Profile data and plot the evidence

// Twelve synthetic observations: teaching data, not measured research evidence.
// Run from the repository root. No Python runtime or map packs are needed.
import aner.data
import aner.plot
import aner.units

var schema = data.schema()
schema = schema.column("row_id", "Int64", false)
schema = schema.column("RecordedAt", "DateTime", false)
schema = schema.unit_column("Temperature_DegC", "degC", true)
schema = schema.unit_column("Distance_cm", "cm", false)
schema = schema.column("signal", "Float64", true)
schema = schema.column("cohort", "String", false)
schema = schema.column("active", "Bool", false)
schema = schema.column("note", "String", false)
let source = data.scan_csv("examples/fixtures/data_profiling.csv", schema, "NA")

// Profile and retain the same validated rows in one source pass. This small
// sample bound retains all present values; larger sources disclose sampling.
let table = source.with_profile(sample_rows: 16, max_categories: 16)
    .collect(max_rows: 0, max_bytes: 1048576, batch_rows: 3)
let report = table.profile()
print("Profile attached to a complete immutable table")
print(table.has_profile() && report.is_profile() && report.complete() && report.origin_complete())
print(table.rows() == 12 && report.rows() == 12 && report.scanned_rows() == 12)
print(report.summary())
print("Exact observation ID extrema and nearest-rank 50th percentile")
print(report.minimum_text("row_id"))
print(report.maximum_text("row_id"))
print(report.quantile_text("row_id", q: 0.5))
print("UTC timestamp bounds preserve exact temporal ordering")
print(report.minimum_text("RecordedAt"))
print(report.maximum_text("RecordedAt"))
print("Missing temperature, observed temperature mean and interval spread")
print(report.missing("Temperature_DegC"))
print(report.unit_mean("Temperature_DegC", "degC"))
print(report.unit_std("Temperature_DegC", "delta_degC"))
print("Exact category diagnostics and Boolean counts")
print(report.distinct_complete("cohort"))
print(report.distinct_count("cohort"))
print(report.true_count("active"))
print(report.false_count("active"))
print("Empty string is present when only NA is the missing token")
print(report.empty_count("note"))
print(report.missing("note"))

// Cached access returns the same report without re-reading CSV or table values.
let cached = table.profile()
print("Cached structured profile retains its population and execution lineage")
print(cached.profile_json() == report.profile_json())
print(cached.provenance() == report.provenance())

// Derived data has its own selected population, recipe and profile. It leaves
// the raw table unchanged; no source CSV is reopened by this snapshot scan.
let warm = table.filter(data.col("Temperature_DegC") >= 30.0@degC)
    .select(List("row_id", "Temperature_DegC", "signal", "cohort"))
    .with_profile()
    .collect(max_rows: 0, max_bytes: 1048576, batch_rows: 2)
print("Source and selected profiles describe different populations")
print(table.rows() == 12 && warm.rows() == 6 && warm.profile().rows() == 6)
print(warm.profile().missing("Temperature_DegC") == 0)

// A small retained sample still keeps exact full-population counts/extrema.
// This is a bounded standalone traversal of the snapshot, not of the CSV.
let sampled = table.scan_table().profile(max_bytes: 1048576, batch_rows: 2,
    sample_rows: 4, max_categories: 2)
print("Sampling and category overflow are explicit")
print(!sampled.quantiles_exact("signal"))
print(sampled.sample_count("signal") == 4)
print(!sampled.distinct_complete("cohort"))
print(sampled.missing("signal") == 2 && sampled.maximum_text("signal") == "12")

// A validated prefix is useful for feedback; it is not a complete-file profile.
let prefix = source.bounded(max_scanned_rows: 4, max_read_bytes: 0, timeout_ms: 0)
    .profile(max_bytes: 1048576, batch_rows: 2)
print("Prefix profiling preserves its stop reason")
print(!prefix.complete() && prefix.stop_reason() == "scan_rows")
print(prefix.rows() == 4 && prefix.scanned_rows() == 4)

// Reading all rows of a preview snapshot completes that snapshot, while its
// lineage continues to disclose the unread original acquisition tail.
let preview_table = source.with_profile().preview(4, 1048576, 2)
let snapshot_report = preview_table.scan_table().profile(1048576, 2)
print("Complete snapshot profiling retains incomplete acquisition ancestry")
print(snapshot_report.complete() && !snapshot_report.origin_complete())
print(snapshot_report.rows() == 4 && !preview_table.origin_complete())

// Named native adapters carry physical labels, lineage and approximation
// disclosures into ordinary immutable charts. Charts need no scalar loops.
let missingness = plot.missingness(report)
    .title("Synthetic sensors: missing observations")
let distribution = plot.distribution(sampled, "signal", bins: 5)
    .title("Synthetic signal: bounded sample distribution")
let cohorts = plot.categories(report, "cohort")
    .title("Synthetic cohorts: exact observed counts")
let relationship = plot.scatter(table, x: "Distance_cm", y: "Temperature_DegC",
    max_points: 12)
    .title("Synthetic temperature and distance")
print("Native charts retain structured lineage")
print(missingness.provenance().contains("source_sha256"))
print(relationship.provenance().contains("max_points"))
missingness.show()
distribution.show()
cohorts.show()
relationship.show()
// Archive report.profile_json(), report.provenance(), chart.provenance(), the
// source snapshot and this program with scientific results. See the guide for
// training/held-out separation and limits; a profile does not prove validity.