Open questions, superseded prose and implementation gaps¶
Snapshot: 2026-09-29. This is not a new release checklist and does not turn unapproved suggestions into required work. Keep three classes separate: a product decision is missing, an accepted decision is not implemented or verified, and historical documentation needs reconciliation.
Decisions not settled by existing authority¶
| ID | Question | Safe disposition and next authority |
|---|---|---|
| Q-01 | Should different DATETIME spellings/offsets be equal mapping keys because they denote the same instant? | Not implied by field formatting or timezone conversion. Retain unsupported/fail-closed execution where no contract exists; owner must select equality/collision semantics before expanding it. ADR-0023. |
| Q-02 | What are nullable internal-formula semantics? | Deferred, not a 1.6 RC blocker. Existing private code remains non-public; a future formula feature needs its own contract. ADR-0026. |
| Q-03 | May a narrow DECIMAL preservation exception override a numeric-shape false positive? | Not active. Provenance of sensitivity evidence must first distinguish shape-only from semantic/content evidence, with exact binding, executable tests and independent review. Do not infer permission from a non-sensitive declaration. ADR-0008, ADR-0021. |
| Q-04 | May external scripts/API callbacks act as replacement providers? | Unapproved suggestion, not implementation scope. Keep unsupported; owner must explicitly select a trust, data-access and execution boundary before any implementation. ADR-0022. |
No approval is requested as part of this documentation task. In particular, Q-02/Q-04 must not be used to stall already-agreed work. If Q-01 or Q-03 is needed for a concrete route, identify that route and ask the narrow question then.
Accepted decisions with remaining engineering evidence¶
| ID | Remaining work | What must not be claimed |
|---|---|---|
| I-01 | Complete the common transformation core and public wizard/noninteractive/agent execution with existing approval gates. | A private helper, temporary publisher or read-only transform-review is not installed public execution. |
| I-02 | Demonstrate capacity across participating SQL capture, mappings, review, execution and output stages; complete required route evidence. | CSV replacement 300k×50 does not prove SQL scale, all twelve routes or arbitrary-width 1M×100 capacity. |
| I-03 | Finish accepted key/domain/relationship and supported native-type integration. | Type parsing, a domain model or one adapter test is not complete cross-input relationship acceptance. |
| I-04 | Reconcile original client findings and inspect/run only available scripts on fictional baseline/candidate inputs. | Private/unavailable client data and future integration feedback are not passed evidence. |
| I-05 | Complete safety activation evidence, exact-candidate independent review, required GitHub gates and public RC verification. | Accepted architecture and AI evidence do not replace human/platform approval or authorize stable 1.6.0. |
The uncommitted query-limit implementation on codex/1-6-query-limits is a WIP
checkpoint with focused evidence in progress.md, not a new approved public
adapter contract. No new hard-RSS/wire guarantee or platform restriction is added
by I-02. Public API details not yet accepted remain design work behind the gate,
not facts reconstructed from private function signatures.
Known documentary conflicts¶
| ID | Earlier text | Governing decision and reconciliation |
|---|---|---|
| D-01 | Later sections of policy-contract.md and safety-boundary.md still describe internal totals/formula recomputation as transformation acceptance. | Explicit owner scope correction on 2026-09-28 and ADR-0026 supersede that requirement. Keep historical/private evidence; do not delete source-free rule support. |
| D-02 | Older policy/report prose describes retention as equality and formatted output percentages as unavailable. | ADR-0025 governs the user-facing three-origin report. Equality safety checks and historical manifest-v1 evidence remain distinct. |
| D-03 | Earlier policy prose treats same-instant DATETIME values as identity replacements, while the later owner clarification explicitly leaves temporal-key equivalence undecided. | Do not promote that sentence into a complete execution contract. Q-01 remains open; the recorded formatting decision does not resolve it. |
| D-04 | Initial-slice policy text says the CSV parser does not change the process-global field limit. | PR #594 introduced scoped coordination/restoration for internal CSV reads. This is stale implementation narration, not a reason to undo accepted scalar-limit support. External unsynchronized changes remain outside the coordination guarantee. |
| D-05 | The general roadmap still names 1.3.x as current work, while release evidence records 1.5.0 and the active change targets 1.6.0rc1. | Use immutable release evidence for shipped status and the active owner-approved plan for current scope. Roadmap reconciliation must not invent dates, integrations or new priorities. |
| D-06 | Historical progress/tasks retain early helper limitations and obsolete blockers. | Use the most recent explicit decision and exact evidence, not the oldest unchecked sentence. A task checkbox is neither authority nor an equal-sized estimate of remaining effort. |
The register preserves these conflicts visibly rather than silently rewriting history or declaring the entire prose corpus reconciled. The notes added to the active decision documents point readers here; detailed cleanup belongs to the next documentation pass for the affected supported behavior.