Skip to content

Releases

v0.4.2

Choose a tag to compare

@phanisaimunipalli phanisaimunipalli released this 14 Nov 18:31

GitHub search is now scoped to the project named in the idea. The planner returns the repo (owner/name) when the idea is a change to an existing project, and issue search is limited to it, so the evidence is about that project instead of unrelated repos.

v0.4.1

Choose a tag to compare

@phanisaimunipalli phanisaimunipalli released this 11 Nov 02:32

Adds --reuse-run and --independent-traits for A/B tests of trait structure, and records the first accuracy test in docs/evals.md. README now states the result: no measurable effect on verdicts in a small test.

v0.4.0

Choose a tag to compare

@phanisaimunipalli phanisaimunipalli released this 09 Nov 00:43

What's new

  • AI population designer. mockcrowd now designs populations of personas whose traits connect the way real people's do. Real posts are scored on six core traits plus audience specific ones, the links between traits are measured, and each persona is drawn from those links and grounded in the three nearest real posts.
  • mockcrowd design "audience" -o crowd.json. Designs a population with no votes and no verdict. --pin TRAIT=low|medium|high|0.8 fixes a trait for everyone and the linked traits adjust.
  • mockcrowd personas <run-id>. Exports any run's personas with traits and the real posts behind them.
  • MCP tool design_personas for other AI agents.
  • Trait checks. Reports show the strongest trait links and how closely the crowd reproduces them. Evals gain trait_error, trait_gain and trait_spread.
  • Rewritten README around persona design.

Limits, stated plainly. In live runs only about 20% of trait scores were stated in a post; the rest are the AI's estimates, so the measured links partly reflect its assumptions. Whether structured traits make verdicts more accurate is not yet measured.

Upgrade: pip install -U mockcrowd or uv tool upgrade mockcrowd.

Full changelog: v0.3.0...v0.4.0

v0.3.0

Choose a tag to compare

@phanisaimunipalli phanisaimunipalli released this 08 Nov 16:44

What's new

  • Free mode. --free (or free: true over MCP) judges free ideas and open source features by whether people would actually use them, against a free control, instead of whether they would pay.
  • GitHub issues as a source. For developer audiences, personas are now grounded in real GitHub issues and their top comments. Set GITHUB_TOKEN for a higher rate limit.
  • --exclude TEXT. Leave out posts whose URL or id matches, so backtests on a known issue cannot see the answer.
  • Update notice. Once a day mockcrowd prints one line when a newer version is on PyPI. It never installs anything. Turn it off with MOCKCROWD_NO_UPDATE_CHECK=1.

Upgrade: pip install -U mockcrowd or uv tool upgrade mockcrowd.

Full changelog: v0.2.0...v0.3.0

v0.2.0

Choose a tag to compare

@phanisaimunipalli phanisaimunipalli released this 07 Nov 20:53

What's new

  • MCP server. pip install "mockcrowd[mcp]", then mockcrowd mcp. Any AI agent can now validate ideas with five tools: validate_idea, get_report, list_runs, record_outcome and list_brains.
  • Persona characters. The README shows the crowd as characters and replays a real debate as a chat.
  • Security hardening. Scraped posts are fenced in prompts, OPENAI_API_KEY only goes to api.openai.com, auth headers drop on cross-origin redirects, saved runs are private to you, and CI actions are pinned to commits.

Full changelog: v0.1.0...v0.2.0

v0.1.0

Choose a tag to compare

@phanisaimunipalli phanisaimunipalli released this 01 Nov 18:17

First public release of MockCrowd — validate a product idea before you build it, with hundreds of AI personas grounded in real Reddit, Hacker News and Bluesky posts, a blind vote against a control pitch, a simulated discussion, and a backtest on known hits and flops.