
  ███╗   ███╗██╗ ██████╗██╗  ██╗██╗
  ████╗ ████║██║██╔════╝██║  ██║██║
  ██╔████╔██║██║██║     ███████║██║      道 
  ██║╚██╔╝██║██║██║     ██╔══██║██║
  ██║ ╚═╝ ██║██║╚██████╗██║  ██║██║
  ╚═╝     ╚═╝╚═╝ ╚═════╝╚═╝  ╚═╝╚═╝

        =[ michi v1.1.0  ·  a local-first ML workbench                        ]
+ -- --=[ 9 verbs  ·  14 models  ·  12 recipe ops  ·  26 explanations        ]
+ -- --=[ 19 console commands  ·  no account, no telemetry, no network call  ]

  心得  michi lists the options; you choose.
  path walks the stages · help lists commands · use <file> loads data

  tip: `clean` never modifies your data. It writes a recipe; `apply` produces a
       new file, always at a path you name.

michi › use data/customers.csv
loaded customers.csv — 14 columns
michi (customers.csv) › set target purchased
target = purchased
michi (customers.csv → purchased) › path

  ✓   見   miru — see what you have             inspect   inspect --explain                 
  ✓   整   totonoeru — decide what to fix       clean     clean --target <col>              
  ✓   直   naosu — produce the fixed data       apply     apply -o clean.parquet            
  ✓   比   kuraberu — compare models honestly   bench     bench --models …                  
  ✓   確   tashikameru — verify one model       eval      eval <model.pkl>                  
  ✓   探   sagasu — search a grid               sweep     sweep sweep.yaml                  
  ·   記   shirusu — write up what happened     report    report runs/                      
  ✓   出   dasu — take the code and go          export    export -o pipeline.py             
  ✓ marks a stage this context can run right now. Every stage stands alone —
  run one and stop, or none of them. michi walks none of it for you.

