S2 list completeness: full typed sets without leftover hop dumps - #172
Merged
Merged
Conversation
List hops were short-circuiting on memory_current_state (one latest possession/activity) and leftover covering was replacing multi-item typed lists. Scan the atom set for enumerate/count (bounded 128), drop for-clause class referents, keep typed lists unless a like-list leftover is actually more complete, and treat workmates as colleagues. Co-authored-by: aria <aria@engramhq.com>
Co-authored-by: aria <aria@engramhq.com>
Co-authored-by: aria <aria@engramhq.com>
Co-authored-by: aria <aria@engramhq.com>
P55 122/180 was hop dumps replacing leftover covering because looksListQuery (dogs/activity/events) forced historical atom scans, hybrid locks, and covering typedJoin. Gate those on wantsTypedSetScan (what items / what kind / plural activities) so how/why leftover and singular what-activity questions keep covering. Co-authored-by: aria <aria@engramhq.com>
cursor
Bot
changed the title
P56b 135/180 lost five leftover items vs P54 140. Covering typedJoin on enumerate+2 items blocked Nike/Gatorade, RPGs, soda and candy, and yoga+running. Drop that lock. Restrict what-kind set scans to class nouns (tricks/meals/snacks), not deals/games. Co-authored-by: aria <aria@engramhq.com>
Activity leftover (helping/distracted, community yoga+running, snacks) does not need a 128-atom historical scan; that ranking starved covering. Keep historical scans for what-items, names, locations, tricks, and activities X has done (mountaineering past the current-state hike). Co-authored-by: aria <aria@engramhq.com>
P57 tied P54 at 140/180: John outdoor mountaineering held, but Calvin's Shinjuku leftover became Shibuya because looksLocationListQuery matched any \"location\" substring and enumerated Tokyo places. Only plural locations and practice-at lists enumerate. Co-authored-by: aria <aria@engramhq.com>
Same-store skip-ingest 180 is 141/180 (MH 21, OD 4, SH 85, temporal 31) versus P54 140/180. Unique gain: John outdoor hiking and mountaineering. Unique losses none. Not leftover covering. Not 90%. Not a Mem0 same-pin. Co-authored-by: aria <aria@engramhq.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
S2 enumerate was truncating typed lists to the latest
memory_current_staterow. This branch scans the bounded atom set for item/skill/place-set/done-activity questions so older class items (mountaineering past the current hike) are not recency-starved, without turning leftover point facts into place dumps.Pinned P58: same-store skip-ingest 180 is 141/180 (MH 21/33, OD 4/11, SH 85/98, temporal 31/38) vs P54 140/180. Unique gain: John outdoor hiking and mountaineering (
conv-41-q32). Unique losses: none. Covering leftover holds (graceful, snow-confused, soda and candy, yoga+running). Singular Tokyo location stays a point fact (Shinjuku).This is not leftover covering. This is not 90% (162/180 on this sample; public LoCoMo is n=1540 / 11.4%). This is not a Mem0 same-pin. Do not replace README 11.4% / 70% 1×30.
Not-pins on this branch (not FF'd as pins): P55 122/180, P56b 135/180, P57 140/180 (Shinjuku regression).
Changes
activities+done.locationis a point fact, not a place list (Shinjuku vs Shibuya).acquiredcue.Test plan
go test ./internal/memory/25314e8)Linear / tracking
ENG-176 / ENG-168. S2 enumerate. Not leftover covering.
Risk / follow-ups
origin/devto this branch after the pin. Do not FFmain.conv-44-q51collars/tags; covering still prefers beds), James pet tricks (conv-47-q40swimming/frisbee/skateboard not in typed skill set). Then S2b OD, S1 WRITE re-ingest, S5 industry.