One source is stored once. Each decision keeps its own rule, review and outcome. Earlier workspaces retain their original formats.
One change. Different consequences.
Which decisions need another look?
Trace a source change through the exact rule behind each decision. See what changed, what stayed within scope, and what we still cannot establish.
When a later declaration clarifies a deferred review, preserve the earlier unanswered question and open a separate review explicitly. Replacement records responsibility and exact evidence links; it does not approve production. Return to the original v5 rehearsal.
01 / Establish a shared baseline
A clear case for comparison.
Create three explicitly simulated starting decisions: a request boundary, a declared-grant boundary, and a production hold pending context. The synthetic baseline requests read, separately declares read granted, and leaves deployment Unknown. Nothing is saved until you submit.
02 / Follow the consequence
Same evidence. Different decision bases.
Select any node to inspect exact references. The trail compares the latest capture with each applicable decision basis; earlier events remain in the history. It is not a replay of what was known at every past moment.
Clarification / preserve both questions
Earlier uncertainty. A separate reassessment.
Inspect the reasoning
Select a trail element.
Explore a source, evidence version, assumption, decision, review or outcome. Inspecting does not write records.
03 / Preserve the next capture
Record once. Compare across three decisions.
For the demonstration, add a write request, retain read-only declared grants, and keep context Unknown. This produces affected, no relevant change, and unresolved applicability. You enter the evidence; no live system is queried. Pending reviews retain their exact earlier captures.
04 / Accountable reassessment
One outcome belongs to one decision.
05 / Keep earlier reasoning
Shared evidence. Preserved histories.
Import history Inspect, then restore
Choose a Decision Continuity JSON export, maximum 1 MiB. Nothing is uploaded. Inspection writes nothing; confirmation restores only a missing record and opens its original workspace. Different existing history cannot be overwritten.
Review before restoring
Consistency checks do not prove authenticity. Reviewer codes are unauthenticated; browser storage is editable.
Synthetic technical demonstration. No practitioner validation, live source monitoring, runtime assurance, authentication, cloud sharing or AI decisions.