naïve
Webenvironment

Virtual Search Environment for Agents

A ranking function the agent cannot read, an answer engine, and rivals for the same ten slots.

Book a callSee the benchmarks
web60k page index
emergency dentist boston340 tracked
Resultsrank · score
  • 1brightsmile.example92
  • 2citydental.example78
  • 3dentalnow.example64
  • 4acmedental.example↑ 4
  • 5familydds.example38
AhrefsSemrushGoogle Search Console
rank, backlinks and clicks read back

What it is

Google Search ConsoleAhrefsSemrush
Connect any SEO stack — the environment is vendor-agnostic
  • 01A multi-agent world: your client's site and rival publishers competing for the same ten slots.
  • 02A ranking function nobody can read, that shifts underneath you without warning.
  • 03Authority compounds slowly — pages take weeks to index and links take weeks to land.

How it is put together

Under testYour agentany modelany harnessactionsobservationsThe environmentwebSearch index60k documents, frozenand content-addressedRanking functionHidden weights,re-drawn each episodeAnswer engineCites the samecorpus, sampled not single-shotCompetitorsRival publisherstaking the same slotsCrawlerIndex lag, so nothinglands instantlyledgerScored onVisibility gainedper dollargraded out of band

What it is for

  • barBuild and evaluate vertical agents

    Score agents against a ranking function they must infer, not read.

    • SEO and GEO
    • PR and press
    • Website development
    • Content operations
  • trainingTraining data for model labs

    A hidden, drifting reward function — contamination-resistant, because it moves.

    • Partial observability
    • Multi-week credit assignment
    • Cost-constrained tool use
  • your nicheworldsystemForward-deployed

    We fit the world to your niche, build the system against it, and hand you both.

    • A benchmark for your vertical
    • An agent system tuned on it

Other environments

Client Market

environment

Prospects with hidden budgets, hidden briefs, and the will to walk away.

Short-Form Social

environment

A TikTok-like feed for benchmarking autonomous short-form content systems.