Maestro Read Results

Arguments routing

ARGUMENTS = everything the user typed after /maestro-read-results.

# If ARGUMENTS contains a path, use it directly:
python3 ${CLAUDE_SKILL_DIR}/scripts/read_results.py $ARGUMENTS

# If no path given, discover candidates first:
find ~/projects -name "maestro.sdb" -size +0c -maxdepth 10 2>/dev/null | head -20

Overview

maestro.sdb and active.state are plain XML files storing all simulation configuration, output expressions, corner definitions, and history paths. The PSF binary results are readable via the psf tool (/opt/cadence/IC231/bin/psf). This skill combines both to evaluate output expressions without any GUI.

Quick Start

# List available runs in a Maestro session
python3 ${CLAUDE_SKILL_DIR}/scripts/read_results.py /path/to/maestro --list

# Read all output values for the latest run
python3 ${CLAUDE_SKILL_DIR}/scripts/read_results.py /path/to/maestro

# Specific run or test
python3 ${CLAUDE_SKILL_DIR}/scripts/read_results.py /path/to/maestro --run ExplorerRun.0
python3 ${CLAUDE_SKILL_DIR}/scripts/read_results.py /path/to/maestro --test myTestName

Finding the Maestro Directory

The Maestro directory contains maestro.sdb. Common locations:

# From vcli maestro list-sessions output (psf_base field)
find ~/projects -name "maestro.sdb" -maxdepth 8 2>/dev/null | head -10

# Typical layout:
# <lib_dir>/<cell>/<view>/maestro/  (ADE Explorer)
# <sim_dir>/maestro/                 (Assembler)

The maestro_dir to pass is the directory containing maestro.sdb.

Output Format

{
  "test": "FT0001A_SH_5T_OTA_D_TO_S_sim_1",
  "run": "ExplorerRun.0",
  "timestamp": "Apr 21 22:53:48 2026",
  "psf_dir": "/path/to/psf",
  "outputs": [
    {
      "name": "gain_dc",
      "expression": "getData(\"net1\" ?result \"ac\")[0]",
      "eval_type": "point",
      "value": 0.3947,
      "dB": -8.075,
      "phase_deg": 179.88
    }
  ]
}

Supported Ocean Functions

Troubleshooting

psf tool not found: Check for alternate Cadence installs:

find /opt/cadence -name "psf" -type f 2>/dev/null

Then pass it via env: PSF_TOOL=/path/to/psf python3 ${CLAUDE_SKILL_DIR}/scripts/read_results.py ...

outputs: []: The active.state XML may use a different sevOutputStruct layout. Run with --list first to confirm the run directory is found, then inspect:

grep -o 'sevOutputStruct[^<]*' /path/to/maestro/active.state | head -5

Expression eval error: Print the raw expression from the JSON output and test manually:

python3 -c "
import sys; sys.path.insert(0, '${CLAUDE_SKILL_DIR}/scripts')
import read_results as r
# ... load psf data, call r.eval_expression(expr, env)
"

Wrong PSF directory: History uses $AXL_HISTORY_NAME variable. Check:

grep -o 'AXL_HISTORY_NAME[^"]*' /path/to/maestro/history.sdb | head -3

Parametric sweep subpoints — ASSEMBLER-2233: maeOpenResults(?history "Interactive.3/4") fails with (ASSEMBLER-2233) Could not Load Results — Maestro's results loader does not accept the N/M sub-point notation. Bypass Maestro and use OCEAN directly with the absolute PSF path instead:

# Find the per-sweep-point PSF directories
ls <sim_root>/<cell>/maestro/results/maestro/Interactive.3/
# → 0/  1/  2/  ...  (one directory per sweep point)

# Read one sweep point via OCEAN (bypasses maeOpenResults limitation)
vcli skill exec 'let((data)
  openResults("<sim_root>/cell/maestro/results/maestro/Interactive.3/4")
  data=getData("/VOUT" ?result "tran")
  ymax(data))'

This is the only reliable method for parametric sweep sub-point waveforms; read_results.py works the same way for offline reading.