tryvinci · cursor · Aug 26, 2026 · Aug 26, 2026 · Aug 26, 2026 · Aug 26, 2026
@@ -196,6 +196,7 @@ Ladder: [research/public-bench-ladder.md](../research/public-bench-ladder.md).
 | P58 historical typed-set lists S0 this-VM (2026-08-26) | Same store, hybrid on. S2 product: scan past current-state for item/skill/place-set/`activities`+`done` questions; singular `location` stays a point fact; leftover how/why does not widen hop scan. **141/180** (MH **21/33**, OD **4/11**, SH **85/98**, temporal **31/38**). [pin](./artifacts/locomo-s0-diag-mh-135-p58-20260826.md). John outdoor hiking+mountaineering. Unique losses none vs P54. Collars/tricks still miss. **Not** leftover covering. **Not** n=1540, **not** a Mem0 same-pin. Not 90% (162/180 on this sample; n=1540 for public LoCoMo). |
 | P59 dest-class lists S0 this-VM (2026-08-26) | Same store, hybrid on. S2 product: dest-being skills without requiring `trick`; transfer-cue item objects (`buy`/`for`); possessed-class dest names (`dog named`, not generic `named`). **143/180** (MH **23/33**, OD **4/11**, SH **85/98**, temporal **31/38**). [pin](./artifacts/locomo-s0-diag-mh-135-p59-20260826.md). Audrey collars/tags/toys/beds; James swim/frisbee/skateboard. Unique losses none vs P58. p59/p59b 142/180 are not pins. **Not** leftover covering. **Not** n=1540, **not** a Mem0 same-pin. Not 90% (162/180 on this sample; n=1540 for public LoCoMo). |
 | P61 beneficiary org sets S0 this-VM (2026-08-26) | Same store, hybrid on. S2 product: who/which beneficiary questions recover raise/for-cue affiliation objects (shelter, homeless, hospital) instead of leftover tournament slogans. **144/180** (MH **24/33**, OD **4/11**, SH **85/98**, temporal **31/38**). [pin](./artifacts/locomo-s0-diag-mh-135-p61-20260826.md). Unique losses none vs P59. p60/p60b 143/180 food-set 180s are not pins. **Not** leftover covering. **Not** n=1540, **not** a Mem0 same-pin. Not 90% (162/180 on this sample; n=1540 for public LoCoMo). |
+| P62 community participation sets S0 this-VM (2026-08-26) | Same store, hybrid on. S2 product: in-what-ways community questions recover joined/organizing/host/article-participating-in objects (activist group, art show, mentorship, pride event) instead of leftover courage slogans; attended compiler junk is ranked after. **145/180** (MH **25/33**, OD **4/11**, SH **85/98**, temporal **31/38**). [pin](./artifacts/locomo-s0-diag-mh-135-p62-20260826.md). Unique losses none vs P61. **Not** leftover covering. **Not** n=1540, **not** a Mem0 same-pin. Not 90% (162/180 on this sample; n=1540 for public LoCoMo). |
 | P42 how-did-start leftover covering S0 this-VM (2026-08-25) | Same store, hybrid on. How-did-start leftover covering admits duration-matched inception leftover on `how did … start … years ago`; covering prefers a multi-stem inception pair over a walking-only duration fact or a gym transformation/journey restatement; lexical search drops start/journey wrappers only on that query shape. **126/180** (MH **18/33**, OD **4/11**, SH **74/98**, temporal **30/38**). [pin](./artifacts/locomo-s0-diag-mh-135-p42-20260825.md). Diet+walking pair leftover recovered. Does not add a diet/walking or gym dictionary. **Not** n=1540, **not** a Mem0 same-pin. |
 | BEAM 100K this cycle | [beam-100k-fresh-20260815.md](./artifacts/beam-100k-fresh-20260815.md) |
 | BEAM 100K historical | [beam-100k-c0-async.md](./artifacts/beam-100k-c0-async.md) |

@@ -0,0 +1,55 @@
+# LoCoMo S0 product `/recall` — P62 community participation sets — 2026-08-26
+
+Same frozen store as [locomo-s0-diag-mh-135-20260822.md](./locomo-s0-diag-mh-135-20260822.md): tenant `diag-mh-135` + conv-30, dataset SHA `79fa87e90f04081343b8c8debecb80a9a6842b76a7aa537dc9fdf651ea698ff4`, stratified 180 seed 1, `--fail-closed --skip-ingest`. Product SHA `c02d70a` (joined/organizing/host/article-participating-in recover onto the activity hop; attended compiler junk is ranked after). Hybrid **on** (`BRAINY_RECALL_LLM=1`). Go default for `BRAINY_RECALL_LLM` stays **off**.
+
+P61 pair: [locomo-s0-diag-mh-135-p61-20260826.md](./locomo-s0-diag-mh-135-p61-20260826.md) (`ee2baa6`, **144/180**). Honesty stop: [benchmax-audit-2026-08-25.md](../../research/competitive/benchmax-audit-2026-08-25.md). This is **S2 enumerate**, not leftover covering.
+
+**Not** n=1540. **Not** a Mem0 same-pin. **Not** SOTA. Does not replace integrity 32/180. Does not replace the reader-off 19/180 no-LLM pin. Does not replace README 11.4% / 70% 1×30. 90% on this 180 is **162/180**. 90% on public LoCoMo is n=1540.
+
+## Scores vs prior pins on this store
+
+| Lane | Overall | multi-hop | open-domain | single-hop | temporal |
+| --- | ---: | ---: | ---: | ---: | ---: |
+| product reader **off** (`453a929`) | **19/180 (0.106)** | **12/33** | 0/11 | 5/98 | 2/38 |
+| product hybrid **on** P28 (`454fbb3`) | **113/180 (0.628)** | **18/33** | **4/11** | **61/98** | **30/38** |
+| product hybrid **on** P53 (`ae15e40`) | **137/180 (0.761)** | **18/33** | **4/11** | **85/98** | **30/38** |
+| product hybrid **on** P54 (`7653135`) | **140/180 (0.778)** | **20/33** | **4/11** | **85/98** | **31/38** |
+| product hybrid **on** P58 (`4817e11`) | **141/180 (0.783)** | **21/33** | **4/11** | **85/98** | **31/38** |
+| product hybrid **on** P59 (`2111b3b`) | **143/180 (0.794)** | **23/33** | **4/11** | **85/98** | **31/38** |
+| product hybrid **on** P61 (`ee2baa6`) | **144/180 (0.800)** | **24/33** | **4/11** | **85/98** | **31/38** |
+| product hybrid **on** P62 (`c02d70a`) | **145/180 (0.806)** | **25/33** | **4/11** | **85/98** | **31/38** |
+| industry search+harness (reader-off pin) | 62/180 (0.344) | 10/33 | 3/11 | 27/98 | 22/38 |
+
+MH **24→25**. OD **held 4**. SH **held 85**. Temporal **held 31**. Product overall still leads this-VM industry 62/180 on the labeled product lane — still not a Mem0 same-pin.
+
+Item flips vs P61: **+1 / −0 = net +1**. Unique losses: **none**.
+
+Named recovery (generic S2, not a LoCoMo-named rule):
+
+- `conv-26-q39` Caroline community participation: leftover transition-courage line → **mentorship program, LGBTQ art show, LGBTQ activist group, LGBT pride event** (joined/organizing/host/article-participating-in objects, then attended; gold activist / pride parades / art show / mentoring accepted as paraphrase).
+
+Held: P61 beneficiary join (dog shelter, homeless, children's hospital); P59 dest-class (Audrey collars/tags/toys/beds; James swim/frisbee/skateboard); leftover covering (soda and candy); Jolene snakes Susie+Seraphim; Maria dogs Coco+Shadow; food-set joins (q18 suggestion list; q37 meals without pie).
+
+Still miss (next generic product, not covering): who-told marriage (`conv-49-q78`); food-set completeness vs WRITE-missing gold; OD 4/11; WRITE 4; polar surfing; Phuket where; companionship leftover (do not cover).
+
+## Failure ledger (35 misses)
+
+| Primary | P61 | P62 |
+| --- | ---: | ---: |
+| RETRIEVAL_MISS | 10 | 9 |
+| PROOF_MISS | 14 | 14 |
+| READER_MISS | 7 | 7 |
+| WRITE_MISS | 4 | 4 |
+| HARNESS_ERROR | 1 | 1 |
+
+Unique leftover losses remain **none**. RETRIEVAL 10→9 is q39 flipping off the leftover courage line.
+
+Largest P62 cells: `single-hop:PROOF_MISS` 8, `open-domain:PROOF_MISS` 5, `multi-hop:READER_MISS` 4, `temporal:RETRIEVAL_MISS` 3, `multi-hop:RETRIEVAL_MISS` 3. WRITE 4 — do not merge #133.
+
+## What this says
+
+1. "In what ways … community" questions were an S2 enumerate bug: leftover covering answered with a recency-top courage slogan while joined/attended/organizing/mentorship facts were already in the frozen store. Recovering those objects onto an activity hop, ranking joined/organizing/host ahead of attended compiler junk (`attended back`, `attended not-so-great`), and keeping the join against leftover covering recovers that set.
+2. 145/180 is 80.6% on this sample, still far from 90% (**162/180**). Remaining mass is who-told lists, incomplete food gold (WRITE), OD hypotheticals, relative dates, and polar/where misses. Do not add leftover covering for companionship or peaceful moments.
+3. Do not add an LGBTQ/activist/parade gold dictionary. Cues are joined / organizing / host / article participating-in / mentorship program / attended.
+
+Report: `locomo-s0-diag-mh-135-p62-product-recall-s1-1a5a7c` (summary JSON + failure ledger in this folder). Auto smoke JSON/md dumps are not committed (secret scanner).
@@ -0,0 +1,56 @@
+{
+  "tenant_prefix": "diag-mh-135",
+  "runs": [
+    {
+      "lane": "product-recall",
+      "run_id": "locomo-s0-diag-mh-135-p62-product-recall-s1-1a5a7c",
+      "accuracy": 0.8055555555555556,
+      "correct": 145,
+      "total": 180,
+      "by_group": {
+        "temporal": {
+          "correct": 31,
+          "total": 38
+        },
+        "multi-hop": {
+          "correct": 25,
+          "total": 33
+        },
+        "open-domain": {
+          "correct": 4,
+          "total": 11
+        },
+        "single-hop": {
+          "correct": 85,
+          "total": 98
+        }
+      },
+      "failure_histogram": {
+        "total": 35,
+        "by_primary": {
+          "PROOF_MISS": 14,
+          "RETRIEVAL_MISS": 9,
+          "READER_MISS": 7,
+          "WRITE_MISS": 4,
+          "HARNESS_ERROR": 1
+        },
+        "by_group": {
+          "single-hop:PROOF_MISS": 8,
+          "open-domain:PROOF_MISS": 5,
+          "multi-hop:READER_MISS": 4,
+          "temporal:RETRIEVAL_MISS": 3,
+          "multi-hop:RETRIEVAL_MISS": 3,
+          "open-domain:RETRIEVAL_MISS": 2,
+          "temporal:WRITE_MISS": 2,
+          "single-hop:READER_MISS": 2,
+          "temporal:READER_MISS": 1,
+          "multi-hop:WRITE_MISS": 1,
+          "single-hop:RETRIEVAL_MISS": 1,
+          "single-hop:WRITE_MISS": 1,
+          "temporal:PROOF_MISS": 1,
+          "single-hop:HARNESS_ERROR": 1
+        }
+      }
+    }
+  ]
+}
@@ -5784,3 +5784,77 @@ Last Brainy pin **4/20**. Not re-run.
 
 **One step:** generic **S2 list completeness remainder** — community participation lists (`conv-26-q39`) and who-told lists (`conv-49-q78`) whose gold objects are in the frozen store. Do not fish food-set completeness against WRITE-missing sandwich snacks / Beef Merlot. Then **S2b OD** (still 4/11; 0/4 diagnostic). Then **S1 WRITE** with re-ingest. Then **S5 industry** (62/180 on this tenant). Do **not** queue leftover covering from this 180. Do not chase camping-peaceful. Do not invent Sunday. Do not steal across speakers. Do not special-case Scout. Do not add companionship covering. Isolated leftover covering is saturating. Remaining 36: RETRIEVAL 10, PROOF 14, READER 7, WRITE 4, HARNESS 1. Fair Mem0 180 waits on quota reset 2026-09-01. n=1540 only at S6. Do not merge #133. Do not write SOTA. Kill list unchanged. Start: [handover-sota-agent-2026-08-21.md](../handover-sota-agent-2026-08-21.md).
 
+## 2026-08-26 — P62 community participation sets (145/180; not leftover covering)
+
+Pin: [locomo-s0-diag-mh-135-p62-20260826.md](../../benchmarks/artifacts/locomo-s0-diag-mh-135-p62-20260826.md).
+Run `locomo-s0-diag-mh-135-p62-product-recall-s1-1a5a7c`. Product SHA **`c02d70a`**.
+Honesty stop **`0c33eb8`** stays on `dev`; this increment is **S2 enumerate**, not covering.
+
+**90% honesty:** 90% on this 180 is **162/180**. 90% on public LoCoMo is **n=1540**, last pin **11.4%**.
+This row is **145/180**. It is **not** 90%. It is **not** beating Mem0.
+
+### Landed
+
+Product: in-what-ways / ways + community questions enumerate activity objects recovered from listed memories with joined / organizing / host / article participating-in / mentorship-program cues, then attended. Compiler `attended back` / `attended not-so-great` rows are thin-stopped and ranked after primary slots so they cannot fill the hop cap before art-show/activist/mentorship objects. Dual-entity join stays off. Leftover covering keeps a ≥2 activity join. Named-community token filter still runs. **No** LoCoMo-named rules. **No** leftover-covering detector. **No** LGBTQ/activist/parade gold dictionary.
+
+Pin docs: this section; [docs/benchmarks/README.md](../../benchmarks/README.md) P62 row.
+
+**Do not merge** leftover-covering PRs **#133**, **#131**, **#143**, **#145**.
+
+### Own pins
+
+Same 180, seed 1, 10 convos, fail-closed skip-ingest, tenant `diag-mh-135`. Hybrid on. Dataset SHA `79fa87e90f04081343b8c8debecb80a9a6842b76a7aa537dc9fdf651ea698ff4`.
+
+| Pin | SHA | Overall | MH | OD | SH | temporal |
+| --- | ---: | ---: | ---: | ---: | ---: | --- |
+| Product reader off | `453a929` | **19/180** | **12/33** | 0/11 | 5/98 | 2/38 |
+| P28 | `454fbb3` | **113/180** | 18/33 | 4/11 | 61/98 | 30/38 |
+| P53 (covering) | `ae15e40` | **137/180** | 18/33 | 4/11 | 85/98 | 30/38 |
+| P54 (entity-scoped counts) | `7653135` | **140/180 (0.778)** | **20/33** | **4/11** | **85/98** | **31/38** |
+| P58 (historical typed-set lists) | `4817e11` | **141/180 (0.783)** | **21/33** | **4/11** | **85/98** | **31/38** |
+| P59 (dest-class lists) | `2111b3b` | **143/180 (0.794)** | **23/33** | **4/11** | **85/98** | **31/38** |
+| P61 (beneficiary org sets) | `ee2baa6` | **144/180 (0.800)** | **24/33** | **4/11** | **85/98** | **31/38** |
+| **P62 (community participation sets)** | **`c02d70a`** | **145/180 (0.806)** | **25/33** | **4/11** | **85/98** | **31/38** |
+| Industry search+harness | same tenant | **62/180** | 10/33 | 3/11 | 27/98 | 22/38 |
+| Full n=1540 product `/recall` | `1b5ab3e` | **175/1540 = 11.4%** | 7.4% MH | 5.2% OD | 10.5% SH | 19.0% temporal |
+| 1×30 conv-26 | `1b5ab3e` | **21/30 (70%)** | 10/10 | **0/4** | — | 11/16 |
+| LME-20 | `1b5ab3e` | **4/20** | | | | |
+
+Unique losses vs P61: **none**. Gain: Caroline community participation (`conv-26-q39`). Industry **62/180**. n=1540 and 1×30 **not** re-run; do not replace README 11.4% / 70%.
+
+### Competitor compare (detailed)
+
+#### 1. LoCoMo — trail (open-domain); this 180 is not public LoCoMo
+
+Public LoCoMo is **n=1540 / 11.4%**. This 180 is a stratified diagnostic. **OD 4/11** (0/4 diagnostic still). Do **not** write lead from 145/180 vs Mem0 11/30 (handicapped, different pin). Published Mem0 **92.5%** (their harness, top-k 200, n=1540) is **context**, never a scoreboard row. Fair Mem0 180 waits on quota **2026-09-01**.
+
+**Trailing axis (open-domain):** product mechanism still missing is **S2b** membership/hypothesis + **S1** write coverage (Xenoblade, yoga 2020, Phuket, Wolves). PoR: compiler coverage then re-ingest; not leftover covering.
+
+**Leading axes (must not regress):** OpMem 13/13, marketing 17/17, 1×30 MH 10/10. P61 beneficiary join, P59 dest-class lists, P54 counts, and P53 covering holds held on live `/recall`.
+
+#### 2. OpMem — lead (stale pin)
+
+Last **13/13**. Last Mem0 Platform ops pin **10/13**. **Lead ops.** Not re-run this increment.
+
+#### 3. Marketing vertical — lead (stale pin)
+
+Last **17/17** vs Mem0 empirical **4/17**. **Lead governed vertical.**
+
+#### 4. LME-20 — no pin this cycle
+
+Last Brainy pin **4/20**. Not re-run.
+
+#### 5. Graphiti / Zep — no pin
+
+**No same-pin.**
+
+**Mem0 OSS** was not re-measured. Platform fair 180 is **quota-blocked** until 2026-09-01.
+
+### Why
+
+`"In what ways is Caroline participating in the LGBTQ community?"` recovered a recency-top courage slogan because leftover covering beat an empty/wrong typed list. The four gold classes were already in listed memories as joined activist group, attended pride, organizing/hosting an art show, and a mentorship program. Recovering those objects onto an activity hop, ranking joined/organizing/host ahead of attended compiler junk, and keeping the join against leftover covering is the product mechanism.
+
+### Next
+
+**One step:** generic **S2 list completeness remainder** — who-told lists (`conv-49-q78`) whose gold objects are partly in the frozen store (work friends + extended family; do not invent Sam-as-told from in-dialogue). Do not fish food-set completeness against WRITE-missing sandwich snacks / Beef Merlot. Then **S2b OD** (still 4/11; 0/4 diagnostic). Then **S1 WRITE** with re-ingest. Then **S5 industry** (62/180 on this tenant). Do **not** queue leftover covering from this 180. Do not chase camping-peaceful. Do not invent Sunday. Do not steal across speakers. Do not special-case Scout. Do not add companionship covering. Isolated leftover covering is saturating. Remaining 35: RETRIEVAL 9, PROOF 14, READER 7, WRITE 4, HARNESS 1. Fair Mem0 180 waits on quota reset 2026-09-01. n=1540 only at S6. Do not merge #133. Do not write SOTA. Kill list unchanged. Start: [handover-sota-agent-2026-08-21.md](../handover-sota-agent-2026-08-21.md).
+