Files
aevolution/docs/adr/0130-cellular-playtest-telemetry-boundary.md
dingoandClaude Opus 5.5 d298d31115 Predeclare the M11.9 playtests and add account-free play telemetry
M11.9a (ADR-0130): the repeated-life playtest plan fixes its questions, sample
and recording before any session. summarize_cellular_play derives aggregates
(population, death causes, lifespans, buds, mutations, variety) from a committed
cellular state, and evo-world-clone summary prints them for a region head,
read-only and without account data. The roadmap records M11 status.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-14 08:09:42 -04:00

2.0 KiB

ADR-0130: Cellular playtest telemetry boundary

Status

Accepted (2026-10-14) for the M11.9 repeated-life playtests. Revisit before any telemetry leaves the development environment.

Context

M11.9 must assess movement, food decisions, predation, tradeoffs, descendants and variety over multiple lives, using human feedback plus bounded, consent-aware telemetry. The durable checkpoint already holds everything needed to count what happened in a region (lives, corpses with causes, the lineage archive with births, deaths and genome revisions, the bud sequence). It also holds account identities, which are not needed for these questions.

Decision

  • Playtest telemetry is derived offline from a region's committed head by evo-world-clone summary (read-only; owner connection; nothing written). It reports only aggregates: the living population by kind and presence; deaths by cause and by kind; minimum, median and maximum lifespan in ticks; buds; applied mutations; distinct living genomes and the deepest living lineage.
  • No account, session, grant, position, IP address, timing of individual inputs or per-player history is collected, printed or stored by this tool. Linking results to a participant happens only in the human notes the participant agrees to.
  • The tool runs on demand by the maintainer, in DEV only. There is no background collection and no new table.
  • Per-player questions (did you understand why you died, did a mutation choice matter) are asked in the playtest notes, not inferred from telemetry.

Alternatives

  • Server-side event logging per action: rejected for M11.9; it would collect per-player behaviour without a consent flow and add a retention problem.
  • Replay-based full-trace analysis: possible later for crash regression; it needs no new data, only the existing history audit.

Consequences

Playtest sessions can be compared by region-level outcomes (death causes, variety, descendants) while participants stay anonymous in the data. Finer behaviour questions remain qualitative.