Decisions and Risks
| Decision | Reason | Remaining limitation |
|---|---|---|
| Contribution map with owners | Make accountability and the output/adoption/benefit distinction inspectable | A displayed owner does not prove acceptance or actual accountability |
| Explicit exposure compatibility | Prevent wrong-population subtraction | Matching exposure does not remove confounding |
| Shared workflow atom union | Avoid counting lookup twice | Atom durations and full use are hypothetical |
| Separate observed and forecast figures | Avoid unsupported combined benefit | Equal values can still be misunderstood; the UI explicitly calls them coincidental |
| Human direction with immutable snapshot | Preserve reasoning and known gaps | A chosen continue direction is not certification |
| Append-only withdrawal | Correct a decision without rewriting original evidence | No multi-user collaboration or audit authority |
| Twelve-review boundary | Keep a bounded understandable local sample | Export and reviewed reset are needed for more history |
| Exact raw/readability comparison | Reject stale reviews and reset previews | Browser storage has no cross-tab atomic compare-and-set; confirmation detects changes observed at its check, not all possible simultaneous races |
| Preserve invalid saved bytes | Avoid silently erasing evidence | Invalid data cannot be imported into active history; export current memory and explicit reset are available |
| Memory-only write failure | Keep recoverable work visible | Refresh or close can lose it; the UI states that consequence |
All data are fictional. There is no authentication, backend, analytics, live AI, certified ROI, or automated external action. Local storage and downloaded reports remain under the browser user’s control. Missing or incompatible evidence never changes the raw observation into a demonstrated benefit.
Open product questions: Would reviewers distinguish the matching 60 values? Is the atom rule intelligible without training? Would owners want a specific next evidence collection action? Would a withdrawal reason satisfy their review process? These are proposed research questions, not findings.