harbor-dsh-evolution
Harbor-side integration for both DeepSeek Harness Candidate evaluation and privacy-preserving Historical Generation Evaluation.
It provides:
DshCandidateAgent: verifies and uploads an immutable Candidate, installs its locked npm dependencies, and runs it through Harbor's ACP runner.EvolutionPlugin: binds each Job to Candidate, Dataset Manifest, Evaluation Stack Manifest, Context v2, Architecture Doctor, Trial assessments, Population report, and Summary.SessionObservationAgent: presents one frozen, credential-redacted DSH Session Observation as one deterministic Harbor Trial without rerunning a Candidate; ordinary paths in visible source text are preserved.HistoricalGenerationEvaluationPlugin: validates the immutable Historical Batch/Dataset/Stack cross-links, runs Evaluator v2, preservescompleted-unscoredabstention, and writes Summary v4 plus a strict completion sentinel. Historical Jobs are diagnostic and cannot enter Promotion Gate.harbor-dsh: initializes strict projects; validates/snapshots Candidates, Datasets, and Stacks; materializes Historical Batch inputs; previews Context v2; diagnoses architecture; summarizes Jobs; and runs the deterministic Promotion Gate.
Install it into the same Python environment as Harbor so the plugin entry point is discoverable:
uv venv .venv
uv pip install --python .venv/bin/python harbor-dsh-evolution==0.9.5
source .venv/bin/activate
harbor plugins list
harbor-dsh --help
The plugin list must contain both entry points:
dsh-evolution
dsh-historical-evaluation
Development from this repository:
uv sync
uv run harbor plugins list
uv run harbor-dsh --help
uv run harbor-dsh historical --help
uv run harbor-dsh dataset validate ../../examples/deep-research/task --project-root ../..
uv run harbor-dsh stack validate ../../examples/deep-research/.harbor/evaluation-stack.yml --project-root ../..
uv run pytest
uv build
Harbor and this package must be installed into the same Python environment for the dsh-evolution and dsh-historical-evaluation entry points to appear in harbor plugins list.
snapshot derives Candidate id and version from package.json unless explicitly supplied. Context v1 is not accepted. Candidate promotion requires Context v2 and emits structured mismatch, artifact, infrastructure, metric, and regression reason codes. Historical materialization instead derives a matching Dataset and immutable Stack from a credential-redacted historical-generation-batch/v1; it never creates a Candidate identity, reports insufficient evidence as completed-unscored, and always returns UNSUPPORTED_JOB_KIND_FOR_PROMOTION if passed to Gate.
1 maintainer
Release files for harbor-dsh-evolution 0.9.5
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| harbor_dsh_evolution-0.9.5.tar.gz | 284.3 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| harbor_dsh_evolution-0.9.5-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 403.7 kB
Release files / harbor_dsh_evolution-0.9.5.tar.gz
| Download URL | harbor_dsh_evolution-0.9.5.tar.gz |
|---|---|
| Size | 284.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
fd6304078349dd6fe34312ace9c7c51bc09e5fb09580d4496ceadb4315a0e9c2
|
|
BLAKE2b-256 checksum How to use checksums |
81ed0857de5beb0021e8135f465cdd086e3b1114a236edab62d7058aa1bfb615
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 7, 2026.
Transparency logRelease files / harbor_dsh_evolution-0.9.5-py3-none-any.whl
| Download URL | harbor_dsh_evolution-0.9.5-py3-none-any.whl |
|---|---|
| Size | 119.5 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
6c43b85b61082d772b68978dce10f6436bfe1da7090c2c36225bb9a8837451dd
|
|
BLAKE2b-256 checksum How to use checksums |
4cadb06a12b253f34d435b214768707cdd02f6b8bdc483e71a6d8312db727b3d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 7, 2026.
Transparency log
