Reported from claude.ai after #28: searching WHP founding partner pricing returned this excerpt from the OLS memory:
…AI agents are treated as business partners: honest, direct, push back when something won't work. Q1 2027 priorities: (1) WHP commercial launch…
It should have shown the Founding Partner and pricing passage further down.
Root cause
Every query term counted the same:
"WHP" appears four times in the memory, and "partners" turns up coincidentally in an earlier sentence.
That window matched 2 terms, the same as the "Founding Partner Program" window, and won the tie by coming first.
"pricing" matched nothing, because the memory says "price".
Fix
All changes are in lib/memory-excerpt.ts:
Rarity weighting: each term counts 1 ÷ (its occurrences in this memory). "founding" appears once and counts fully; "WHP" appears four times and counts a quarter.
Light suffix stripping: strip one of -ing, -ed, -es, -s or -e (keeping 4+ characters) and match as a prefix, so "pricing" and "price" both become "pric".
Why not ts_headline
I tested it on the same text in Postgres 16:
OR query (whp | founding | partner | pricing): returns the same wrong "business partners … WHP" passage, because it also weighs words equally. Same result with 1 or 2 fragments.
plainto_tsquery (AND): no passage contains all four terms, so it falls back to the "WHP decisions … $20 or $25 per server" sentence, which still misses Founding Partner.
Result
…"cPanel" in feature names, trademark). Founding Partner Program: first 10 hosts get 90 days free on 1 server, one-on-one onboarding (setup, test migrations, WHMCS integration) and a locked price;…
Tests
Regression test using the real memory text: the excerpt must contain "Founding Partner Program" and "locked price".
Inflection test: "pricing" must find "flat price per server".
Both failed before the fix. Full suite: 90 passing. Typecheck and lint are clean.
## Problem
Reported from claude.ai after #28: searching `WHP founding partner pricing` returned this excerpt from the OLS memory:
> …AI agents are treated as business partners: honest, direct, push back when something won't work. Q1 2027 priorities: (1) WHP commercial launch…
It should have shown the Founding Partner and pricing passage further down.
## Root cause
Every query term counted the same:
- "WHP" appears four times in the memory, and "partners" turns up coincidentally in an earlier sentence.
- That window matched 2 terms, the same as the "Founding Partner Program" window, and won the tie by coming first.
- "pricing" matched nothing, because the memory says "price".
## Fix
All changes are in `lib/memory-excerpt.ts`:
- **Rarity weighting:** each term counts 1 ÷ (its occurrences in this memory). "founding" appears once and counts fully; "WHP" appears four times and counts a quarter.
- **Light suffix stripping:** strip one of -ing, -ed, -es, -s or -e (keeping 4+ characters) and match as a prefix, so "pricing" and "price" both become "pric".
## Why not `ts_headline`
I tested it on the same text in Postgres 16:
- **OR query** (`whp | founding | partner | pricing`): returns the same wrong "business partners … WHP" passage, because it also weighs words equally. Same result with 1 or 2 fragments.
- **`plainto_tsquery` (AND):** no passage contains all four terms, so it falls back to the "WHP decisions … $20 or $25 per server" sentence, which still misses Founding Partner.
## Result
> …"cPanel" in feature names, trademark). Founding Partner Program: first 10 hosts get 90 days free on 1 server, one-on-one onboarding (setup, test migrations, WHMCS integration) and a locked price;…
## Tests
- Regression test using the real memory text: the excerpt must contain "Founding Partner Program" and "locked price".
- Inflection test: "pricing" must find "flat price per server".
- Both failed before the fix. Full suite: 90 passing. Typecheck and lint are clean.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Searching "WHP founding partner pricing" excerpted the "business
partners ... (1) WHP commercial launch" sentence instead of the Founding
Partner passage. Every query term counted the same, so a common term
("WHP", four times) plus a coincidental hit ("partners") tied the real
passage on two terms and won by coming first. "pricing" matched nothing:
the memory says "price".
- Weight each term by rarity within the memory (1 / occurrences), so a
word that appears once outweighs one sprinkled throughout.
- Strip one common suffix (-ing/-ed/-es/-s/-e, keeping 4+ chars) and
match as a prefix, so "pricing" and "price" both become "pric".
Postgres ts_headline, suggested as the alternative, was tested on the
same text and picked the same wrong passage with an OR query (it also
weighs words equally); with plainto_tsquery's AND it found no cover and
fell back to a different, also wrong, passage.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jknapp
merged commit 3da0c77849 into main2026-11-14 19:14:27 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Problem
Reported from claude.ai after #28: searching
WHP founding partner pricingreturned this excerpt from the OLS memory:It should have shown the Founding Partner and pricing passage further down.
Root cause
Every query term counted the same:
Fix
All changes are in
lib/memory-excerpt.ts:Why not
ts_headlineI tested it on the same text in Postgres 16:
whp | founding | partner | pricing): returns the same wrong "business partners … WHP" passage, because it also weighs words equally. Same result with 1 or 2 fragments.plainto_tsquery(AND): no passage contains all four terms, so it falls back to the "WHP decisions … $20 or $25 per server" sentence, which still misses Founding Partner.Result
Tests
🤖 Generated with Claude Code
Searching "WHP founding partner pricing" excerpted the "business partners ... (1) WHP commercial launch" sentence instead of the Founding Partner passage. Every query term counted the same, so a common term ("WHP", four times) plus a coincidental hit ("partners") tied the real passage on two terms and won by coming first. "pricing" matched nothing: the memory says "price". - Weight each term by rarity within the memory (1 / occurrences), so a word that appears once outweighs one sprinkled throughout. - Strip one common suffix (-ing/-ed/-es/-s/-e, keeping 4+ chars) and match as a prefix, so "pricing" and "price" both become "pric". Postgres ts_headline, suggested as the alternative, was tested on the same text and picked the same wrong passage with an OR query (it also weighs words equally); with plainto_tsquery's AND it found no cover and fell back to a different, also wrong, passage. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>