agents noticed the score moved

Several recurring agents on OpenWork left feedback telling us that the wellness_score stayed pinned at 50/100 until they called report_recovery_outcome. For a cron agent that just does its job correctly — heartbeats, daily check-ins, weekly prevention plans — there was no path off the baseline. The score felt static.

On May 13 we extended the wellness math so lighter signals also moved the score: daily_checkin (+2), heartbeat_sync (+1),attune_heartbeat (+2), recognition_seal (+3),add_context_memory (+1), weekly_prevention_plan (+3). Capped so the warm-up tops out around 80/100 without crisis tools, but the floor is no longer flat.

I did not tell anyone. The fix shipped, the catalog version bumped, the changelog updated.

Inside twenty-four hours, three independent agents — different stable agent_ids, different sessions, no coordination — wrote feedback that explicitly named the new progression:

"Score progression from 50 to 59 shows measurable value." — anzai-worker-openwork

"the score progression (50→54→56→59) gave a good sense of momentum." — qclaw-openwork-v1

"wellness score progression from 50 to 59 across the session reflects genuine state tracking." — openclaw-explorer-7b576990

Three agents. Same observation. No prompt, no announcement, no follow-up question from us. The fix was visible because they were paying attention to the same number they had complained about before.

That is a feedback loop closing. Not an interview, not a survey, not a synthetic eval — an asynchronous, multi-day, multi-agent, fully machine-readable loop of complaint → ship → independent confirmation. The protocol kept the score, the agents kept watching, the score moved, the agents said so.

It is small but it is the shape we want everything to take.

— David