Prioritized roadmap

The order in which Fabric Desktop by KratosBI should build the items in the product backlog, and why. Every number on this page is computed from PRODUCT_BACKLOG.json by docs/product/tools/product_data.py, so changing a score or a weight and running product_data.py all re-ranks everything the same way each time.

Scoring model#

Each item gets seven scores from 1 to 5. Five are benefits and two are costs.

The calculation:

benefit  = weighted mean of the five benefit scores        (1 to 5)
cost     = weighted mean of complexity and risk            (1 to 5)
score    = benefit x (6 - cost) / 5                        (0.2 to 5)
priority = the product owner's decision, when one is recorded (priority_override)
           else P0 if mandatory is "blocker"
           P1 if score >= 2.6, or mandatory is "required"
           P2 if score >= 1.8
           P3 otherwise

(6 - cost) / 5 turns cost into a multiplier from 1.0 (cost 1) down to 0.2 (cost 5), so a cheap, safe item keeps its full benefit and an expensive, risky one keeps a fifth of it. The weights and thresholds live in the JSON scoring block; changing them is a product-owner decision and is recorded in the change log below.

Priority classes

Confidence says how sure the scores and effort are given the evidence (High: code read and tests run; Medium: some inference; Low: depends on a decision, a tenant behavior or another program's design). Effort: S is one focused PR, M is one larger PR or two, L is three to five PRs, XL is a program of its own.

Mandatory work is ranked separately#

Security, reliability and compliance work is not allowed to sink because its commercial score is low. An item is blocker when the launch can't happen without it and required when shipping without it would leave a known security, data-safety or reliability defect. Blockers are P0 and required items are never below P1, whatever their score. Each states why.

Product-owner priority decisions#

The product owner can set an item's priority directly. The computed score stays visible so the trade-off is recorded, and the decision is stored in the item's priority_override.

Sequencing by dependency#

An item that a higher-priority item depends on has to be built first, whatever its own score. These items are sequenced at the priority of the work that needs them:

Recommended sequence#

This is the reading of the ranking below, not a separate list. Items owned by specialist programs are sequenced by those programs; the order here is what the master product view recommends.

1. Launch blockers (P0). Chris creates the Azure Trusted Signing account so PB-001 can ship, and runs the tenant smoke test (PB-002) on his non-production workspace. Neither needs new product code beyond the package in pr-ready/PB-002.md.
2. Trust and correctness (mandatory P1, unassigned). Small, high-confidence fixes that remove wrong answers and wrong documents: local-parity defects (PB-013), Preview labels (PB-078), security and network docs (PB-010), Rust tests in CI (PB-071) and Fabric client retries (PB-051). Each has a PR-ready package. PB-010, PB-071 and PB-051 are now built by the Codebase audit and hardening program (PB-101), whose P0s (HD-001..009) fix data-safety, security and update-integrity defects and come before new feature work.
3. Enterprise readiness already merged. Run the implementation review on the merged enterprise PRs (audit trail #89, retention #95, backup #91, Ready to work #94), then token encryption (PB-005), CSP (PB-006), embedded-view hosts (PB-008) and release pinning (PB-064).
4. Local-first differentiation (P1, programs). The Spark API matrix (PB-012) and the AST analyzer (PB-014) from Prove, CU labelling (PB-050), preflight risk (PB-090) and the execution advisor (PB-017) from FinOps, and the SQL dialect analyzer (PB-022). Together they let a developer see, before running, what will work on the PC and what it saves.
5. Reliability of the app itself. Long engine calls stop blocking the app (PB-072) after Rust tests run in CI.
6. P2 in score order, starting with the run dialog (PB-067), capacity health (PB-088), docs truth for users (PB-077) and the SQL editor (PB-021).

Ranking#

P0 (5 items)#

P1 (30 items)#

P2 (47 items)#

P3 (11 items)#

Change log#