fix: QOTD/WOTD freshness — pick within the freshest cohort, not the rotated pool

Both selectors ordered candidates least-recently-shown, then daily.seeded_order()
ROTATED the whole list and took [0] — an arbitrary date-hashed item, undoing the
ordering. Result: repeats (quote id 2 on 6/28+6/29; word "harmony" on 6/25+6/28),
no guarantee a pool item is shown before it recurs.

Fix: daily.freshest(rows) returns the freshest cohort only — every NEVER-shown
item while any remain, else the oldest-shown group. quote/wotd _candidates use it;
seeded_order now picks deterministically WITHIN that cohort. So every pool item is
featured once before any repeat, then cycles oldest-first. Dropped the unused
_NO_REPEAT_POOL window. Tests: no-repeat-until-exhausted (quote + wotd) + a
freshest() unit test. 428 backend tests green.

(Separate follow-up: expand the QOTD pool from 16 → 90+ vetted public-domain
quotes for a longer no-repeat window.)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
jay
2026-06-29 05:39:06 -04:00
parent 414a4c4b8b
commit 0ae789752e
5 changed files with 64 additions and 13 deletions
+15
View File
@@ -33,6 +33,21 @@ def seeded_order(ids: list, date_str: str) -> list:
return ids[seed:] + ids[:seed] return ids[seed:] + ids[:seed]
def freshest(rows: list) -> list:
"""The cohort to feature today, from pool rows carrying `id` + `shown_at`: every
NEVER-shown item (shown_at NULL) while any remain, else every item tied for the
OLDEST shown_at. Guarantees each pool item is featured ONCE before any repeat, then
cycles oldest-first. Pick deterministically *within* this cohort (seeded_order) —
NEVER across the whole pool, which re-feeds recent items (the QOTD/WOTD repeat bug)."""
never = [r["id"] for r in rows if r["shown_at"] is None]
if never:
return sorted(never)
if not rows:
return []
oldest = min(r["shown_at"] for r in rows)
return sorted(r["id"] for r in rows if r["shown_at"] == oldest)
def content_key(*parts) -> str: def content_key(*parts) -> str:
"""A stable dedup key for a pool item (so re-harvesting never duplicates a row).""" """A stable dedup key for a pool item (so re-harvesting never duplicates a row)."""
raw = "|".join("" if p is None else str(p) for p in parts) raw = "|".join("" if p is None else str(p) for p in parts)
+4 -7
View File
@@ -14,8 +14,6 @@ import sqlite3
from . import daily from . import daily
from .localtime import local_today from .localtime import local_today
_NO_REPEAT_POOL = 60
# Public-domain (ancient / author died well over a century ago), uplifting. Admin curates. # Public-domain (ancient / author died well over a century ago), uplifting. Admin curates.
SEED = [ SEED = [
("Very little is needed to make a happy life; it is all within yourself, in your way of thinking.", "Marcus Aurelius", "Meditations"), ("Very little is needed to make a happy life; it is all within yourself, in your way of thinking.", "Marcus Aurelius", "Meditations"),
@@ -65,11 +63,10 @@ def _candidates(conn: sqlite3.Connection, avoid: int | None = None) -> list[int]
if featured: if featured:
ids = [r[0] for r in featured] ids = [r[0] for r in featured]
else: else:
rows = conn.execute( # The freshest cohort only (never-shown, else the oldest-shown group) — picking
"SELECT id FROM quote_pool WHERE blocked=0 ORDER BY shown_at IS NOT NULL, shown_at, id LIMIT ?", # across the whole pool is what re-fed recent quotes day to day.
(_NO_REPEAT_POOL,), rows = conn.execute("SELECT id, shown_at FROM quote_pool WHERE blocked=0").fetchall()
).fetchall() ids = daily.freshest(rows)
ids = [r[0] for r in rows]
if avoid is not None: if avoid is not None:
ids = [i for i in ids if i != avoid] or ids ids = [i for i in ids if i != avoid] or ids
return ids return ids
+4 -6
View File
@@ -25,7 +25,6 @@ from .localtime import local_today
DICT_BASE = "https://api.dictionaryapi.dev/api/v2/entries/en" DICT_BASE = "https://api.dictionaryapi.dev/api/v2/entries/en"
_UA = {"User-Agent": "upbeatBytes/1.0 (+https://upbeatbytes.com)"} _UA = {"User-Agent": "upbeatBytes/1.0 (+https://upbeatbytes.com)"}
_NO_REPEAT_POOL = 60
_TARGET_POOL = 30 # keep harvesting (a batch/day) until the pool reaches this _TARGET_POOL = 30 # keep harvesting (a batch/day) until the pool reaches this
_HARVEST_BATCH = 12 _HARVEST_BATCH = 12
_MIN_AUDIO_BYTES = 500 _MIN_AUDIO_BYTES = 500
@@ -222,11 +221,10 @@ def _candidates(conn: sqlite3.Connection, avoid: int | None = None) -> list[int]
if featured: if featured:
ids = [r[0] for r in featured] ids = [r[0] for r in featured]
else: else:
rows = conn.execute( # The freshest cohort only (never-shown, else the oldest-shown group) — picking
"SELECT id FROM wotd_pool WHERE blocked=0 ORDER BY shown_at IS NOT NULL, shown_at, id LIMIT ?", # across the whole pool is what re-fed recent words day to day.
(_NO_REPEAT_POOL,), rows = conn.execute("SELECT id, shown_at FROM wotd_pool WHERE blocked=0").fetchall()
).fetchall() ids = daily.freshest(rows)
ids = [r[0] for r in rows]
if avoid is not None: if avoid is not None:
ids = [i for i in ids if i != avoid] or ids ids = [i for i in ids if i != avoid] or ids
return ids return ids
+25
View File
@@ -56,3 +56,28 @@ def test_get_today_never_empty(conn):
def test_run_daily_seeds_then_picks(conn): def test_run_daily_seeds_then_picks(conn):
r = quote.run_daily(conn) r = quote.run_daily(conn)
assert r["pool"] == len(quote.SEED) and r["picked"] assert r["pool"] == len(quote.SEED) and r["picked"]
def test_no_repeat_until_pool_exhausted(conn):
"""Every quote is featured exactly once before ANY repeat; then the oldest-shown
repeats first. (Regression for the rotate-the-whole-pool selector bug.)"""
import datetime
quote.seed(conn)
n = len(quote.SEED)
d0 = datetime.date(2026, 1, 1)
picks = [quote.pick_daily(conn, feature_date=(d0 + datetime.timedelta(days=i)).isoformat())["pool_id"]
for i in range(n)]
assert len(set(picks)) == n # full coverage, no repeat within the pool
nxt = (d0 + datetime.timedelta(days=n)).isoformat()
assert quote.pick_daily(conn, feature_date=nxt)["pool_id"] == picks[0] # oldest repeats first
def test_freshest_cohort():
from goodnews import daily
# never-shown win outright (the oldest shown item is ignored while any never-shown remain)
assert daily.freshest([{"id": 1, "shown_at": "2026-01-02"},
{"id": 2, "shown_at": None}, {"id": 3, "shown_at": None}]) == [2, 3]
# all shown → only the oldest-shown cohort
assert daily.freshest([{"id": 1, "shown_at": "2026-01-03"},
{"id": 2, "shown_at": "2026-01-01"}, {"id": 3, "shown_at": "2026-01-01"}]) == [2, 3]
assert daily.freshest([]) == []
+16
View File
@@ -142,3 +142,19 @@ def test_polish_returns_none_with_empty_examples():
def chat_text(self, m): def chat_text(self, m):
return '{"gloss": "A warm clear gloss.", "examples": []}' return '{"gloss": "A warm clear gloss.", "examples": []}'
assert wotd._polish(C(), "serene", "adjective", "x") is None assert wotd._polish(C(), "serene", "adjective", "x") is None
def test_no_repeat_until_pool_exhausted(conn):
"""Same freshness guarantee as QOTD: every word featured once before any repeat,
then the oldest-shown repeats first. (Regression for 'harmony' repeating after 3 days.)"""
import datetime
for w in ["alpha", "bravo", "charlie", "delta"]:
conn.execute("INSERT INTO wotd_pool (word, definition) VALUES (?, 'a definition')", (w,))
conn.commit()
n = 4
d0 = datetime.date(2026, 2, 1)
picks = [wotd.pick_daily(conn, feature_date=(d0 + datetime.timedelta(days=i)).isoformat())["pool_id"]
for i in range(n)]
assert len(set(picks)) == n
nxt = (d0 + datetime.timedelta(days=n)).isoformat()
assert wotd.pick_daily(conn, feature_date=nxt)["pool_id"] == picks[0]