# Ppa Diagnose

> Explain why a PPA number is where it is — which paths, cells, nets or flow decisions account for a timing, power or area result — starting from what the deterministic diagnosis already found and adding only falsifiable hypotheses on top. Use when the user says "why is timing failing", "where is the area going", "what is burning the power", "diagnose this PPA result", or when a closure loop needs a reason before it needs a candidate.

- Skill: `vibeic/ppa-diagnose` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx get-skill add vibeic/ppa-diagnose`
- Raw SKILL.md: https://api.skillmd.com/api/skills/vibeic/ppa-diagnose/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: vibeic (https://skillmd.com/u/vibeic)
- Updated: 2026-11-15
- Page: https://skillmd.com/skills/vibeic/ppa-diagnose

---


# Diagnose PPA

Read the selected rows in `<project>/reports/flow_engine.json` and open their
native `reports`. `flow/steps.json` names each current tool and
`flow/native_limits.json` retains its original limits. Review `tool_ran`,
`native_steps`, `native_metrics`, and each metric's value, operator and limit.
A missing native measurement is `NOT_MEASURED`. AI reviews the engineering
meaning of every result and fixes problems in the design, tool or plugin.

```bash
python3 programs/flow_engine.py report --project <project>
```

Start with the measured failing metric and the original requirement. Read the
native path, cell, net, power or congestion breakdown that accounts for it.

1. For timing, separate cell delay, interconnect, logic depth, fanout and skew.
   Check constraints and I/O loading before proposing an RTL change.
2. For area, identify the dominant cells, memories, macros and unused reserve.
   Keep intentional ECO spare area separate from removable logic.
3. For power, identify activity, capacitance, voltage, leakage and the dominant
   switching structures. State any missing activity or model coverage.
4. For congestion or IR, connect the hotspot to the actual geometry, grid and
   measured demand.
5. Propose a bounded change and a falsifiable expected effect. Preserve the
   previous design, run the owning tools, and compare the same metric and scope.

Report observations separately from hypotheses. Cite source paths and native
numbers. An untested explanation remains a hypothesis. Keep original limits
and disclose tradeoffs in every other affected metric. No context receipt or
separate diagnostic authorization is required.

Large repeated physical-verification counts need a setup diagnosis before a
claim about the design. Confirm the actual GDS/DEF/Magic view selected by the
step, merged cell geometry, cell GDS/CDL pin interfaces, metal/via/track maps,
taps/endcaps/fillers and supply/substrate connections. Group native markers by
rule and inspect their coordinates. Report the first LVS mismatch class and
separate missing library annotations from device/net/property differences.

A measured GF180 7-track case illustrates both causes. A vendor two-supply
GDS/LEF set lacked the declared library's explicit well/substrate pin texts;
DEF-based extraction and a powered netlist from that foreign database produced
thousands of spurious LVS mismatches. Native declared-SCL power connections
and missing library texts restored unique LVS matching without changing any
source polygon. The same layouts had real PDN errors: 0.26 µm cuts repeated
at 0.128 µm pitch overlapped, and a 100 µm tap-placement distance violated the
15 µm tap-distance rules. Name the measured shipped platform revision and
actual rule families; do not generalize its result to every ORFS version.

