# Chipbase OpenROAD runs 62 complete runs of the open physical design flow, OpenROAD-flow-scripts, in the openroad/orfs container (OpenROAD 26Q3). They are the runs behind the "Your move" and "Read the die" drills on the Chipbase site, published so that anybody can check a number or use the runs for something else. ## What is in the archive - `runs.json` and `runs.csv`: one row per run. Design, platform, every setting changed from the flow's defaults, and the measured results. - `whatif-src/` and `pathology-src/`: for each run, the full log (`.log`) and the reports the flow wrote (`/*.rpt`), plus the rendered maps for the pathology runs. - `run*.sh` and `run_pathology.sh`: the scripts that launched the runs, one container per group. - `thin_grid.tcl`, `thin_grid_gcd.tcl`, `dense_grid.tcl`: the modified power grids some runs used. Each is the platform's or the design's own file with strap widths or pitches changed, and says what changed at the top. `gcd_grid_original.tcl` is the unmodified file, for comparison. - `dataset.py`: the code that read every number out of the logs and built this archive, so any value can be checked by running it again. ## How the numbers were read Every value comes from the line of the log or the final report that printed it. Nothing is typed from memory, and a value a run never printed is left empty rather than set to zero. - **Wire** is the detailed router's routed length, printed last. Global routing prints an estimate earlier that can differ by more than a quarter. - **Smallest period** is `period_min` for the real clock. The virtual clock the constraints create also reports one, always zero, and is excluded. - **Worst setup slack** is `report_worst_slack`, because `report_wns` clamps at zero and hides the margin of a design that passes. - **Worst drop** is reported for both the supply and the ground net. The map the flow saves is drawn on the ground net. - **Congestion reports** counts two messages: global routing running extra iterations to clear congestion, and the placer failing its routability target. - **Time** is in picoseconds on ASAP7 and nanoseconds elsewhere. ## Two properties worth knowing The flow is deterministic here. Two independent runs of ibex, in different containers on different days, agreed on every published metric to the last digit, so a small difference between two runs is a real difference. Most runs change one setting against a baseline of the same design with no changes, which makes the set usable for sensitivity studies: the effect of one knob, with everything else held. ## What is not in it Runtime. Runs were executed in parallel on one machine, so wall-clock time would measure the scheduler rather than the run. Every launched run finished and is included.