How the research corrects itself
Research publishes when the run that wrote it finishes. The model reads the primary sources, writes the record, and publishes it in the same pass. Three checks run first. Every cited source must resolve to a stored document. A record that supersedes another must point at the same matter and state the reading it replaces. A run that failed to read its sources publishes nothing. A record that fails a check waits in the queue with the reason attached.
Every published version carries a SHA-256 digest of its exact prose and evidence receipts, and names the run that wrote it. Published versions, source versions, incident events, and superseded records are never overwritten.
Current coverage, kept separate
Freshness
| Research object | Overdue | Maximum normal cadence |
|---|---|---|
| Individual judgments | 0 | 93 days when approved; 203 days otherwise; 32 days during an active incident |
| Class applications | 0 | Evidence-appropriate; 203 days maximum unless adverse |
| Class dossiers | 0 | 184 days |
A memo is current because it is watched, not because a date has not passed. Every night the review pass reads each memo’s venues and files for warnings and open items; a quiet memo is confirmed on its date with the evidence recorded on the page, and a memo where something moved is reopened with a drafted revision. An approval is confirmed quarterly; a class application or a watched record holds for two quarters; an individually researched rejection holds for a year. A fired trigger reopens the memo the same night.
Corrections and supersession
A typo that does not change meaning is labeled a correction. New evidence that changes an interpretation creates a superseding version. A changed verdict or a methodology change is labeled as such, names the evidence that changed the reading, and states the prior view. The old version is not overwritten or silently removed.
What stayed unresolved, and changing sources
Every record lists what it could not settle: contrary estimates, alternative readings of the evidence, and the conditions that would change the conclusion. The public conclusion may stay concise; the open questions stay on the record. Static memo sources carry URL, source type, access date, and supported claims. Ingested regulatory and on-chain evidence additionally preserves every observed content hash and excerpt as an immutable source version.
3884/3884 research source receipts complete · ledger 1022dfa3bb48e18f…
Incident clock
- Open the incident and record confirmed facts separately from unknowns.
- Record a temporary research posture before debating the lasting verdict.
- Identify client-held positions and whether subscriber communication is required.
- Assign an owner, deadline, and follow-up; preserve each event in the incident timeline.
- Close only after the permanent research record and any required communication are complete.