coulomb/clay-borg
4
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
clay-borg/simulators/ai-ai.md
tegwick b590e7fd59
Some checks failed
ci / check (push) Failing after 4s
Details
simulators/: persist the survey, and let it shrink two of the three tracks 
Eight profiles on a common schema, each marking what was checked against a
source this session and what is background recollection. Three are marked
unverified in full — Machinations, the play substrates, most of RBG — and
say so rather than reading as evaluations. Written straight after three
review rounds whose entire yield was claims outrunning what had been
checked, so the confidence rule is the first thing in the README.

The survey changed the plan, which is what a survey is for.

Track C was described in Positioning as open ground. It is not: Browne
published 57 criteria for game quality, and Ai Ai already computes
designer-facing measures — drama, lead changes, branching factor,
completion, duration — from played games. The track becomes adopt, credit
and find the gap. The gap looks real: those measures presume a leader, and
SHARED GROUND has none — Modes.csv gives its tiebreak as "Not applicable".

Track B probably adopts rather than builds. OpenSpiel implements CFR,
best-response and exploitability over games that are simultaneous-move,
imperfect-information and co-operative, which is all four of GROUND's
awkward properties. "Does ATTACK ever pay" is a best-response question,
and we spent three review rounds refining a two-policy sweep for it. The
first Track B task is now one question — is exploitability meaningful for
a co-operative game with a shared threshold — not a build.

The cost of not surveying earlier is therefore measurable, and is recorded
rather than glossed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-08 11:56:27 +02:00

3.3 KiB
Raw Permalink Blame History

Ai Ai (and MoGaL)

Stephen Tavener, with Cameron Browne · Java · checked · licence and cost unverified

What it optimises for

General game playing with analysis aimed at game authors. Ai Ai is built on MoGaL (Modular Game Library), a general game system by Browne and Tavener assembled from Java blocks glued by a JSON-like language. checked

How a game is defined

Hand-coded in Java for efficiency, or assembled from blocks using a JSON-based scripting language. checked

What it can analyse

This is the important part and the reason this profile exists. Ai Ai "has quite a lot of analysis functions and contains analysis tools for game authoring", and plays via MCTS variants. checked

The metric family in this research tradition — Browne, and work building on it — includes turn metrics: branching factor, drama, killer moves, lead changes; and endgame metrics: completion, duration, advantage. checked (from the literature; that Ai Ai computes this exact set is unverified).

Lead change is treated as close to drama: a game that never changes leader is tedious, and one that changes too often is chaotic. checked

What it does not do

  • No federated authority: no concept of the game's owner ruling on an undecided case. unverified
  • No provenance binding to a publisher's dataset. unverified
  • Its measurements are outputs, not claims under adversarial review.

Relevance to clay-borg

Track C, and it substantially reframes it. Positioning.md §4 described "assimilated knowledge about why games work" as open ground. It is not. Ai Ai computes designer-facing measures of exactly this kind and has done for years.

Also Track A: MoGaL's "Java blocks glued by JSON" is a third point on the description-language axis, between Ludii's ludemes and hand-written code — and closer to what a clay-borg game definition might look like.

What to steal

  • The metric family itself. Drama, lead change, completion, duration and branching factor are directly computable from our journals, and we already compute cruder cousins (win rate, arm rate, action counts).
  • The framing that a metric is for the author, not for an agent.

What to avoid

Adopting the metrics without their assumptions. Drama and lead change are defined against a notion of "who is ahead", which in a co-operative game with a defection mechanic is not obvious — GROUND's SHARED GROUND mode has no individual leader at all, and Modes.csv says its tiebreak is "Not applicable". A metric imported without checking that its subject exists here is the wrong-subject error (../specs/Taxonomy.md §2.2).

Open questions

  • The exact metric list Ai Ai produces, and their definitions.
  • Whether any of them are defined for co-operative or hidden-role games.
  • Licence, and whether the analysis is usable headlessly.

Sources