Evidpath

Evidpath documentation · Swarm testing

Learn the swarm model, choose a domain, and read the evidence.

This is the main reading path for Evidpath. GitHub is still there for source, releases, and issues, but the product documentation lives here as a structured site. Start with the swarm model, then choose the recommender, search, or agent domain product that matches the system you ship.

Start here

The shortest useful path.

Install the package, run one domain audit, then compare a candidate against a baseline when a release decision needs evidence.

QuickstartRequest a domain pilot
Basic flow
python -m pip install evidpath

evidpath audit --domain recommender \
  --target-url http://127.0.0.1:8051 \
  --scenario returning-user-home-feed

evidpath audit --domain search \
  --target-url http://127.0.0.1:8051 \
  --scenario time-sensitive-query

evidpath audit --domain agents \
  --driver-config-path ./driver_config.json \
  --scenario current-info-tool-use

Guides

Read by job, not by repository file.

Source repository
1

Question

2

Domain

3

Swarm

4

Trace

5

Judge

6

Evidence

Understand swarm testing

Learn how release questions become seeded behavior swarms and trace-backed evidence.

Open guide

Choose a domain product

Compare recommender, search, and agent trajectory testing surfaces.

Open guide

Test recommenders

Review slates, novelty, repetition, trust collapse, cold start, and abandonment.

Open guide

Test search rankers

Review relevance, freshness, ambiguity, typo recovery, and zero-result behavior.

Open guide

Test agents

Review tool use, grounding, refusal, multi-turn state, unsafe requests, and latency.

Open guide

Pick an integration path

Choose native HTTP, schema-mapped HTTP, Python callable, or agent protocol drivers.

Open guide

Run the CLI

See audit, compare, generated coverage, and plan-first workflow boundaries.

Open guide

Read the evidence packet

Understand reports, JSON, traces, manifests, semantic sidecars, and regression outputs.

Open guide

Compare before launch

Use baseline/candidate runs to make release review concrete.

Open guide

Fix common issues

Debug endpoint, contract, driver, output, and provider setup problems.

Open guide