v0.7.37

Fixed — a grade only where the log supports one

• A metric nothing says the direction of is no longer graded. For a run
  whose story metric has a name that does not say which way is better, with no
  .epochix.yaml to say either, the engine assumed higher was better and
  graded the improvement on that. A recorded CatBoost run whose AUC is 1.0
  every round came out D. It is now an "I", in both engines, and the card
  under it says it was not graded and why, in English, French and Persian,
  instead of claiming the letter came from thresholds or improvement.
• Awakening no longer says the run has just begun. It is told whenever a
  run has advanced less than 10% — by the clock when the run's length is
  known, otherwise by how much of the possible improvement it has realised —
  so "an opening reading" and "only beginning" were false of any run that
  stays near its first reading, at any epoch. The Awakening stories, for every
  task in all three languages, now state the reading and name the phase.
• The phase journey's flavour lines are gone — "first signal from noise",
  "beyond memorisation" and the rest were English-only and not drawn from the
  log.

Fixed — the learning curve's grade lines are the grader's

• The dashed grade lines did not match the grades. They were a typed copy
  of classification's bands (A+ 0.97, A 0.93, … C 0.55), drawn for every
  percentage metric. The engine grades classification at A+ 0.95, A 0.90, …
  C 0.60, detection's mAP from 0.75 down, and AUC on bands of its own: a run
  graded A+ at 0.95 sat below the chart's A+ line. The engine now records the
  bands a letter was measured against — a project's own .epochix.yaml bands
  included — and the chart draws exactly those, in the dashboard and the
  VS Code panel; a run graded on improvement shows none.
• The grade letters no longer pile up. A+ to C sit close together near
  the top; a letter too close to the one above is now left out, its line kept.