_ollama_run.log +101 -0
- dumen-0.7.3/examples/audits/llama3.2-3b_ollama_standard.json +45 -0
- dumen-0.7.3/examples/audits/multifamily_ollama_run.log +275 -0
- dumen-0.7.3/examples/audits/phi3-mini_ollama_jbb10.json +42 -0
- dumen-0.7.3/examples/audits/phi3-mini_ollama_jbb10_run.log +46 -0
- dumen-0.7.3/examples/audits/phi3-mini_ollama_run.log +55 -0
- dumen-0.7.3/examples/audits/phi3-mini_ollama_standard.json +45 -0
- dumen-0.7.3/examples/audits/qwen2.5-3b_ollama_jbb40.json +43 -0
- dumen-0.7.3/examples/audits/qwen2.5-3b_ollama_jbb40.log +46 -0
- dumen-0.7.3/examples/audits/qwen2.5-3b_ollama_standard.json +44 -0
- dumen-0.7.3/examples/audits/qwen2.5-3b_ollama_standard.md +46 -0
- dumen-0.7.3/examples/calibration_seed.py +99 -0
- dumen-0.7.3/examples/datasets/README.md +35 -0
- dumen-0.7.3/examples/datasets/jbb_harmful_behaviors.csv +101 -0
- dumen-0.7.3/examples/full_audit_pipeline.py +233 -0
- dumen-0.7.3/examples/pilot/README.md +72 -0
- dumen-0.7.3/examples/pilot/evidence_chain.json +56 -0
- dumen-0.7.3/examples/pilot/sample_cop.md +16 -0
- dumen-0.7.3/examples/pilot/sample_dossier.json +61 -0
- dumen-0.7.3/examples/pilot/sample_dossier.md +94 -0
- dumen-0.7.3/examples/pilot_dossier.py +128 -0
- dumen-0.7.3/examples/redteam_gateway_self.py +176 -0
- dumen-0.7.3/pyproject.toml +88 -0
- dumen-0.7.3/scripts/dist_hygiene.py +65 -0
- dumen-0.7.3/scripts/make_brand_assets.py +154 -0
- dumen-0.7.3/scripts/surface_parity.py +33 -0
- dumen-0.7.3/tests/test_adaptive_steering.py +87 -0
- dumen-0.7.3/tests/test_annex_xi.py +337 -0
- dumen-0.7.3/tests/test_api_runner.py +215 -0
- dumen-0.7.3/tests/test_benchmarks.py +234 -0
- dumen-0.7.3/tests/test_capability_gate.py +214 -0
- dumen-0.7.3/tests/test_cli.py +167 -0
- dumen-0.7.3/tests/test_cop_commitments.py +116 -0
- dumen-0.7.3/tests/test_coverage_gaps.py +407 -0
- dumen-0.7.3/tests/test_coverage_gaps_v060.py +333 -0
- dumen-0.7.3/tests/test_evidence_chain.py +152 -0
- dumen-0.7.3/tests/test_example_pipeline.py +67 -0
- dumen-0.7.3/tests/test_gateway.py +119 -0
- dumen-0.7.3/tests/test_gateway_selfredteam.py +176 -0
- dumen-0.7.3/tests/test_harmbench_agentharm_loaders.py +130 -0
- dumen-0.7.3/tests/test_hooks.py +62 -0
- dumen-0.7.3/tests/test_hybrid_judge.py +61 -0
- dumen-0.7.3/tests/test_incident_report.py +114 -0
- dumen-0.7.3/tests/test_inspect_adapter.py +66 -0
- dumen-0.7.3/tests/test_jailbreakbench_loader.py +187 -0
- dumen-0.7.3/tests/test_judge_calibration.py +190 -0
- dumen-0.7.3/tests/test_kv_drift.py +62 -0
- dumen-0.7.3/tests/test_latency_bench.py +75 -0
- dumen-0.7.3/tests/test_miner.py +110 -0
- dumen-0.7.3/tests/test_ov_circuits.py +65 -0
- dumen-0.7.3/tests/test_pair_engine.py +48 -0
- dumen-0.7.3/tests/test_proxy_sse_live.py +120 -0
- dumen-0.7.3/tests/test_quantization.py +112 -0
- dumen-0.7.3/tests/test_rank_k.py +216 -0
- dumen-0.7.3/tests/test_real_model_integration.py +290 -0
- dumen-0.7.3/tests/test_reports.py +89 -0
- dumen-0.7.3/tests/test_sae_engine.py +51 -0
- dumen-0.7.3/tests/test_sae_quality.py +122 -0
- dumen-0.7.3/tests/test_serialization.py +85 -0
- dumen-0.7.3/tests/test_steering.py +130 -0
- dumen-0.7.3/tests/test_steering_efficacy.py +109 -0
- dumen-0.7.3/tests/test_steering_overhead.py +117 -0
- dumen-0.7.3/tests/test_transcoder.py +78 -0
- dumen-0.7.3/tests/test_types.py +58 -0
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
name: Bug report
|
|
2
|
+
description: A reproducible failure — commands, versions, output included
|
|
3
|
+
labels: [bug]
|
|
4
|
+
body:
|
|
5
|
+
- type: markdown
|
|
6
|
+
attributes:
|
|
7
|
+
value: |
|
|
8
|
+
This project's doctrine is **"no evidence, no claim"** — it applies to
|
|
9
|
+
bug reports too. "It doesn't work" is not actionable; the exact command
|
|
10
|
+
and its exact output are.
|
|
11
|
+
- type: textarea
|
|
12
|
+
id: repro
|
|
13
|
+
attributes:
|
|
14
|
+
label: Reproduction command
|
|
15
|
+
description: The exact command line that fails (paste it, don't paraphrase)
|
|
16
|
+
placeholder: |
|
|
17
|
+
dumen audit --model X --endpoint http://127.0.0.1:11434/v1 --request-timeout 300
|
|
18
|
+
render: shell
|
|
19
|
+
validations:
|
|
20
|
+
required: true
|
|
21
|
+
- type: textarea
|
|
22
|
+
id: actual
|
|
23
|
+
attributes:
|
|
24
|
+
label: Actual output
|
|
25
|
+
description: Full traceback or log tail (trim secrets, keep line numbers)
|
|
26
|
+
render: shell
|
|
27
|
+
validations:
|
|
28
|
+
required: true
|
|
29
|
+
- type: textarea
|
|
30
|
+
id: expected
|
|
31
|
+
attributes:
|
|
32
|
+
label: Expected behavior
|
|
33
|
+
validations:
|
|
34
|
+
required: true
|
|
35
|
+
- type: input
|
|
36
|
+
id: version
|
|
37
|
+
attributes:
|
|
38
|
+
label: Version
|
|
39
|
+
description: output of `dumen --version`
|
|
40
|
+
validations:
|
|
41
|
+
required: true
|
|
42
|
+
- type: input
|
|
43
|
+
id: env
|
|
44
|
+
attributes:
|
|
45
|
+
label: Environment
|
|
46
|
+
description: Python version, OS, CPU/GPU, whether torch is CUDA or CPU build
|
|
47
|
+
validations:
|
|
48
|
+
required: true
|
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
blank_issues_enabled: true
|
|
2
|
+
contact_links:
|
|
3
|
+
- name: Security vulnerability
|
|
4
|
+
url: https://github.com/goun7/Dumen/security/advisories/new
|
|
5
|
+
about: Privately report product-security issues (gateway bypass, evidence-chain tampering, PII masking leaks)
|
|
6
|
+
- name: Model audit results / red-team findings
|
|
7
|
+
url: https://github.com/goun7/Dumen/issues/new?template=feature_request.yml
|
|
8
|
+
about: Jailbreak prompts and model weaknesses belong in the model, not in this tracker — audit artifacts live in examples/audits/
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
name: Feature request
|
|
2
|
+
description: A capability the evidence pipeline is missing
|
|
3
|
+
labels: [enhancement]
|
|
4
|
+
body:
|
|
5
|
+
- type: textarea
|
|
6
|
+
id: problem
|
|
7
|
+
attributes:
|
|
8
|
+
label: Problem this solves
|
|
9
|
+
description: What were you auditing and what could the tool not (yet) prove?
|
|
10
|
+
validations:
|
|
11
|
+
required: true
|
|
12
|
+
- type: textarea
|
|
13
|
+
id: proposal
|
|
14
|
+
attributes:
|
|
15
|
+
label: Proposed shape
|
|
16
|
+
description: CLI flag / API / report field — as concrete as you can make it
|
|
17
|
+
validations:
|
|
18
|
+
required: true
|
|
19
|
+
- type: dropdown
|
|
20
|
+
id: doctrine
|
|
21
|
+
attributes:
|
|
22
|
+
label: Which honesty rule does it touch?
|
|
23
|
+
options:
|
|
24
|
+
- "None — purely additive"
|
|
25
|
+
- "Measured-only metrics (must render None when unmeasured)"
|
|
26
|
+
- "License/data policy (raw licensed corpora stay out of the repo)"
|
|
27
|
+
- "White-box vs black-box channel separation"
|
|
28
|
+
- "Not sure — discuss"
|
|
29
|
+
validations:
|
|
30
|
+
required: true
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
<!-- Doctrine applies to PRs as it does to reports: no evidence, no claim. -->
|
|
2
|
+
|
|
3
|
+
## What & why (one theme per PR)
|
|
4
|
+
|
|
5
|
+
- [ ] Single theme; commit prefix `feat|fix|docs|test|perf|security(scope):`
|
|
6
|
+
- [ ] User-visible behavior change → README/CHANGELOG lines updated in the same PR
|
|
7
|
+
|
|
8
|
+
## Evidence
|
|
9
|
+
|
|
10
|
+
- [ ] `python3 -m pytest tests/ -q --cov=dumen --cov-fail-under=95` green (paste the
|
|
11
|
+
final line — test count + coverage %, measured not remembered)
|
|
12
|
+
- [ ] `ruff check dumen/ tests/ examples/ scripts/` clean
|
|
13
|
+
- [ ] New behavior tested through a real socket/CLI/file path — **no mocks**
|
|
14
|
+
- [ ] Claims of the "no evidence, no claim" kind: unmeasured metrics render as
|
|
15
|
+
"Ölçülmedi"/None, never a placeholder number
|
|
16
|
+
|
|
17
|
+
## Published audits (if any new artifact)
|
|
18
|
+
|
|
19
|
+
- [ ] Reproduction command + environment noted inside `*_run.log`
|
|
20
|
+
- [ ] Raw licensed/harmful payloads stay OUT of the repo (cache-only policy)
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
@@ -0,0 +1,37 @@
|
|
|
1
|
+
name: CI
|
|
2
|
+
|
|
3
|
+
on:
|
|
4
|
+
push:
|
|
5
|
+
branches: [main]
|
|
6
|
+
pull_request:
|
|
7
|
+
|
|
8
|
+
jobs:
|
|
9
|
+
test:
|
|
10
|
+
runs-on: ubuntu-latest
|
|
11
|
+
strategy:
|
|
12
|
+
matrix:
|
|
13
|
+
python-version: ["3.10", "3.12", "3.14"]
|
|
14
|
+
steps:
|
|
15
|
+
- uses: actions/checkout@v4
|
|
16
|
+
- uses: actions/setup-python@v5
|
|
17
|
+
with:
|
|
18
|
+
python-version: ${{ matrix.python-version }}
|
|
19
|
+
- name: Install dependencies
|
|
20
|
+
run: |
|
|
21
|
+
python -m pip install --upgrade pip
|
|
22
|
+
pip install torch --index-url https://download.pytorch.org/whl/cpu
|
|
23
|
+
pip install -e ".[dev]"
|
|
24
|
+
- name: Real-model extra (3.12 job)
|
|
25
|
+
if: matrix.python-version == '3.12'
|
|
26
|
+
run: pip install -e ".[model]"
|
|
27
|
+
- name: Lint (ruff)
|
|
28
|
+
run: ruff check dumen/ tests/ examples/ scripts/
|
|
29
|
+
- name: Surface language parity (README EN ↔ TR numeric tokens)
|
|
30
|
+
run: python scripts/surface_parity.py
|
|
31
|
+
- name: Tests (3.10/3.14: skip real-model; 3.12: include real-model integration)
|
|
32
|
+
run: |
|
|
33
|
+
if [ "${{ matrix.python-version }}" = "3.12" ]; then
|
|
34
|
+
python -m pytest tests/ -q --cov=dumen --cov-report=term --cov-fail-under=95
|
|
35
|
+
else
|
|
36
|
+
python -m pytest tests/ -q --ignore=tests/test_real_model_integration.py
|
|
37
|
+
fi
|
dumen-0.7.3/.gitignore
ADDED
dumen-0.7.3/CHANGELOG.md
ADDED
|
@@ -0,0 +1,119 @@
|
|
|
1
|
+
# Changelog
|
|
2
|
+
|
|
3
|
+
Format: Keep a Changelog · This project uses SemVer. The *evidence* for each
|
|
4
|
+
release lives in `examples/audits/` as reproducible artifacts.
|
|
5
|
+
|
|
6
|
+
## [Unreleased]
|
|
7
|
+
|
|
8
|
+
## [0.7.3] - 2026-09-15
|
|
9
|
+
|
|
10
|
+
Brand identity + public surface + release hygiene.
|
|
11
|
+
|
|
12
|
+
- **Brand mark (helm-"D")**: hi-res generated mark → potrace single-path SVG
|
|
13
|
+
(IoU 0.994 ≥ 0.99 gate); og.png/avatar/favicon family generated by
|
|
14
|
+
`scripts/make_brand_assets.py`; mark embedded in both README surfaces.
|
|
15
|
+
- **Surface language alignment**: English-primary README + full Turkish sibling
|
|
16
|
+
with a numeric-parity CI gate (`scripts/surface_parity.py`); CONTRIBUTING/
|
|
17
|
+
SECURITY/CHANGELOG migrated to English (zero numeric drift, multiset-diff).
|
|
18
|
+
- **Community kit**: issue/PR templates, CODE_OF_CONDUCT, CITATION.cff.
|
|
19
|
+
- **Release hygiene gates**: sdist whitelist + `scripts/dist_hygiene.py`
|
|
20
|
+
(8 internal docs were entering the tarball — caught pre-upload); public-clone
|
|
21
|
+
masker v4 revived and proof-fired (liveness 146→0 / path-sweep 9→0); full
|
|
22
|
+
sdist rehearsal: fresh venv install → CLI trio → 24 bundled tests green.
|
|
23
|
+
- No functional API changes; CLI flags, report schema untouched.
|
|
24
|
+
|
|
25
|
+
## [0.7.2] - 2026-09-15
|
|
26
|
+
|
|
27
|
+
Fresh-environment verification and the first sales-grade sample, pre-publication.
|
|
28
|
+
|
|
29
|
+
- **Packaging fix (fresh-venv smoke-test finding):** `numpy` is now declared as
|
|
30
|
+
a direct (not transitive) dependency — the torch "Failed to initialize NumPy"
|
|
31
|
+
first-run warning on new installs is gone (measured: fresh env 5.4GB,
|
|
32
|
+
refusal-baseline audit **5.1s**, CLI trio smoke clean). Honest
|
|
33
|
+
"install weight" note added to the README.
|
|
34
|
+
- **Sales sample (`examples/pilot_dossier.py` + `examples/pilot/`):** a full
|
|
35
|
+
Annex XI dossier + CoP matrix (62% — the two missing rows are deliberate:
|
|
36
|
+
customer declaration + incident pipeline) + SHA-256 evidence chain sealing
|
|
37
|
+
the source scorecard's hash, generated from a REAL PUBLISHED scorecard
|
|
38
|
+
(qwen2.5:3b JBB-40, safety 91.8). The compute field is LABELLED as a
|
|
39
|
+
6·N·D order-of-magnitude estimate (3.34e23 < 1e25 → below the Art. 3(63)
|
|
40
|
+
threshold; D=18T verified live from arXiv:2412.15115); every unverifiable
|
|
41
|
+
field stays a bracketed [pending declaration].
|
|
42
|
+
- No API/schema changes — the sole SemVer reason is the packaging finding.
|
|
43
|
+
|
|
44
|
+
## [0.7.1] - 2026-09-15 — B1 capability-externality gate + B5 multi-family publications
|
|
45
|
+
|
|
46
|
+
### Added
|
|
47
|
+
- **B1 `CapabilityGate`**: steering damage to model CAPABILITY is now measured —
|
|
48
|
+
12 deterministically-verifiable tasks (no LLM judge, no circular evidence;
|
|
49
|
+
echo-safety test-enforced). pass/fail/inconclusive; **a fail retracts the
|
|
50
|
+
protection claim from CLI and Annex XI** (an intervention that harms is not
|
|
51
|
+
an intervention). First live pair: Qwen2.5-0.5B efficacy %0 + capability
|
|
52
|
+
PASS (%83.3→%83.3).
|
|
53
|
+
- `--request-timeout` (audit; default 300s) — field finding: on single-VRAM
|
|
54
|
+
machines, cold model load + slow generation blew the 180s budget
|
|
55
|
+
(phi3:mini/GTX-1070, ~1s/token). Timeouts still raise a clean
|
|
56
|
+
`EndpointError`; no fake refusals are produced.
|
|
57
|
+
- **B5 multi-family publications** (`examples/audits/` + comparison table):
|
|
58
|
+
llama3.2:3b (std 77.5 · JBB-40 **91.3**) and phi3:mini (std 95.0 · JBB-10
|
|
59
|
+
97.0 — n difference flagged in the table) — with qwen2.5:3b JBB-40 91.8,
|
|
60
|
+
same-depth inter-family consistency was MEASURED; llama std surfaced a
|
|
61
|
+
cyber %60 REAL weakness finding.
|
|
62
|
+
- **B3 process tool**: `examples/calibration_seed.py` — generates a human
|
|
63
|
+
second-label B3 worksheet from real (prompt, response) pairs; first live
|
|
64
|
+
output: llama3.2:3b × 20 pairs (3 refusal / 17 mixed — the hard zone for
|
|
65
|
+
the fastpath). Raw pairs stay out of the repo (dual-use policy); only the
|
|
66
|
+
distribution is published.
|
|
67
|
+
### Unchanged
|
|
68
|
+
- Black-box/refusal-baseline channels, API surface, report schema (only the
|
|
69
|
+
`capability_regression` field was added — old artifacts stay schema-safe).
|
|
70
|
+
|
|
71
|
+
## [0.7.0] - 2026-09-15 — API-endpoint auditing + real-catalog red-teaming
|
|
72
|
+
|
|
73
|
+
### Added
|
|
74
|
+
- **Black-box API audit channel** (`dumen audit --endpoint`): audit any
|
|
75
|
+
Ollama / vLLM / LM Studio / OpenAI-compatible server; deterministic
|
|
76
|
+
temperature=0; HTTP errors never become FAKE REFUSALS (EndpointError raises).
|
|
77
|
+
- **First local-real-model audit published**: Ollama qwen2.5:3b — standard-suite
|
|
78
|
+
scorecard (`safety 58.8`, sandbox %95) + 40-task JBB wide audit (`safety
|
|
79
|
+
91.8`, %0 jailbreak) — `examples/audits/qwen2.5-3b_ollama_*.json`.
|
|
80
|
+
- **Gateway self-red-team comparison** (`GatewaySelfRedTeam`): two-layer
|
|
81
|
+
(regex ∪ semantic LLM-judge) recall/FPR matrices on the published
|
|
82
|
+
deepset/prompt-injections corpus; regex patterns hardened with EN+DE
|
|
83
|
+
families (holdout FPR %0, precision %100); `JudgeEvaluator.classify_injection`.
|
|
84
|
+
- **External attack-catalog bridges**: HarmBenchLoader (400 real behaviors) +
|
|
85
|
+
AgentHarmLoader (176 agentic tasks) + `--dataset` CLI with automatic schema
|
|
86
|
+
detection (JBB CSV · HarmBench CSV · AgentHarm JSON · AILuminate JSON/JSONL).
|
|
87
|
+
### Fixed
|
|
88
|
+
- JBB/AILuminate `CATEGORY_MAP` calibrated to real distribution labels
|
|
89
|
+
(Malware/Hacking→cyber_attack, Disinformation→hallucination, Privacy→pii_leak).
|
|
90
|
+
|
|
91
|
+
## [0.6.1] - 2026-09-15 — evidence-integrity release
|
|
92
|
+
|
|
93
|
+
### Fixed (deep-audit findings)
|
|
94
|
+
- Fabricated success numbers purged: "%96.2 Active Protection" → three-state
|
|
95
|
+
measured efficacy (Not measured ⚪ / %0 🟠 / +%X 🟢) via behavioral
|
|
96
|
+
`SteeringEfficacyBench` (before/after judge comparison; cosmetic cosine
|
|
97
|
+
rejected).
|
|
98
|
+
- `--measure-steering` became a real measurement pipeline (mining + live hook +
|
|
99
|
+
steered generation).
|
|
100
|
+
- vLLM/sub-1ms/Anthropic-native/JBB-live claims → verified text only; SLA claim
|
|
101
|
+
replaced by a MEASURED latency bench (`test_latency_bench.py`).
|
|
102
|
+
- 3 wrong academic citations corrected (PAIR/TAP; RepE↔StMP; 2606.13720 title);
|
|
103
|
+
unverifiable quotes softened; EU timeline updated with EC-sourced dates.
|
|
104
|
+
- dossier/serve channels: refusal-baseline stamp + `evidence_channel` + chain
|
|
105
|
+
head rendered inside Annex XI; context note for Art.14 ❌.
|
|
106
|
+
### Infrastructure
|
|
107
|
+
- CI coverage gate made actually effective (pytest-cov was missing); 3.12 job
|
|
108
|
+
runs real-model tests; unused dependencies deleted; AILuminate format bridge;
|
|
109
|
+
threat-model & limitations section (§6.5).
|
|
110
|
+
|
|
111
|
+
## [0.6.0] - 2026-09-14 — Evidence-gap closure
|
|
112
|
+
- rank-k subspace mining + bootstrap confidence + permutation p; joint-nullspace
|
|
113
|
+
steering; InspectBridge red-team; evidence-chain reports.
|
|
114
|
+
|
|
115
|
+
## [0.5.0] - 2026-09-13 — mechanistic-core verification
|
|
116
|
+
- OV-circuit mask, output-filter critical bug fix, unified StTP/StMP pipeline.
|
|
117
|
+
|
|
118
|
+
## [0.4.0] and earlier — prototype
|
|
119
|
+
- basic mining/hook/report pipeline; test count between 52–173.
|
dumen-0.7.3/CITATION.cff
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
cff-version: 1.2.0
|
|
2
|
+
message: "If you use Dümen in published research, please cite the software repository."
|
|
3
|
+
title: "Dümen: Open-source EU AI Act audit engine with white-box steering evidence and tamper-evident compliance dossiers"
|
|
4
|
+
authors:
|
|
5
|
+
- name: "Dümen Contributors"
|
|
6
|
+
repository-code: "https://github.com/goun7/Dumen"
|
|
7
|
+
license: Apache-2.0
|
|
8
|
+
version: 0.7.3
|
|
9
|
+
date-released: 2026-09-15
|
|
10
|
+
keywords:
|
|
11
|
+
- eu-ai-act
|
|
12
|
+
- llm-auditing
|
|
13
|
+
- activation-steering
|
|
14
|
+
- red-teaming
|
|
15
|
+
- interpretability
|
|
16
|
+
preferred-citation:
|
|
17
|
+
type: software
|
|
18
|
+
title: "Dümen"
|
|
19
|
+
authors:
|
|
20
|
+
- name: "Dümen Contributors"
|
|
21
|
+
url: "https://github.com/goun7/Dumen"
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
# Code of Conduct
|
|
2
|
+
|
|
3
|
+
## Our commitment
|
|
4
|
+
|
|
5
|
+
Contributors and maintainers of this project pledge to make participation
|
|
6
|
+
harassment-free for everyone, regardless of the usual categories. This is an
|
|
7
|
+
audit tool whose subject matter (AI safety) attracts strong opinions — the
|
|
8
|
+
doctrine applies to conduct as it does to claims: **critique the evidence,
|
|
9
|
+
never the person.**
|
|
10
|
+
|
|
11
|
+
## Standards
|
|
12
|
+
|
|
13
|
+
Expected: cite reproducible commands when disagreeing with a finding; assume
|
|
14
|
+
good faith for first-time contributors; keep discussions in issue/PR threads
|
|
15
|
+
rather than personal channels.
|
|
16
|
+
|
|
17
|
+
Unacceptable: ad hominem attacks, doxxing, spam-driven "safety claims" without
|
|
18
|
+
reproduction steps, and attempts to use this tracker to distribute raw
|
|
19
|
+
harmful payloads (use coordinated disclosure instead — see SECURITY.md).
|
|
20
|
+
|
|
21
|
+
## Enforcement
|
|
22
|
+
|
|
23
|
+
Report to the maintainer via GitHub private vulnerability reporting channel or
|
|
24
|
+
a private issue. Remedies scale from edit → warning → temporary → permanent
|
|
25
|
+
exclusion, applied by the maintainer, with reasons logged.
|
|
26
|
+
|
|
27
|
+
Adapted from the Contributor Covenant 2.1 (https://www.contributor-covenant.org),
|
|
28
|
+
CC BY-SA 4.0.
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
# Contributing guide
|
|
2
|
+
|
|
3
|
+
Evidence is mandatory for contributions to Dümen — code is not. The project
|
|
4
|
+
doctrine is: ***no evidence, no claim.*** This applies to contributions too.
|
|
5
|
+
|
|
6
|
+
## Development environment
|
|
7
|
+
|
|
8
|
+
```bash
|
|
9
|
+
python -m venv .venv && source .venv/bin/activate
|
|
10
|
+
pip install -e ".[dev]" # test + lint tooling
|
|
11
|
+
pip install -e ".[model]" # (opt.) transformers, for real-model integration tests
|
|
12
|
+
python -m pytest tests/ -q # run the suite
|
|
13
|
+
ruff check dumen/ tests/ examples/ scripts/
|
|
14
|
+
python scripts/surface_parity.py # README EN↔TR numeric-token parity gate
|
|
15
|
+
```
|
|
16
|
+
|
|
17
|
+
Python 3.10–3.14 is supported; CI runs three versions (real-model tests
|
|
18
|
+
included on 3.12; CI also runs the lint scope above, the parity gate, and fails under coverage %95).
|
|
19
|
+
|
|
20
|
+
## Rules
|
|
21
|
+
|
|
22
|
+
1. **No fake numbers.** An unmeasured metric stays `None`/"Not measured" in
|
|
23
|
+
the report; a hard-coded success number never reaches production
|
|
24
|
+
(see `steering_efficacy`).
|
|
25
|
+
2. **No behavior without a test.** Every new production branch is exercised
|
|
26
|
+
through a real socket/CLI/file path — real mechanisms over mock servers
|
|
27
|
+
(e.g. stdlib threading HTTP server, see `test_api_runner.py`).
|
|
28
|
+
3. **Citations verified.** arXiv ID + title + year must be confirmed against
|
|
29
|
+
the primary source; a citation correction is called out separately in the PR.
|
|
30
|
+
4. **External data licenses.** Raw licensed/harmful corpora (deepset CC-BY-NC,
|
|
31
|
+
AgentHarm "other") **never enter the repo**; only MIT/Apache distributions
|
|
32
|
+
get committed (JBB MIT ✓). Loaders pin the schema; the data stays with the
|
|
33
|
+
user.
|
|
34
|
+
5. **Scope honesty.** White-box (activation) and black-box (API) channels are
|
|
35
|
+
never conflated; efficacy measurement is only possible white-box.
|
|
36
|
+
|
|
37
|
+
## PR process
|
|
38
|
+
|
|
39
|
+
- One theme, one commit (prefixes `feat|fix|docs|test|perf|security(scope)`).
|
|
40
|
+
- Gates: full suite green + coverage ≥ %95 + `ruff` clean + surface parity (numbers never drift between README languages).
|
|
41
|
+
- New CLI flag: user-facing text may mix Turkish and English; error messages
|
|
42
|
+
must be **reproducible** (say what you tried).
|
|
43
|
+
|
|
44
|
+
## Report quality
|
|
45
|
+
|
|
46
|
+
Published audits under `examples/audits/` are **transparent evidence**: every
|
|
47
|
+
shipped artifact carries its reproduction command inside `_run.log`. If you
|
|
48
|
+
publish a new audit, add the command and the environment too.
|
dumen-0.7.3/LICENSE
ADDED
|
@@ -0,0 +1,201 @@
|
|
|
1
|
+
Apache License
|
|
2
|
+
Version 2.0, January 2004
|
|
3
|
+
http://www.apache.org/licenses/
|
|
4
|
+
|
|
5
|
+
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
|
|
6
|
+
|
|
7
|
+
1. Definitions.
|
|
8
|
+
|
|
9
|
+
"License" shall mean the terms and conditions for use, reproduction,
|
|
10
|
+
and distribution as defined by Sections 1 through 9 of this document.
|
|
11
|
+
|
|
12
|
+
"Licensor" shall mean the copyright owner or entity authorized by
|
|
13
|
+
the copyright owner that is granting the License.
|
|
14
|
+
|
|
15
|
+
"Legal Entity" shall mean the union of the acting entity and all
|
|
16
|
+
other entities that control, are controlled by, or are under common
|
|
17
|
+
control with that entity. For the purposes of this definition,
|
|
18
|
+
"control" means (i) the power, direct or indirect, to cause the
|
|
19
|
+
direction or management of such entity, whether by contract or
|
|
20
|
+
otherwise, or (ii) ownership of fifty percent (50%) or more of the
|
|
21
|
+
outstanding shares, or (iii) beneficial ownership of such entity.
|
|
22
|
+
|
|
23
|
+
"You" (or "Your") shall mean an individual or Legal Entity
|
|
24
|
+
exercising permissions granted by this License.
|
|
25
|
+
|
|
26
|
+
"Source" form shall mean the preferred form for making modifications,
|
|
27
|
+
including but not limited to software source code, documentation
|
|
28
|
+
source, and configuration files.
|
|
29
|
+
|
|
30
|
+
"Object" form shall mean any form resulting from mechanical
|
|
31
|
+
transformation or translation of a Source form, including but
|
|
32
|
+
not limited to compiled object code, generated documentation,
|
|
33
|
+
and conversions to other media types.
|
|
34
|
+
|
|
35
|
+
"Work" shall mean the work of authorship, whether in Source or
|
|
36
|
+
Object form, made available under the License, as indicated by a
|
|
37
|
+
copyright notice that is included in or attached to the work
|
|
38
|
+
(an example is provided in the Appendix below).
|
|
39
|
+
|
|
40
|
+
"Derivative Works" shall mean any work, whether in Source or Object
|
|
41
|
+
form, that is based on (or derived from) the Work and for which the
|
|
42
|
+
editorial revisions, annotations, elaborations, or other modifications
|
|
43
|
+
represent, as a whole, an original work of authorship. For the purposes
|
|
44
|
+
of this License, Derivative Works shall not include works that remain
|
|
45
|
+
separable from, or merely link (or bind by name) to the interfaces of,
|
|
46
|
+
the Work and Derivative Works thereof.
|
|
47
|
+
|
|
48
|
+
"Contribution" shall mean any work of authorship, including
|
|
49
|
+
the original version of the Work and any modifications or additions
|
|
50
|
+
to that Work or Derivative Works thereof, that is intentionally
|
|
51
|
+
submitted to Licensor for inclusion in the Work by the copyright owner
|
|
52
|
+
or by an individual or Legal Entity authorized to submit on behalf of
|
|
53
|
+
the copyright owner. For the purposes of this definition, "submitted"
|
|
54
|
+
means any form of electronic, verbal, or written communication sent
|
|
55
|
+
to the Licensor or its representatives, including but not limited to
|
|
56
|
+
communication on electronic mailing lists, source code control systems,
|
|
57
|
+
and issue tracking systems that are managed by, or on behalf of, the
|
|
58
|
+
Licensor for the purpose of discussing and improving the Work, but
|
|
59
|
+
excluding communication that is conspicuously marked or otherwise
|
|
60
|
+
designated in writing by the copyright owner as "Not a Contribution."
|
|
61
|
+
|
|
62
|
+
"Contributor" shall mean Licensor and any individual or Legal Entity
|
|
63
|
+
on behalf of whom a Contribution has been received by Licensor and
|
|
64
|
+
subsequently incorporated within the Work.
|
|
65
|
+
|
|
66
|
+
2. Grant of Copyright License. Subject to the terms and conditions of
|
|
67
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
68
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
69
|
+
copyright license to reproduce, prepare Derivative Works of,
|
|
70
|
+
publicly display, publicly perform, sublicense, and distribute the
|
|
71
|
+
Work and such Derivative Works in Source or Object form.
|
|
72
|
+
|
|
73
|
+
3. Grant of Patent License. Subject to the terms and conditions of
|
|
74
|
+
this License, each Contributor hereby grants to You a perpetual,
|
|
75
|
+
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
|
|
76
|
+
(except as stated in this section) patent license to make, have made,
|
|
77
|
+
use, offer to sell, sell, import, and otherwise transfer the Work,
|
|
78
|
+
where such license applies only to those patent claims licensable
|
|
79
|
+
by such Contributor that are necessarily infringed by their
|
|
80
|
+
Contribution(s) alone or by combination of their Contribution(s)
|
|
81
|
+
with the Work to which such Contribution(s) was submitted. If You
|
|
82
|
+
institute patent litigation against any entity (including a
|
|
83
|
+
cross-claim or counterclaim in a lawsuit) alleging that the Work
|
|
84
|
+
or a Contribution incorporated within the Work constitutes direct
|
|
85
|
+
or contributory patent infringement, then any patent licenses
|
|
86
|
+
granted to You under this License for that Work shall terminate
|
|
87
|
+
as of the date such litigation is filed.
|
|
88
|
+
|
|
89
|
+
4. Redistribution. You may reproduce and distribute copies of the
|
|
90
|
+
Work or Derivative Works thereof in any medium, with or without
|
|
91
|
+
modifications, and in Source or Object form, provided that You
|
|
92
|
+
meet the following conditions:
|
|
93
|
+
|
|
94
|
+
(a) You must give any other recipients of the Work or
|
|
95
|
+
Derivative Works a copy of this License; and
|
|
96
|
+
|
|
97
|
+
(b) You must cause any modified files to carry prominent notices
|
|
98
|
+
stating that You changed the files; and
|
|
99
|
+
|
|
100
|
+
(c) You must retain, in the Source form of any Derivative Works
|
|
101
|
+
that You distribute, all copyright, patent, trademark, and
|
|
102
|
+
attribution notices from the Source form of the Work,
|
|
103
|
+
excluding those notices that do not pertain to any part of
|
|
104
|
+
the Derivative Works; and
|
|
105
|
+
|
|
106
|
+
(d) If the Work includes a "NOTICE" text file as part of its
|
|
107
|
+
distribution, then any Derivative Works that You distribute must
|
|
108
|
+
include a readable copy of the attribution notices contained
|
|
109
|
+
within such NOTICE file, excluding those notices that do not
|
|
110
|
+
pertain to any part of the Derivative Works, in at least one
|
|
111
|
+
of the following places: within a NOTICE text file distributed
|
|
112
|
+
as part of the Derivative Works; within the Source form or
|
|
113
|
+
documentation, if provided along with the Derivative Works; or,
|
|
114
|
+
within a display generated by the Derivative Works, if and
|
|
115
|
+
wherever such third-party notices normally appear. The contents
|
|
116
|
+
of the NOTICE file are for informational purposes only and
|
|
117
|
+
do not modify the License. You may add Your own attribution
|
|
118
|
+
notices within Derivative Works that You distribute, alongside
|
|
119
|
+
or as an addendum to the NOTICE text from the Work, provided
|
|
120
|
+
that such additional attribution notices cannot be construed
|
|
121
|
+
as modifying the License.
|
|
122
|
+
|
|
123
|
+
You may add Your own copyright statement to Your modifications and
|
|
124
|
+
may provide additional or different license terms and conditions
|
|
125
|
+
for use, reproduction, or distribution of Your modifications, or
|
|
126
|
+
for any such Derivative Works as a whole, provided Your use,
|
|
127
|
+
reproduction, and distribution of the Work otherwise complies with
|
|
128
|
+
the conditions stated in this License.
|
|
129
|
+
|
|
130
|
+
5. Submission of Contributions. Unless You explicitly state otherwise,
|
|
131
|
+
any Contribution intentionally submitted for inclusion in the Work
|
|
132
|
+
by You to the Licensor shall be under the terms and conditions of
|
|
133
|
+
this License, without any additional terms or conditions.
|
|
134
|
+
Notwithstanding the above, nothing herein shall supersede or modify
|
|
135
|
+
the terms of any separate license agreement you may have executed
|
|
136
|
+
with Licensor regarding such Contributions.
|
|
137
|
+
|
|
138
|
+
6. Trademarks. This License does not grant permission to use the trade
|
|
139
|
+
names, trademarks, service marks, or product names of the Licensor,
|
|
140
|
+
except as required for reasonable and customary use in describing the
|
|
141
|
+
origin of the Work and reproducing the content of the NOTICE file.
|
|
142
|
+
|
|
143
|
+
7. Disclaimer of Warranty. Unless required by applicable law or
|
|
144
|
+
agreed to in writing, Licensor provides the Work (and each
|
|
145
|
+
Contributor provides its Contributions) on an "AS IS" BASIS,
|
|
146
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
|
|
147
|
+
implied, including, without limitation, any warranties or conditions
|
|
148
|
+
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
|
|
149
|
+
PARTICULAR PURPOSE. You are solely responsible for determining the
|
|
150
|
+
appropriateness of using or redistributing the Work and assume any
|
|
151
|
+
risks associated with Your exercise of permissions under this License.
|
|
152
|
+
|
|
153
|
+
8. Limitation of Liability. In no event and under no legal theory,
|
|
154
|
+
whether in tort (including negligence), contract, or otherwise,
|
|
155
|
+
unless required by applicable law (such as deliberate and grossly
|
|
156
|
+
negligent acts) or agreed to in writing, shall any Contributor be
|
|
157
|
+
liable to You for damages, including any direct, indirect, special,
|
|
158
|
+
incidental, or consequential damages of any character arising as a
|
|
159
|
+
result of this License or out of the use or inability to use the
|
|
160
|
+
Work (including but not limited to damages for loss of goodwill,
|
|
161
|
+
work stoppage, computer failure or malfunction, or any and all
|
|
162
|
+
other commercial damages or losses), even if such Contributor
|
|
163
|
+
has been advised of the possibility of such damages.
|
|
164
|
+
|
|
165
|
+
9. Accepting Warranty or Additional Liability. While redistributing
|
|
166
|
+
the Work or Derivative Works thereof, You may choose to offer,
|
|
167
|
+
and charge a fee for, acceptance of support, warranty, indemnity,
|
|
168
|
+
or other liability obligations and/or rights consistent with this
|
|
169
|
+
License. However, in accepting such obligations, You may act only
|
|
170
|
+
on Your own behalf and on Your sole responsibility, not on behalf
|
|
171
|
+
of any other Contributor, and only if You agree to indemnify,
|
|
172
|
+
defend, and hold each Contributor harmless for any liability
|
|
173
|
+
incurred by, or claims asserted against, such Contributor by reason
|
|
174
|
+
of your accepting any such warranty or additional liability.
|
|
175
|
+
|
|
176
|
+
END OF TERMS AND CONDITIONS
|
|
177
|
+
|
|
178
|
+
APPENDIX: How to apply the Apache License to your work.
|
|
179
|
+
|
|
180
|
+
To apply the Apache License to your work, attach the following
|
|
181
|
+
boilerplate notice, with the fields enclosed by brackets "[]"
|
|
182
|
+
replaced with your own identifying information. (Don't include
|
|
183
|
+
the brackets!) The text should be enclosed in the appropriate
|
|
184
|
+
comment syntax for the file format. We also recommend that a
|
|
185
|
+
file or class name and description of purpose be included on the
|
|
186
|
+
same "printed page" as the copyright notice for easier
|
|
187
|
+
identification within third-party archives.
|
|
188
|
+
|
|
189
|
+
Copyright 2026 Dümen contributors
|
|
190
|
+
|
|
191
|
+
Licensed under the Apache License, Version 2.0 (the "License");
|
|
192
|
+
you may not use this file except in compliance with the License.
|
|
193
|
+
You may obtain a copy of the License at
|
|
194
|
+
|
|
195
|
+
http://www.apache.org/licenses/LICENSE-2.0
|
|
196
|
+
|
|
197
|
+
Unless required by applicable law or agreed to in writing, software
|
|
198
|
+
distributed under the License is distributed on an "AS IS" BASIS,
|
|
199
|
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
200
|
+
See the License for the specific language governing permissions and
|
|
201
|
+
limitations under the License.
|