Repository navigation
Close phase 7 in the roadmap and open phase 8 - #23
Conversation
Phase 7 closed with PR #21. The roadmap records what it changed for later phases, and lists more extractor prompts under Deferred. The plan drops phase 7, adds the detail results to its start point, and gives phase 8 the detail baseline, its target groups, and what the fact sources mean for the episode key.
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
| @@ -0,0 +1,348 @@ | |||
| # mem0 survey: how it extracts, stores, and recalls memories | |||
There was a problem hiding this comment.
nit: Name docs/internal/mem0-survey.md in the PR description, or move it to its own PR (not blocking).
The description says the PR closes phase 7 and opens phase 8, and lists only roadmap.md and plan.md. A reviewer who reads only the description misses 348 added lines. AGENTS.md also lists every internal doc under "Where things are", and the survey has no entry there.
|
|
||
| The accuracy comes mostly from the extraction prompt, not from the storage or the retrieval | ||
| machinery. This matches the hotmemory diagnosis: 19 of the 35 memory misses on sample A were facts | ||
| that the extractor never wrote (`hotmemory/docs/benchmarks.md:180-191`). |
There was a problem hiding this comment.
nit: Change the citation to docs/benchmarks.md:221-228 (not blocking).
Lines 180-191 of docs/benchmarks.md now hold the replay notes and the start of sample B. The "19 of 35" counts are at lines 221 and 226. The hotmemory/ prefix also does not resolve inside this repository. The same prefix appears at line 10.
| 1. Extraction prompt (19 of 35 misses). Our prompt asks for "facts that a personal assistant | ||
| must remember about the user", as "one short sentence". mem0 asks for both speakers, says | ||
| "when in doubt, extract", sets a minimum count, and wants 15 to 80 words of context per | ||
| memory. Phase 7 is the place to test a mem0-style variant. Candidate changes: |
There was a problem hiding this comment.
nit: Mark this section as written before phase 7 ran, or update it to the phase 7 results (not blocking).
This PR closes phase 7. The survey still says "Phase 7 is the place to test a mem0-style variant". Line 4 also places hotmemory at branch phase-6-accuracy-loop. roadmap.md now records that detail ran and that more prompts are deferred. A reader of the survey cannot tell which candidate changes detail already covers.
1 participant
This PR closes phase 7 in the roadmap and opens phase 8 as #22. It changes docs only.
How:
docs/internal/roadmap.md: phase 7 is closed on 2026-10-14 with PR Phase 7: the extractor #21, and phase 8 is open as Phase 8: episodes behind facts #22. A closing paragraph for phase 7 records its results and how it changes later phases. "More extractor prompts" goes under Deferred, becausedetailmet a stop condition.docs/internal/plan.md: the phase 7 section is removed. Its plan stays in git history atab93da1. The start table gains thedetailresults: 31 of 54 on sample A and 32 on sample B. The extractor fact and the fixed-models decision now namedetailand the fact source.detailconfiguration as its baseline, and the groups of misses it targets.Why: the roadmap rule says that the next phase opens only after the previous phase closes. Issue #22 holds the same phase 8 brief as the plan.
Verification:
make verifypasses. A search of the added lines found no private names.