container_stats.py

#!/usr/bin/env python3
"""Statistics over listings.csv, shared by the dataset build and the PDF.

Both `build.py` and `scripts/build_reference_pdf.py` state per-vendor internal
dimensions as fact. They must therefore compute them the same way, from one
definition — hence this module rather than a copy in each.

Pure functions over already-loaded rows; nothing here reads a path except
`load_listings`, and nothing touches the network.
"""

import csv

# Listings that are not straight-walled open-top boxes, and so must not feed the
# internal-dimension statistics. Matched on the notes build_listings.py writes.
EXCLUDE_FROM_INTERNAL_STATS = ("Nesting or folding crate", "Sold with a lid")


def load_listings(path):
    if not path.exists():
        return []
    with path.open(encoding="utf-8") as f:
        return list(csv.DictReader(f))


def median(xs):
    s = sorted(xs)
    mid = len(s) // 2
    return s[mid] if len(s) % 2 else (s[mid - 1] + s[mid]) / 2


def footprint(row):
    """(type, L_mm, W_mm) for a listing, or None if unparseable."""
    try:
        return (row["type"], round(float(row["external_length_cm"]) * 10),
                round(float(row["external_width_cm"]) * 10))
    except ValueError:
        return None


def internal_index(listings):
    """(type, L_mm, W_mm) -> {vendor: (median_int_L, median_int_W, median_height_deduction)}, mm.

    One figure per *vendor*, not per listing. Salesbridges alone accounts for two
    thirds of the 600x400 listings, mostly colour variants of the same mould; a
    straight median over listings would just report Salesbridges' number and call
    it a consensus. Taking each vendor's median first gives every catalogue one
    vote, which is what the vendor capacity span means.
    """
    per_vendor = {}
    for r in listings:
        key = footprint(r)
        if (not key or not r["internal_length_cm"]
                or any(x in r["notes"] for x in EXCLUDE_FROM_INTERNAL_STATS)):
            continue
        try:
            il, iw = float(r["internal_length_cm"]) * 10, float(r["internal_width_cm"]) * 10
            ded = float(r["external_height_cm"]) * 10 - float(r["internal_height_cm"]) * 10
        except ValueError:
            continue
        per_vendor.setdefault(key, {}).setdefault(r["vendor"], []).append((il, iw, ded))

    return {
        key: {v: tuple(median([o[i] for o in obs]) for i in range(3))
              for v, obs in vendors.items()}
        for key, vendors in per_vendor.items()
    }


def vendor_listing_counts(listings):
    """(type, L_mm, W_mm) -> {vendor: n listings publishing internal dimensions}."""
    counts = {}
    for r in listings:
        key = footprint(r)
        if (not key or not r["internal_length_cm"]
                or any(x in r["notes"] for x in EXCLUDE_FROM_INTERNAL_STATS)):
            continue
        counts.setdefault(key, {})
        counts[key][r["vendor"]] = counts[key].get(r["vendor"], 0) + 1
    return counts


def capacity_span(index, kind, L, W, H):
    """(low, high) litres implied by the least and most generous vendor, or ("", "").

    Same box, same external size, different published internal dimensions: at
    600x400 the vendors range from 555x355 (Plastic Box Shop, measured near the
    base) to 570x370 (Salesbridges, the top opening), which is an 11 % spread in
    litres. Quoting one number without that range would be false precision.
    """
    vendors = index.get((kind, L, W))
    if not vendors:
        return "", ""
    caps = [il * iw * max(H - ded, 0) / 1e6 for il, iw, ded in vendors.values()]
    return round(min(caps), 1), round(max(caps), 1)