defmodule Microwaveprop.Pskr.FeatureBin do
  @moduledoc """
  One feature-bucket row from a `Pskr.RecalibrationRun`. Stores the
  spot-density statistics for a (band, feature, bin) cell of the
  corpus so an operator can read whether a feature actually
  discriminates propagation at the granularity the scorer cares
  about.

  Example: at run R, on the 10 GHz band, the `pwat_mm` feature has
  five rows (one per bin label: `<15`, `15-25`, `25-40`, `40-55`,
  `>55`). Each row carries:
    * `sample_count` — how many corpus samples landed in the bin
    * `spot_count_total` — sum of spot_count across those samples
    * `spot_count_avg` / `spot_count_p50` / `spot_count_p90` —
      central tendency + tail for the spot-density distribution

  The unique key (run_id, band, feature, bin_label) means a rerun
  of the same recalibration run is idempotent.
  """
  use Ecto.Schema

  import Ecto.Changeset

  @primary_key {:id, :binary_id, autogenerate: true}
  @foreign_key_type :binary_id

  schema "pskr_feature_bins" do
    belongs_to :run, Microwaveprop.Pskr.RecalibrationRun, foreign_key: :run_id

    field :band, :string
    field :feature, :string
    field :bin_label, :string
    field :bin_min, :float
    field :bin_max, :float
    field :sample_count, :integer, default: 0
    field :spot_count_total, :integer, default: 0
    field :spot_count_avg, :float
    field :spot_count_p50, :float
    field :spot_count_p90, :float

    timestamps(type: :utc_datetime)
  end

  @type t :: %__MODULE__{}

  @cast_fields ~w(run_id band feature bin_label bin_min bin_max sample_count
                  spot_count_total spot_count_avg spot_count_p50 spot_count_p90)a

  @required_fields ~w(run_id band feature bin_label sample_count)a

  @spec changeset(t(), map()) :: Ecto.Changeset.t()
  def changeset(record, attrs) do
    record
    |> cast(attrs, @cast_fields)
    |> validate_required(@required_fields)
    |> unique_constraint([:run_id, :band, :feature, :bin_label])
  end
end