bspdn-goal-driverlisted

# BSPDN Goal Driver ## When to use Use this skill when the user asks to push objective/performance toward explicit PPA goals, not just run exploratory experiments. ## Goal contract Target A: - dynamic power reduction about `8%`, - area/timing non-worse (`delta_area<=0`, `delta_wns>=0`, `delta_tns>=0`). Target B: - power/area non-worse, - frequency uplift `>=5%` validated by period-sweep evidence. ## Milestone policy (incremental, mandatory) Do not pursue one-shot final target. Use staged gates: 1. M0: evidence-valid baseline and repeatable A/B flow (`execution-contract PASS`). 2. M1: `power <= -1%` with non-worse area/timing. 3. M2: `power <= -3%` with non-worse area/timing. 4. M3: `power <= -5%` with non-worse area/timing. 5. M4: final target A (`~ -8%`) or target B (`>=5%` frequency uplift with non-worse power/area). Promotion to next milestone requires current milestone PASS + retrospective note. ## Workflow 1. Build/update goal tracker: ```bash python3 scripts/debug/track_bspdn_goal_progress.py --campaign-md <ppa_md> --out-prefix <prefix> ``` 2. Select testcase tier: - Tier-1: high backside-usage potential with route feasibility, - Tier-2: mechanism-isolation testcase for fast iteration. 3. Run pre-submit checks: - `eda-preflight-reflect` - route smoke (`smoke_targeted_backside_route_toycase.sh`) for targeted net lists - execution policy lock (`vanilla_replace` primary baseline) 4. Execute A/B campaign with strict gates: - shared baseline and route policy, -