Appendix B — Questions to answer next

A live snapshot of TODO.md at the repo root, kept in the appendix so the roadmap is always in front of us. Edit TODO.md and this page updates on the next render.

from pathlib import Path
from IPython.display import Markdown


def _find_todo() -> Path:
    # walk up from the notebook working dir to the repo root
    for d in [Path.cwd(), *Path.cwd().parents]:
        if (d / "TODO.md").exists():
            return d / "TODO.md"
    raise FileNotFoundError("TODO.md not found above notebooks/")


_text = _find_todo().read_text()
# this chapter supplies the appendix heading, so drop the file's own H1
_lines = _text.splitlines()
if _lines and _lines[0].startswith("# "):
    _lines = _lines[1:]
    while _lines and not _lines[0].strip():
        _lines.pop(0)
Markdown("\n".join(_lines))

The book is organized as questions (see notebooks/). These are candidate questions for future chapters or sections, with notes on feasibility given the data we already pull from Neon (accounts, donations, events, registrations).

B.1 Giving

• Do class-takers and patrons become donors? First-activity date vs first-gift date per household: how often does participation precede giving, and by how long? (Feasible now: registered_at + donation_date.)
• Who has lapsed? Donor households that gave in earlier years but not recently — recency/frequency segmentation, and how much annual giving the lapsed group used to represent. (Feasible now.)
• How concentrated is giving? Share of lifetime dollars from the top 10 / 50 households; dependence on a few large gifts, and whether concentration is rising. (Feasible now; publish only aggregate shares, never named households.)
• When during the year do donations arrive? Seasonality, year-end share, spikes around events or appeals. (Feasible now.)
• Do donors who attend give more than donors who don’t? Giving levels by engagement tier (donate-only vs donate+attend vs all three). (Feasible now via households_summary.)

B.2 Audience

• How far in advance do people buy tickets? Lead time (registered_at to starts_on) by genre and day of week; the walk-up share. (Feasible; first check how much of registered_at is real timestamps vs batch-entered dates.)
• Which shows bring in first-timers? For each production, how many attending households were brand new — and did they ever come back? Acquisition value of programming. (Feasible now; builds on the production mapping.)
• How many attendees come back? Repeat rate and one-and-done share by genre; do class households attend more performances than performance-only households attend classes? (Feasible now.)
• Has the audience gotten more or less local over time? Distance mix of registrations by period, overall and by genre — the audience analogue of the donor-geography over-time section. (Feasible now.)
• Are sell-outs predictable? “SOLD OUT” in event names is a weak label; check whether the events extract includes a capacity field so utilization can be computed properly. (Needs a look at events.jsonl — possible data gap.)

B.3 Data gaps to investigate

• Event capacity (for utilization / sell-out analysis) — is it populated in Neon?
• Memberships — if Hubbard Hall uses Neon’s membership module, pulling it would open “do members behave differently?” questions.
• Campaign/appeal codes on donations — would let “when do donations arrive?” separate appeal-driven from spontaneous gifts.

B.4 Fall 2026 mailing list — deliverables

• Separate printer CSV (Don, 2026-09-15). The printer needs its own file rather than a tab inside the workbook: the printer columns only (id, mailing_name, salutation, address, city, state, zip) in the printer sort (surname, then first listed name). Plain CSV, not xlsx - it is a label feed, not something anyone reads. Watch the zip: it must stay 5-digit text with leading zeros, so decide deliberately whether the consumer wants the ="05773" formula wrapper (needed only if it will be opened in Sheets/Excel) or a bare quoted 05773 (right for a mail-merge that reads the file directly). Ask the printer which before generating.