shalintripathi Measure outcomes (closing the loop)
- Scan outcomes/ for records with a due measurement date <= today that lack that measurement.
- Per record: pull GSC for the target URL - clicks/impressions/CTR/position for the 7 (or 28) days after the change vs the same window before. GSC data lags 2-3 days: fetch the day-7 window no earlier than day 10, and the day-28 window no earlier than day 31. No GSC connector: record "unmeasured (no GSC)" honestly.
- Append the delta to the outcome record + one signal line
("outcome : position 8.2 -> 5.9 after title rewrite").
"After" is a sequence, not a cause. Attributing the move to the change
follows the attribution rule (canonical in hoo-daily's anomaly
section, step 2.7): the claim, the comparison actually run, and what
would falsify it - for example "compared the 28 days after against the
same 28-day window before, same URL; if the site-wide trend moved the
same way over those windows, the change is not what did it." Where
that comparison was not run, the record reads
cause: unknown (before/after window not compared)and keeps the delta. A measured number with an honest unknown is what lets the reflector score a skillbook entry later; a confident story that nobody checked teaches the loop the wrong lesson. - Wins and losses BOTH matter: the reflector reads outcomes to score
skillbook entries helpful/harmful. At day-28, set the item to measured
via
PYTHONPATH="$CLAUDE_PLUGIN_ROOT/lib" python3 -m core status <item-path> measured --actor agent. Never edit brain frontmatter directly. The contract CLI is the only write path for status and approvals.