Preserve AI provenance, pre-event delivery, actor-separated exposure, later event, diagnosis, accepted repair, corrected delivery, maintainer recheck, affected-user or independent-operator recheck, prevention, accepted second-person change, and transferred authority. Zero case clears all twelve. OpenClaw issue 30 and PR 31 remain open; Prism issue 81 is closed after a maintainer check but without the affected operator's return. Ticket status is not a recovery receipt. Read the complete account →

The CHI 2024 MAIDR study gives 11 blind participants a pre-AI multimodal chart system; the current project declares a complete TypeScript rewrite and an AI description layer. That is a valuable baseline plus a new artifact state—not an exact-version participant recheck. A public-health copilot paper describes a 16-person trust/usability method, then says the experiment was removed and reports no human result. Join exact artifact and version, AI state, release, later event, actor, custody, and outcome before counting a lifecycle.

Public code narrows the later eight-participant AI study to a legacy box-plot surface and v2.10.0 as the first containing tag. It does not name the participant-tested build or model. Later model swaps, verification fixes, deprecation, and the separate TypeScript rewrite trigger new representative checks; they do not supply them.

Graphy follows the same three blind co-designers while selected interaction changes are implemented between rounds. The final own-data workshop shows the evolved interaction in use, but the paper explicitly is not a formal usability or performance evaluation. Its seven-commit public repository has no workshop-version binding, tag, release, or representative recheck. Count repeated co-design—not a release afterlife.

Every required receipt appears somewhere in the held evidence, but none stays attached to one artifact lineage. A complete row needs repeated representative use, immutable tested build, exact exposure model, versioned release, later material event, and representative or actor-separated post-change recheck. This is zero of seven in named surfaces through 15 August 2026—not a claim about every private or inaccessible case.

A 2021–2025 HealthTech program followed 21 projects through 84 recorded calls, notes and decision logs, backlogs and issue trackers, one governance-dashboard case, and a 16-startup survey. The case reached deployment with patients. It is longitudinal visualization evidence—not a generative-authoring study—and it does not measure total cost, accessible reader outcomes, calibrated trust, or comparative maintenance efficacy.

DV-World adds 50 native Excel repair tasks and 80 new-data or evolving-requirement tasks. The best reported agents reached 48% repair success and 51.44% evolution. Use a prepared change as a pre-delivery gate; the benchmark does not follow an accepted artifact through elapsed maintenance, handoff, reader use, or whole cost.

A four-month KubeStellar Console report and commit-pinned QA record follow one AI-assisted dashboard through a live blank-page regression, two-step repair, restored deployment, and a manually added post-build check. This is project-authored single-maintainer evidence—not a comparative rate, whole-cost ledger, handoff, accessible-reader result, consequential decision, or trust result.

OpenClaw Agent Dashboard says it was built with Claude Code and shipped v3.0.0. Thirteen days later, a non-owner operator reported that its cost view showed $0 on a custom provider despite present token data. The owner acknowledged that configuration had not been tested. A different non-owner's open PR 31 proposes a matching fallback, but has no maintainer review, merge, corrected release, or reporter recheck. A proposal is not recovery.

Prism says its family dashboard was built with Claude Code under human product direction. One non-owner operator's Home Assistant route failed at install, repair build, and later startup before they used ordinary Docker instead. v1.8.11 and its entrypoint contain the named socket and schema-replay repairs; the maintainer reports install, start, and restart checks on real Home Assistant OS. The operator never rechecked it. Route substitution is not recovery of the failed route, and no defect is attributed to Claude Code.

Prism's non-owner PR 23 carries a Claude-assisted weather visualization through owner-found defects, contributor repair, owner integration, and a credited release; the contributor later returns with accepted PR 41. OpenClaw's non-owner PR 15 puts Claude-assisted multi-provider pricing into the product after owner integration. These establish second-person change—and one repeat contributor—not release, incident, or recovery authority.