"""Close calls of Reddit's The Button, by week: how often the countdown fell below 2 seconds, against chance.

Usage:  python3 button-close-calls.py thebutton_presses.csv
Data:   https://github.com/reddit/thebutton-data (thebutton_presses.csv, 1,008,315 rows). Standard library only.

A close call is a press made with less than 2 seconds left: flair "0s" or "1s", the time left that Reddit recorded at
the press. Chance: if a week's presses arrived at random at its average rate (λ presses a second), the countdown would
fall below 2 seconds before a press, that is a gap over 58 seconds, with probability e^(−58λ); times the week's
presses, that is the close calls chance would bring.

We count from the recorded time left, not from the gaps between timestamps: hundreds of gaps in the log are 60 seconds
or longer although the countdown only ran out once, so the timestamps are not precise enough to the second.
Rows without a press ("non presser", or no flair) are left out: 1,001,291 presses. Weeks are counted from the first.
(Used by unpressed.app/same-game since 27 Oct 2026, unpressed.app/why from 20 Oct; replaces the "protection factor" of research/insights-memo.md § 1.)
"""

import csv
import math
import statistics
import sys
from datetime import datetime

WEEK = 7 * 86400

presses = []
with open(sys.argv[1], newline="") as f:
    for row in csv.DictReader(f):
        flair = row["flair"]
        if not flair or flair == "non presser":
            continue
        presses.append((datetime.fromisoformat(row["press time"]).timestamp(), int(flair.rstrip("s"))))
presses.sort()
start = presses[0][0]

weeks = {}
for t, left in presses:
    weeks.setdefault(int((t - start) // WEEK) + 1, []).append((t, left))

print(f"{len(presses):,} presses")
print("week  presses/day  median left  under 11 s  close calls  by chance  share of chance")
for w, ps in sorted(weeks.items()):
    n = len(ps)
    rate = n / (ps[-1][0] - ps[0][0])  # presses a second
    lefts = [left for _, left in ps]
    close = sum(left <= 1 for left in lefts)
    chance = n * math.exp(-58 * rate)
    share = f"{100 * close / chance:.1f} %" if chance >= 1 else "–"
    print(
        f"{w:>4} {rate * 86400:>12,.0f} {statistics.median(lefts):>10.0f} s {100 * sum(l < 11 for l in lefts) / n:>9.1f} %"
        f" {close:>12,} {chance:>10,.0f} {share:>16}"
    )