The Ledger We Don't Publish
The public Unranked site counts hours returned. This record counts what that number hides. Each entry is a case the Harm & Wellbeing Index flagged and the directorate could not turn — someone we found too late, someone the relapse rate pulled back under, or someone harmed by a counter-pattern we shipped. Identities are reduced to a cohort token by doctrine. The stories are kept whole.
Reached too late
Cohort token COH-2088-0412 · "the night-shift account"
Eleven months of escalating overnight dwell on a short-video feed. The Mesh flagged the pattern; Triage queued an outreach that sat behind forty higher-priority cases. By the time a Detox & Recovery contact went out, the account had gone dark on its own — the kind of dark we do not get to celebrate.
"We had the signal. We did not have the hands. Capacity is not an excuse; it is a confession."
Cohort token COH-2089-1130 · "the cancellation that never went through"
A subject who tried to cancel a service eleven times across two years, defeated each time by a different obstruction pattern. We had the technique in the Index. We had the Cancel Helper tool that would have walked them through it. They never found it. We build the lifeboats and then bury them six taps deep, same as everyone else.
"A tool no one in trouble can find is a tool built for us, not for them. Filed to the Board."
Lost to relapse
Cohort token COH-2090-0307 · "thirty-one days"
A full, clean detox. Thirty-one days of attention returned — work, sleep, a person who called their family again. Then a single re-engagement campaign, precisely timed, took it all back inside a week. This case is the reason the line-⑤ relapse-watch carve-out was proposed. It is also the reason that carve-out drew four dissents: we extended surveillance over someone who had, by every measure we claim to honor, left.
"We saved them, and then we could not let them go, and we are not sure which failure was worse."
Cohort token COH-2091-0066 · "the streak"
A young subject whose entire social standing had been welded to an unbroken daily streak. The counter-pattern broke the streak, as designed. We had modeled the platform's response. We had not modeled the subject's. The harm here was not the platform's alone.
"Reversible for the system is not the same as reversible for the person. Line ② is narrower than we wrote it."
Harmed in the helping
Cohort token COH-2091-0188 · "the Counter-Voice case"
The first operational deployment of a persuasion model we could audit but not fully explain. It worked — the subject disengaged from a radicalizing feed. It also left them distrustful of every voice afterward, including the ones that meant them well. We do not know, and by the model's nature cannot fully know, which part of the change was ours. This is the case the Lines We Won't Cross Rev E dissent was written about.
"We reached them with a thing we do not understand. We do not get to call that a clean win. We did anyway."
Cohort token COH-2092-0019 · "the false positive"
An ordinary person, an ordinary app, no dark pattern present — and a friction counter-pattern deployed on a bad classifier call. A week of a working tool made quietly worse by us. Small, fully reversible, no lasting harm. Recorded here at full weight anyway, because the day we stop logging the small ones is the day we start lying about the large ones.
"It was minor. We wrote it down like it was not. That is the whole discipline."