dsh-llm-verifier-pro
LLM-as-a-Verifier fine-grained reward tools, Probabilistic Pivot Tournament, and Best-of-N mode for DeepSeek Harness.
Install
$ dsh plugin --profile web add github:jmche/dsh-llm-verifier-proPlugin Overview & Capabilities
AI-assisted organization based on the public repository snapshot. The content must be grounded in source evidence and does not replace compatibility or security verification.
Key Capabilities
- verify_compare tool for directed pairwise comparisons with fine-grained logprob rewards
- verify_select tool using Probabilistic Pivot Tournament (PPT) for efficient O(N·k) best-of-N candidate selection
- verify_track tool for offline multi-checkpoint trajectory progress tracking
- Best-of-N (Bo-N) turn replay mode with parallel or serial candidate generation
- Model mix support allowing diverse candidate generation across different providers and models
- Support for TypeSafe Jev System One selector as an alternative pairwise scoring engine
- Web settings panel integration for dynamic Cordis patch configuration
Useful For
- Filtering and selecting the highest-quality assistant answers before returning them to users
- Self-verification and quality grading of multi-step agent reasoning chains
- Comparing and ranking alternative code or text candidate generations
- Tracking task progress curves across multi-step execution checkpoints
Who It Fits
- DeepSeek Harness users seeking test-time compute scaling and candidate verification
- Agent developers building automated self-checking and reasoning workflows
- Power users desiring Best-of-N answer sampling across single or heterogeneous models
Documented Limitations
- Requires LLM endpoints supporting token-level logprobs or a compatible Jev System One endpoint
- Does not support multimodal image or video inputs
- Best-of-N mode only scores final text answers and deliberately skips tool-calling action turns
- Progress tracking uses an offline single-call model rather than per-prefix online tracking
DSH Compatibility
Version-specific runtime evidence collected by DSH Plugin. A missing result means we have not tested that combination yet.
Security Signals
Objective signals discovered from package metadata and source inspection. These are not a guarantee that a plugin is safe.
package.json declares 0 runtime, 10 development, 6 peer, and 0 optional dependencies.
package.json declares DSH bundle metadata.
GitHub reports the repository license as NOASSERTION.
A root package.json was captured and can be inspected by the registry.
Public GitHub source metadata is available for this registry snapshot.
Source & Registry Notes
Public provenance, Registry classification, and the latest source check for this entry, kept separate from runtime verification.
- Source repository
- jmche/dsh-llm-verifier-pro
- Registry source
- GitHub · dsh-plugin topic
- Registry classification
- Plugin
- Source checked
- 9ae95d7 · 2026-09-30
This project is independently indexed from public source information. DSH Plugin is not affiliated with DeepSeek or the plugin author. Always check the author repository before installation.
Repository Activity
- GitHub stars
- 0GitHub stars
- Forks
- 0
- Open issues
- 0
- Last commit
- 2026-09-29
- Last release
- No release detected
Related DSH Plugins
Ranked by overlapping capabilities, use cases, plugin type, categories, and DSH profile.
Vision routing and pixel-level visual tool suite for DeepSeek Harness agents with built-in free fallbacks.
View plugin →DeepSeek Harness web-search plugin that adds web search, X search, and focused page fetching with structured evidence.
View plugin →Local-first AI agent runtime and DSH plugin bundle providing sandboxed sessions, MCP tools, audit logs, and session replay.
View plugin →Persistent multi-model Agent teams and bounded DAG workflows with GUI controls for DeepSeek Harness.
View plugin →