Open questions
This section is informative.¶
These decisions remain open; each names a suggested owner and the phase it blocks. Questions the earlier documents raised and this draft settles (lineage standard, Tier A payload, Tier C format, platform lift use, use-by derivation, and storage: Postgres is the source of truth and BigQuery is a CDC mirror, section 20) are removed.¶
| Question | Suggested owner | Blocks |
|---|---|---|
| Who owns the shared-names registry, and what is the turnaround for a new placement?¶ | Framework owner | P0 |
| Default split rule when no better assumption exists¶ | Measurement lead | P1 |
| Which copulas, if any, are allowed as declared dependence assumptions?¶ | Measurement lead | P1 |
| Power-prior discount for class B platform Lift and weakly fitted class C tests¶ | Measurement lead | P2 |
| Benjamini–Hochberg q for calibration conflicts, and who approves overrides on accounts without a measurement lead. Settling q also resolves the jump in flagging between m = 1 and m >= 2 tests in one cycle¶ | Measurement lead | P2 |
| Does interval lift go upstream into an in-house MMM framework or live in a UMP calibration layer?¶ | Measurement lead | P2 |
| One acceptance bar for all families, or per-family pass marks?¶ | Framework owner | P2 |
| Standard and full materiality thresholds per pilot account¶ | Approver | P0 |
| Who is the framework owner, and who is the key custodian?¶ | Executive sponsor | P0 |
| Legal review: SR 26-2 scope text, EU AI Act position, erasure versus records retention¶ | Framework owner with legal | P1 |
| Excel policy: registered manual steps only, or an add-in¶ | Framework owner | P3 |
| Can Newton call our MCP or OpenAPI tool server, run in our GCP project, support reader and actor separation, and export session logs as run records?¶ | Agent host steward | P1 |
| Who writes and refreshes the golden question set, and how often?¶ | Measurement lead | P1 |
| What can Google’s geo planning tool export, and can results be pulled automatically?¶ | Experimentation analyst | P1 |
| Minimum segment sizes and query budgets per client contract¶ | Data owner | P2 |
| Name, licence and publishing home for the public spec; invite outside reviewers before v1?¶ | Framework owner | P4 |
Publishing base for schema $id values, problem type URIs and the profile URI. Until it is settled, the current values are identifiers that need not dereference, and the schema files are mirrored under /schemas/v1/ on the docs site¶ | Framework owner | P4 |
Grace window when a proposed threshold binds or changes. Default: the next scheduled refresh or 90 days (proposed), whichever is sooner, after which certificates are re-run and a failure moves the capability to unvalidated, not to failed (section 14)¶ | Measurement lead | P2 |
Added by the fallback-first gate design (Draft 2.1)¶
| Question | Suggested owner | Blocks |
|---|---|---|
| Default assumptions for undeclared estimand fields, by source type, and the plausible range each envelope covers¶ | Measurement lead | P0 |
| Default uncertainty to impute by tier and channel family when the library has nothing comparable¶ | Measurement lead | P1 |
| Which gates, if any, run strict by default for public claims or particular client contracts¶ | Framework owner | P1 |
| What share of decisions resting on a given fallback should trigger a test or a vendor data request¶ | Measurement lead | P2 |
| How the approver’s acknowledgement of fallbacks is shown in briefs without burying the recommendation¶ | Planning lead | P1 |