12-week roadmap

Rolling plan from week 1 (starting Monday 2026-11-02) to week 12 (ending 2027-01-24). The weekly review moves it on by a week, records what changed and why, and keeps it in line with the work queues. Items with a PB id are in the backlog. Anything marked needs Chris waits for his mockup approval, his go on a new feature, or his yes on capacity spend.

Why this order: as of 2026-10-28 about 370 PRs merged into main in the week from 2026-10-21, main CI is green, and 3 of 142 features are verified end to end. A Fortune 100 rollout needs proof and reliability before breadth, so weeks 1 to 4 turn built work into verified work.

Weeks 1 to 4: stabilize and verify (2026-11-02 to 2026-11-29)#

Workstream	Goal	Stories
Microsoft Fabric	Live evidence for the core flows	PB-002 ten-flow tenant smoke test in a throwaway workspace on Trial capacity (needs Chris: capacity yes); PB-003 OneLake leases and MERGE writes live; PB-043 cancel a Fabric run where people watch it; PB-013 last parity defect
Platform Engineering	Nothing freezes, nothing ships unaudited	PB-074 release waits for the audit job; PB-072 every long engine call has a limit and never blocks a quick one; PB-051 REST client retries and token refresh; PB-001 signing (waits on Microsoft); PB-010 security docs match the code
Local Execution	Correct numbers and the most common write pattern on the PC	PB-094 kernel-process measurements; PB-068 key lookups without whole-table reads; PB-015 MERGE, UPDATE and DELETE semantics (after PB-012)
Enterprise Services	Policies checked without Fabric	PB-057 offline policy validation; PB-083 endorsed-items wording; PB-053 scheduled shortcut health check; PB-052 central platform on a tenant (needs Chris)
Developer Experience	Docs and Preview tags tell the truth; editor regressions get caught	PB-078 Preview ids for every Preview feature; PB-077 docs match the code; PB-070 notebook editor UI tests
Databricks and Snowflake	Architecture before code	Fabric Spark (Livy) Preview engine draft; Snowflake audit, ADRs and parity matrix (research only); Databricks conversion loss report
All	Acceptance	One acceptance batch for the 39 items In review, with evidence prepared by quality

Exit: PB-002 run with defects filed; no P0 or P1 reliability item open without an owner; every In review item accepted, rejected or given a dated fix.

Weeks 5 to 8: expand the three platforms (2026-11-30 to 2026-12-27)#

Workstream	Goal	Candidate stories
Databricks and Snowflake	Databricks verified live; Snowflake first slice	Databricks Connect and SQL warehouse on a live workspace (needs Chris: workspace and spend); Asset Bundles validate and deploy through the Databricks CLI; first Snowflake slice from its ADRs, likely SQL validation and local execution through a Snowflake SQL dialect (needs Chris: go)
Local Execution	Pick the right engine for each workload	Engine recommendation across local Python, local SQL, Fabric and Databricks, built on the PB-017 advisor; PB-073 process kernel by default with limits; PB-067 bounded-memory runs; real local Spark detection
Microsoft Fabric	Deployment and Git parity	PB-040 Fabric workspace Git integration; PB-041 deployment preview and item deploy; PB-049 pipeline mailbox on a tenant (needs Chris); PB-048 schedule editing (needs Chris: mockup)
Platform Engineering	Release confidence	PB-063 assistive technology audit; PB-061 missing policy keys; release notes from merged PRs
Developer Experience	Engine picker in the notebook	Engine picker UI from the approved mockup (needs Chris: mockup review)

Weeks 9 to 12: developer experience, enterprise services, optimization (2026-12-28 to 2027-01-24)#

Workstream	Goal	Candidate stories
Developer Experience	Debug and test like a desktop IDE	Debugger maturity (PB-026); PB-085 stale-cell marking; PB-034 notebook wizard from a table; PB-027 column completion
Local Execution and Enterprise Services	Defensible savings	Measured vs estimated CU rolled up across installations in the central Optimize lakehouse, with the method documented (PB-089, PB-050)
Enterprise Services	Data quality and lineage	PB-060 sensitive data classification beside Purview; PB-019 late-arriving rows; PB-031 schema evolution
Databricks and Snowflake	Snowflake parity grows	Snowpark local testing, dbt projects, deployment, per the Snowflake plan
Platform Engineering	Scale	Multiple workers (workloads design), PB-016 reproducible environments

Changes#

Week of	Change	Why
2026-10-28	First version	Orchestrator brief; built from the backlog, inventory and open PRs on main d0747a6
2026-10-28 (evening)	Queues refreshed after the first day's work; a limit of four PRs waiting for review per workstream	Review and CI became the bottleneck: about 20 drafts waited for review and CI queued over 100 runs