sudacode/SubMiner
1
0 Fork 0
mirror of https://github.com/ksyasuda/SubMiner.git synced 2026-11-10 04:46:07 -08:00
Code Issues Packages Projects Releases Wiki Activity

fix(stats): keep first_seen for words with retention-pruned history

Browse Source 
- Only repair stale first_seen when frequency matches surviving occurrences
- Rows whose older lines were pruned (or predate occurrence tables) keep their original first_seen
- Extend rollup rebuild test with a retention-pruned word
This commit is contained in:
sudacode committed 2026-10-01 23:27:40 -07:00
1 parent ce680b746a
commit bf2ea99178
2 files changed
+19 -6
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/core/services/immersion-tracker/lexical-rollups.test.ts
+8 -5
View File
@@ -201,15 +201,17 @@ test('lexical rollup rebuild repairs seen bounds left pointing at deleted histor
const staleLast = seen + 86_400.631;
db.prepare(
`INSERT INTO imm_words(id, headword, word, reading, first_seen, last_seen, frequency)
VALUES (1, '猫', '猫', 'ねこ', ?, ?, 1), (2, '犬', '犬', 'いぬ', ?, ?, 1)`,
).run(staleFirst, staleLast, seen - 0.1, seen - 0.1);
VALUES (1, '猫', '猫', 'ねこ', ?, ?, 1), (2, '犬', '犬', 'いぬ', ?, ?, 1),
(3, '鳥', '鳥', 'とり', ?, ?, 5)`,
// Word 3's earlier history was pruned by retention: frequency still counts it.
).run(staleFirst, staleLast, seen - 0.1, seen - 0.1, staleFirst, seen);
db.prepare(
`INSERT INTO imm_kanji(id, kanji, first_seen, last_seen, frequency) VALUES (1, '猫', ?, ?, 1)`,
).run(staleFirst, staleLast);
db.prepare(
`INSERT INTO imm_word_line_occurrences(line_id, word_id, occurrence_count, seen_ms)
VALUES (?, 1, 1, ?), (?, 2, 1, ?)`,
).run(lineId, BASE_MS + 251, lineId, BASE_MS + 251);
VALUES (?, 1, 1, ?), (?, 2, 1, ?), (?, 3, 1, ?)`,
).run(lineId, BASE_MS + 251, lineId, BASE_MS + 251, lineId, BASE_MS + 251);
db.prepare(
`INSERT INTO imm_kanji_line_occurrences(line_id, kanji_id, occurrence_count, seen_ms)
VALUES (?, 1, 1, ?)`,
@@ -231,8 +233,9 @@ test('lexical rollup rebuild repairs seen bounds left pointing at deleted histor
[
[seen, seen],
[seen - 0.1, seen - 0.1],
[staleFirst, seen],
],
'healthy rows within flush lag are untouched',
'flush-lag and retention-pruned rows keep their first_seen',
);
assert.deepEqual(bounds('imm_kanji'), [[seen, seen]]);
} finally {
src/core/services/immersion-tracker/lexical-rollups.ts
+11 -1
View File
@@ -35,6 +35,12 @@ export function localEpochDaySql(value: string): string {
// stamped at queue time, before seen_ms is stamped at flush, so healthy rows
// never exceed their newest occurrence and only trail their oldest by flush lag.
// Rows with undated occurrences are skipped.
//
// first_seen can also legitimately predate every surviving occurrence: retention
// pruning cascades away old lines, and early history predates the occurrence
// tables, while both keep the lifetime frequency. Deletes subtract frequency
// exactly, so only rows whose frequency matches their surviving occurrences have
// their whole history on hand and a provably stale first_seen.
function repairStaleSeenBounds(db: DatabaseSync): void {
for (const [table, occurrenceTable, col] of [
['imm_words', 'imm_word_line_occurrences', 'word_id'],
@@ -48,7 +54,11 @@ function repairStaleSeenBounds(db: DatabaseSync): void {
db.exec(`
UPDATE ${table}
SET first_seen = ${seenSec('MIN')}
WHERE first_seen < ${seenSec('MIN')} - ${FIRST_SEEN_FLUSH_LAG_SLACK_SEC} AND ${hasNoUndated};
WHERE first_seen < ${seenSec('MIN')} - ${FIRST_SEEN_FLUSH_LAG_SLACK_SEC}
AND ${hasNoUndated}
AND frequency = (
SELECT SUM(o.occurrence_count) FROM ${occurrenceTable} o WHERE o.${col} = ${table}.id
);
UPDATE ${table}
SET last_seen = ${seenSec('MAX')}
WHERE last_seen > ${seenSec('MAX')} AND ${hasNoUndated};