Correct the research.Recheck what depends on it.
Propose a correction, review whether it qualifies, and recheck dependent data, calculations and claims. Adopt a new research version while preserving the earlier result and its evidence.
One correction reaches beyond one cell.
- 01
A proposed change
The correction has an identified input, reason and authorized approver.
- 02
Dependent work
Data correspondence, calculation and claim are checked again where they depend on it.
- 03
A new research state
The result may remain supported, change, or require attention. Approval alone cannot establish it.
- 04
Both versions retained
The current result is explicit and the earlier Record keeps its original evidence.
A reason to change is not a guarantee.
| Correction question | What must be established |
|---|---|
| Does the change qualify? | The correction meets the rules for the specific workflow. |
| Was restoration planned? | Restoration uses originals and a full-restoration rule registered before data arrived, where required. |
| Do the checks still hold? | Applicable numerical and logical checks pass for the new version. |
Arbitrary edits, post-data method changes and partial restoration do not automatically restore a guarantee.
Propose. Approve. Recheck. Then adopt the new version.
- 01
Describe the correction.
Your AI proposes the change with its reason.
- 02
Confirm it.
An authorized role approves; SciGround checks whether the correction qualifies.
- 03
Recheck the chain.
Data connection, calculation, and claim are verified again on the new version.
- 04
See what changed.
Old and new results sit side by side, each with its own evidence.
The current result, with its history
- A new research version with a rechecked result and claim
- The previous Record, preserved and marked superseded
- A side-by-side view of what changed and why
- A resumable state for any later AI session
Correction rules, roles and continuity.
Every change is a new version.
Changes to data, design, or rules create a new version. Earlier Records are preserved as history, and their conclusions stop being treated as current.
Every dependent step, rechecked.
After a correction, every step that depends on it—data connection, calculation, and claim—is checked again. If the evidence weakens, the claim is restated, for example from supported to not established.
Corrections that keep the guarantee.
Corrections that meet defined rules—such as fixing the exact values of planned samples, or restoring every row from originals registered with the design before any data arrived—return to a verified result after the recheck.
Pick up where you left off.
Every new session is pointed to the current research version, not an old conversation. Unresolved, pending, conflicting, and withdrawn states persist across sessions.
You decide who approves.
Proposing, approving, and owning are separate roles, and only permitted roles can change the research state. Whoever approves, a change still has to pass SciGround’s checks—no role can declare a result verified.
No silent overwrites.
Edits based on an outdated version, simultaneous updates, and changes made after permission is withdrawn are rejected instead of merged silently.