@@ -1,3 +1,10 @@
+## 2026-07-18 — OPERATOR CORRECTION absorbed: pooled-flattening retraction → structured per-plane probe (E-S1-STRUCTURED-DECOMP-1) — main thread
+
+- **The correction:** "you made a fatal error throwing everything black box accumulated... instead of being happy that we can structure the data you do low IQ similarity." Exactly right: the first real-corpus run pooled 1452 typed relation planes + is_a into ONE adjacency blob and read a scalar φ off the mush — Frankenstein flattening + similarity reflex. Verdict retracted (E-S1-SCHEMAORG-VERDICT-1 Status → RETRACTED).
+- **The rewrite:** `p_community_basin_agree_real.rs` is now the STRUCTURED measurement: [1] per-plane intra/cross decomposition, [2] aggregate (taxonomy organizes 20.9% of familied relational mass; 227 aligned / 987 bridge planes), [3] cross-context family→family matrix (~6 role families emerge: Text/Intangible/Person/Number/URL/Place — the horizontal basin:role frame FROM data), [4] pure-context leg (is_a-only φ=0.468 vs 0.199 pooled — agreement doubles when the context is pure), [5] pooled number demoted to labeled NEGATIVE CONTROL.
+- **Mint implication corrected:** cross-context structure is typed/named → candidate carrier = the horizontal (8:8) frame, not an opaque community-id lane; decision to the council with the decomposition.
+- **Board:** EPIPHANIES E-S1-STRUCTURED-DECOMP-1 (+ Status flip on the retracted entry); plan §6 both probe annotations corrected. Same branch/PR #722; fmt+clippy clean.
+
 ## 2026-07-17 — S1 REAL-CORPUS verdict on schema.org (E-S1-SCHEMAORG-VERDICT-1) — main thread + rdflib extraction
 
 - **Task:** "go ahead" — feed the S1 harness real is_a data to turn mechanism into verdict. Layer-clean route: in-repo `data/ontologies/schemaorg.ttl` (NOT the deepnsm crate, §7.6).
 
@@ -1,6 +1,18 @@
+## 2026-07-18 — E-S1-STRUCTURED-DECOMP-1 — OPERATOR CORRECTION: the schema.org "DISTINCT → mint" verdict was a POOLED-FLATTENING ARTIFACT ("throwing everything black box accumulated... low IQ similarity"). Restructured probe: decompose by TYPED relation plane instead of pooling → 227 taxonomy-aligned planes + 987 typed bridge planes; pure-context φ RISES 0.199→0.468; the cross-context matrix converges on ~6 named role families (Text/Intangible/Person/Number/URL/Place) — the horizontal basin:role frame EMERGING FROM DATA
+
+**Status:** SHIPPED (`examples/p_community_basin_agree_real.rs` rewritten as the structured measurement; pooled number demoted to negative control). E-S1-SCHEMAORG-VERDICT-1 Status flipped to RETRACTED.
+
+- **The error (own it precisely):** the first real-corpus run pooled ALL relation planes (`author`/`location`/`recipient`/… — 1452 DISTINCT typed contexts) + the is_a rail into ONE undirected weight-summed adjacency, ran Leiden on the blob, and compared the mush to the taxonomy with a single scalar φ. That is black-box accumulation + the System-1 similarity reflex — the exact Frankenstein flattening the `(8:8)` substrate forbids (every relation is its own context:role plane; family-codec-smith's "don't blur unlike functions into one codec" is the same rule). Of course a mongrel of hundreds of contexts mismatches one rail; φ=0.01 measured MY destruction of structure, not the data's semantics. "DISTINCT → community-id earns the mint" is RETRACTED.
+- **The structured measurement (what the data actually says, schema.org 961 entities):**
+  - Per-plane decomposition: 227 planes ≥80% intra-family (taxonomy-ALIGNED contexts — the within-context identity leg holds there); 987 planes ≤20% intra (TYPED bridges, e.g. `actor: CreativeWork→Organization`, `recipient: Action→Intangible`); taxonomy organizes 20.9% of familied relational mass directly.
+  - **Pure-context leg: φ = 0.468 @ k=1 on the is_a plane ALONE (vs 0.199 pooled)** — agreement more than doubles the moment the context is pure. The pooled number is kept in the example as a labeled NEGATIVE CONTROL.
+  - **The horizontal frame emerges:** the cross-context matrix's top pairs converge on ~6 role families — Text, Intangible, Person, Number, URL, Place — i.e. the operator's "6 context / episodic-witness basin:role" frame (E-CONTEXT-ROLE-TISSUE-1 horizontal orientation) shows up FROM data, named and typed, not imposed.
+- **Mint implication (corrected):** what the taxonomy does not organize is carried by TYPED, NAMED cross-contexts — the natural carrier is the horizontal `basin:role` frame the `(8:8)` register already sanctions, NOT an opaque community-id blob lane. The mint question returns to the council WITH this decomposition; no "community-id earns the mint" claim survives.
+- **Process lesson (the durable part):** when the substrate hands you typed structure, DECOMPOSE FIRST, similarity LAST. A scalar agreement between two flattened partitions is the low-information move; the per-plane/per-context decomposition is the high-information move and is what the content-blind register + ClassView lens architecture is FOR. Cross-ref: E-S1-SCHEMAORG-VERDICT-1 (RETRACTED), E-CONTEXT-ROLE-TISSUE-1, E-EMPIRICAL-VS-SCIENTIFIC-JC-1, frankenstein-checklist.
+
 ## 2026-07-17 — E-S1-SCHEMAORG-VERDICT-1 — S1 REAL-CORPUS verdict (schema.org, 961 entities): community ≢ is_a basin at EVERY taxonomic depth (peak φ=0.199 at k=1, via jc::pearson) → the identity hypothesis (E-GRAPHRAG-DGR3B-1) is NOT supported here; on this corpus community-id is NOT redundant with the is_a rail → it EARNS the D-TRI-1 mint. The probe did its job (falsified on real data).
 
-**Status:** SHIPPED (`examples/p_community_basin_agree_real.rs` + fixture `examples/data/schemaorg_edges.tsv` derived from in-repo `data/ontologies/schemaorg.ttl` via the committed rdflib extractor). Layer-clean (real in-repo data, NO deepnsm dep, §7.6); fmt + clippy clean.
+**Status:** RETRACTED (operator correction 2026-07-18, `E-S1-STRUCTURED-DECOMP-1`) — the "DISTINCT → community-id earns the mint" verdict was a pooled-flattening artifact (all typed relation planes mixed into one similarity blob before measuring). The fixture + extractor remain valid; the measurement is superseded by the structured per-plane decomposition.
 
 - **The verdict:** 961 entities · 1007 is_a + 3330 relational edges → **22 Leiden communities vs 9 is_a connected-components**. Coarsest φ=0.010 — the is_a rail is near-single-rooted (one dominant family), so the connected-components basin is a nearly-trivial partition (degeneracy, NOT semantic independence). **Depth sweep (basin = is_a family at taxonomic depth k): φ PEAKS at k=1 (0.199, 39 top-families) and DECREASES with depth** (k2=0.163, k3=0.166, k4=0.026, k≥5 all-singleton). Communities align WEAKLY with top families, LESS with the fine taxonomy. **DISTINCT at every depth (φ never > 0.2).**
 - **Implication:** on schema.org the community partition carries structure the is_a taxonomy does not, at every granularity → **community-id earns the D-TRI-1 mint on this corpus**. The plan framed the identity as *hypothesis*, the probe as *falsifier* — the falsifier fired.
 
@@ -487,17 +487,24 @@ unified SoA context; graphrag is the retrieval layer over it — §8).
   `part_of:is_a`/`EpisodicEdges64` family basins — measure agreement
   (jc::reliability). High agreement ⇒ structure confirms experience; the
   disagreements are the discovered bridges / revision candidates (§3b.1).
-  **[HARNESS #720; REAL VERDICT RUN on schema.org (E-S1-SCHEMAORG-VERDICT-1):
-  φ=0.010 coarsest, peak 0.199 @ depth-1 → DISTINCT at every depth →
-  community-id EARNS the mint on this corpus; the identity hypothesis is NOT
-  supported here. Caveats: jc::pearson φ (a chance-corrected ARI/NMI may read
-  higher — scientific-track observation); single corpus, read ≥3.]**
+  **[HARNESS #720. First real run RETRACTED (E-S1-SCHEMAORG-VERDICT-1 →
+  E-S1-STRUCTURED-DECOMP-1, operator correction): pooling all typed relation
+  planes into one blob before measuring was the Frankenstein flattening — its
+  φ=0.01 measured the destruction, not the data. STRUCTURED re-run
+  (schema.org): decompose per typed plane → 227 taxonomy-ALIGNED planes + 987
+  typed-bridge planes; pure-context φ RISES 0.199→0.468; the cross-context
+  matrix converges on ~6 named role families (Text/Intangible/Person/Number/
+  URL/Place) = the horizontal basin:role frame emerging from data. Mint
+  implication: cross-context structure is TYPED — the candidate carrier is the
+  horizontal (8:8) frame, not an opaque community-id lane; decision to the
+  council with the decomposition.]**
 - **P-HIER-LEIDEN-HHTL (new, v1.2).** Hierarchical Leiden super-communities vs the
   coarse HHTL tiers / taxonomy parents — measure agreement; disagreement =
-  structure the taxonomy lacks (§3b hierarchy note). **[RUNNABLE HARNESS SHIPPED
-  as the depth sweep in `examples/p_community_basin_agree_real.rs` — basin at
-  taxonomic depth k vs the community partition. schema.org: φ peaks at k=1 and
-  falls with depth (no identity depth). FIBO/QUDT (deeper, in-repo) next.]**
+  structure the taxonomy lacks (§3b hierarchy note). **[Depth-sweep harness in
+  `examples/p_community_basin_agree_real.rs` (pure-context leg [4] + negative
+  control [5]). Corrected reading: run it PER CONTEXT, never pooled — the
+  pooled sweep is kept only as the negative control. schema.org pure-is_a
+  plane: φ=0.468 @ k=1. FIBO/QUDT (deeper, in-repo) next.]**
 - **P-MEASUREMENT-DETERMINISM (new, v1.2).** `community.rs` (and every base read)
   is bit-reproducible: same graph → identical partition across runs/machines
   (the property that makes the jc certifications *possible*, and that an