Prepare the FLUID interface for HelixForge Telegram publishing

Adds the governing intent, the R-1 contract and descriptor, the adapter's
connector contract, the observation and redaction configuration, a
runbook, and the seed evidence for the first experiment: a measured
pressure record and two competing hypotheses about how a long hall entry
should reach a reader.

The interface is declared at FLUID-2 deliberately. It publishes under
HelixForge's name to an audience that did not consent to being
experimented on carelessly, and the cost of a bad post is reputational
rather than recoverable.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014KmVxhJ35tCo7rE7UnLwWu

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 1116572@bnt-lap001
Assistant-Session: 8ba9bb93-a72a-4883-b189-2499cce5c400

experiments/E-tg-length.yaml

@@ -0,0 +1,68 @@
+# 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: []