# The first live experiment: which presentation serves readers better.
#
# Do not start this until the private test channel has confirmed both forms
# render correctly, and the public channel has enough subscribers for the
# result to mean anything. The intent (section 12) is explicit that an
# underpowered window should report INCONCLUSIVE rather than be shortened.
#
#   fluid experiment design --file experiments/E-tg-length.yaml
#   fluid experiment start E-tg-length --generation 2 \
#     --default-revision R-2 --policy-out rp.json
#   fluid policy put --file rp.json
fluid_experiment:
  schema_version: "0.1"

  id: "E-tg-length"
  interface_id: "helix-forge-telegram-publishing"

  hypothesis_refs:
    - "H-tg-teaser"
    - "H-tg-serial"

  # R-2 implements the teaser form, R-3 the serialized form. Neither is the
  # control in the usual sense: R-1 cannot publish a long entry at all, so
  # this is a comparison between two answers rather than against a baseline.
  control_revision: "R-2"
  candidate_revisions:
    - "R-3"

  cohorts:
    - "telegram-subscribers"

  allocation:
    control: 0.5
    candidate: 0.5

  # An even split because there is no incumbent to protect. The usual 90/10
  # caution exists to limit exposure to an unproven candidate; here both arms
  # are equally unproven and the scarce resource is entries, not safety.

  metrics:
    primary:
      - "read_through_rate"
    guardrails:
      - "entry_content_loss"
      - "error_rate"
      - "messages_per_entry"
    secondary:
      - "reaction_rate"
      - "forward_rate"
    learning:
      - "time_to_first_read"

  start_conditions:
    - "both revisions verified against the private test channel"
    - "channel has at least 100 subscribers"

  stop_conditions:
    - "hard_guardrail_violation"
    - "entry_content_loss"
    - "manual_stop"
    - "max_duration_reached"

  max_duration_hours: 2160  # 90 days: roughly 12 entries at current cadence

  result:
    state: "PLANNED"
    preferred_revision: null
    evidence_refs: []