contract(unicharset): direction + mirror leaf — byte-parity 112/112, varied-field surface complete

.claude/board/EPIPHANIES.md

@@ -1805,6 +1805,19 @@ Why this is the right move, not just a bug patch:
 3. **Flexibility + the one cost.** A node mixes in up to 16 family adjacencies (huge flexibility, any-to-any within 256). The named limitation is **mixin dependency**: a referenced family must exist or the slot is a dangling adapter (skipped). That is the honest trade — and it is cheap, because a missing family is a render no-op, not a corruption.
 
 The general rule for graph edges on this substrate: **resolve to the stable grouping (family), not the volatile leaf (member)** — unless a richer flavor (8×16-bit, 32×4 residue, member→member second-hop) is measured to be needed. Cross-ref: `E-ANCHOR-IS-A-HEAD-FIELD-NOT-A-VALUE-TYPE` (the static dual), `E-GUID-IS-THE-GRAPH`, the operator's deferred helix-basin-anchor (CLAM ⇄ Louvain turbovec edge residue) as the eventual richer flavor; `aiwar.rs` (the POC: 221 aiwar entities → 60 category family hubs).
+## 2026-06-20 — E-CPP-PARITY-6 — the UNICHARSET `direction` + `mirror` columns are byte-identical to libtesseract; the sixth leaf, and the first to read PAST the bbox CSV into the multi-column tail
+
+**Status:** FINDING (in-env, real trained data). `lance_graph_contract::unicharset::UniCharSet::{get_direction, get_mirror}` dump the `eng.lstm-unicharset` per-id bidi direction codes and mirror ids **byte-identical to tesseract's own `get_direction` / `get_mirror`, 112/112 each** (same self-validating oracle, `direction` + `mirror` modes). Sixth + seventh proven accessor surfaces.
+
+**Why this was the "multi-tier parser" leaf — and why it turned out simple.** `direction`/`mirror` sit two/three columns past the script, after the bbox+stats CSV. Tesseract places them via a 5-tier `istringstream` fallback (`unicharset.cpp:833-868`). But the bbox+stats group is always a SINGLE whitespace token (comma-separated, no spaces), so on a whitespace split the columns land at fixed offsets regardless of tier: `script`, `other_case`, `direction`, `mirror` are simply the 1st/2nd/3rd/4th tokens after the optional CSV. Continuing the existing per-line token walk one and two positions past `other_case` reads them; a tier without the columns leaves the walk exhausted → defaults. No bespoke tier detector needed — the token walk IS the tier collapse. (The float stats inside the CSV still need decimal parsing; that's the remaining sub-leaf.)
+
+**Two transcode subtleties the oracle pinned (read-the-truth-first, again).** (1) `direction`'s load default is `U_LEFT_TO_RIGHT` (0) for an absent column, but `get_direction`'s OUT-OF-RANGE return is `U_OTHER_NEUTRAL` (10) — two different "defaults" for two different conditions (`unicharset.h:712-714`). (2) `mirror` is clamped at load exactly like `other_case` (`>= size` → self) and returns `INVALID_UNICHAR_ID` (-1) out of range. The oracle confirmed direction is genuinely varied on eng (55× LTR=0, 33× OTHER_NEUTRAL=10, plus 2/3/4/6 for digit-class chars) and mirror has 10 real pairs (bracket/paren/brace mirrors, e.g. `(`↔`)`), so this exercises the parse, not just the defaults.
+
+**Pattern holds (E-CPP-KEYSTONE-1).** +2 accessors + 2 dumps + one `diff` each, no new architecture, no Core gap. +3 contract tests (26 unicharset total). Consumed by `tesseract-core::CharSet::{get_direction,get_mirror}`. Reproducible via the committed `examples/unicharset_dump.rs {direction,mirror}`.
+
+**Tooling note (TECH_DEBT filed):** the contract crate is NOT fmt-gated in CI (`style.yml` checks only `lance-graph` + `deepnsm`), so merged symbiont/SoA PRs left rustfmt-1.9.0 drift in `hhtl.rs`/`nan_projection.rs`/`soa_graph.rs`. My leaf files are fmt-clean; I did not reformat others' merged files. See TECH_DEBT.
+
+Cross-ref: `E-CPP-PARITY-1..5` (the prior leaves), `E-CPP-KEYSTONE-1`, `.claude/knowledge/core-first-transcode-doctrine.md`. Branch `claude/happy-hamilton-0azlw4`, lance-graph + tesseract-rs.
 
 ---
 

.claude/board/LATEST_STATE.md

@@ -36,6 +36,8 @@ Membrane consumers can now pull BOTH halves of a render `classid` BBB-safely fro
 
 ---
 
+> **2026-06-20 — branch work (`claude/happy-hamilton-0azlw4`)** — **UNICHARSET `direction` + `mirror` transcoded + byte-parity proven (E-CPP-PARITY-6), the sixth leaf — first to read PAST the bbox CSV.** `UniCharSet` now parses the two columns after `other_case` into `directions: Vec<i32>` + `mirrors: Vec<i32>` by continuing the per-line token walk (the bbox+stats group is one whitespace token, so columns land at fixed offsets regardless of the 5-tier fallback — no bespoke tier detector). `get_direction` (`unicharset.h:712`, load default `U_LEFT_TO_RIGHT` 0, out-of-range → `U_OTHER_NEUTRAL` 10) + `get_mirror` (`unicharset.h:721`, clamped like other_case, out-of-range → -1) + `dump_direction`/`dump_mirror`. **Byte-identical 112/112 each** on real `eng.lstm-unicharset` (self-validating oracle; direction varied: 55× LTR / 33× OTHER_NEUTRAL / 2·3·4·6 for digit chars; mirror has 10 bracket/paren pairs). Additive, zero-dep; +3 contract tests (26 unicharset total), my files clippy + fmt clean; reproducible via `examples/unicharset_dump.rs {direction,mirror}`. Consumed by `tesseract-core::CharSet::{get_direction,get_mirror}`. No Core gap. Remaining UNICHARSET sub-leaf: the float stats (bbox ints + width/bearing/advance) inside the CSV. EPIPHANIES `E-CPP-PARITY-6`; TECH_DEBT (contract crate not fmt-gated in CI).
+>
 > **2026-06-20 — IN PR (`claude/jirak-math-theorems-harvest-rfii13`)** — **kanban×Rubicon SoA value tenant + per-tenant counters (capstone S1 green).** NEW `ValueTenant::Kanban = 9` at value-slab `[112,120)` (8 B: `phase|exec|reserved|cycle`), added to `ValueSchema::{Cognitive,Full}` — reserve-don't-reclaim, **layout-preserving** (Full 112→120 B, stride 512 untouched, no version bump). `KanbanTenant` Copy view + `NodeRow::{kanban,set_kanban}` (owner-gated write / surreal read-only / Rubicon); `KanbanColumn`/`ExecTarget` `from_u8`. **Subsumes the envelope-pointer G1** — the node carries its own phase+cycle, pinning SoA↔kanban in the LE blob (a `FixedSizeBinary(512)` store reads kanban zero-copy at any version). NEW `tenant_counter` module + feature `tenant-counters` (default OFF, zero-cost no-op; one relaxed atomic/tenant-write when on) — the capstone NaN-census instrument; `set_kanban` is the first wired cascade point. Decisions kept (I-VSA-IDENTITIES + AGI-glove): thinking-style is ClassView+`Meta`, NOT a 128-bit tenant; plan-shape ClassView-derived; MUL flow-trigger is a function, not a tenant. Contract lib **714**/715(tenant-counters)/720(guid-v2-tail), clippy `-D warnings` + fmt clean all three. Refs: AGENT_LOG (cont.¹⁷), EPIPHANIES `E-KANBAN-IS-A-VALUE-TENANT-SUBSUMES-G1`, plan `capstone-cognitive-loop-wiring-nan-census-v1` (S1 green).
 >
 > **2026-06-20 — IN PR (`claude/jirak-math-theorems-harvest-rfii13`)** — **Zero-copy SoA read contract: `node_rows_from_le_bytes` (the surrealdb "second brain" primitive).** The inverse of `NodeRowPacket::as_le_bytes` (WRITE) — `canonical_node::node_rows_from_le_bytes(&[u8]) -> Option<&[NodeRow]>`, a CHECKED zero-copy cast (`len % 512 == 0` AND `ptr % 64 == 0`, else `None` → caller copies, no UB; empty→Some(empty)). This IS the LE contract a backing store satisfies so its bytes ARE the SoA the cognitive shader reads in place. **Brutal verdict:** lance-graph side now zero-copy-ready end-to-end; surrealdb's kv-lance does NOT qualify as scaffolded (`val: DataType::Binary` variable-length → needs `FixedSizeBinary(512)`), and value zero-copy holds only if stored UNcompressed (key/address always zero-copy). 712 contract lib green, clippy `-D warnings` both configs + fmt clean. Refs: AGENT_LOG 2026-06-20 (cont.¹⁴), EPIPHANIES `E-SURREALDB-SECOND-BRAIN-IS-ZERO-COPY-IFF-FIXEDSIZEBINARY`.
@@ -160,6 +162,8 @@ Membrane consumers can now pull BOTH halves of a render `classid` BBB-safely fro
 
 > **2026-06-18 — ADDED (D-DO-ARM-1, the OGAR DO arm)**: `lance_graph_contract::action::{ActionState, StateGuard, ActionDef, ClassActions, actions_for, effective_actions, ActionInvocation}` — the Perdurant DO arm completing the OGAR IR (the action-axis sibling of `codegen_manifest`'s `MethodSig`/THINK). Both the 4-agent `sale_order` AR→DO probe (runtime-archaeologist) AND the merged cross-repo PR survey (ruff/OGAR/lance-graph/openproject/tesseract) agreed this was the ONE missing wire: the THINK arm (`classid → ClassView`, `has_function → MethodSig`) is converged + merged; the DO-arm `ActionInvocation`/`ActionDef` type was ABSENT. **`ActionDef`** (static, `const`-constructible, all `&'static`/`Copy`): `predicate` (= harvested `has_function` method), `object_class` (classid), `exec` (`ExecTarget` incl `SurrealQl`), `guard` (`StateGuard` = KausalSpec field==value), `required_role` (RBAC), `overrides` (OGAR `classid→ClassView` inheritance). **`ClassActions`+`actions_for`** (zero-fallback) mirror `ClassMethods`/`methods_for`. **`effective_actions(parent, child)`** = OGAR inheritance on the action axis (child overrides parent by predicate). **`ActionInvocation`** (dynamic, `Copy`): lifecycle `ActionState{Pending→Committed|Failed|Cancelled}` (sticky terminals), S2.5 `cycle` stamp, idempotency/trace keys, HLC `emitted_at_millis`. **`ActionInvocation::commit(def, actor, impact, now)`** is the gated egress — RBAC FIRST (`auth::ActorContext` must hold `required_role` or be admin → else `Failed`), THEN MUL impact (`mul::GateDecision`: `Flow→Committed`+stamped, `Hold→`Pending/escalate, `Block→Cancelled`). This IS "commit to the external consumer (odoo/openproject/woa/tesseract) after the cycle decides sound." Dispatched via `UnifiedStep`/`ExecTarget`, NOT a per-crate endpoint. Additive, zero-dep. +5 tests green. Consumer reference: `docs/OGAR_CONSUMER_API.md`. Branch `claude/soa-write-deinterlace-inc2`.
 
+> **2026-06-20 — ADDED (D-UNICHARSET-DIR-MIRROR, the bidi-direction + mirror leaf)**: `lance_graph_contract::unicharset::UniCharSet` gained `get_direction(id) -> i32` + `get_mirror(id) -> i32` + `dump_direction()` + `dump_mirror()`, backed by `directions: Vec<i32>` + `mirrors: Vec<i32>`. The two columns after `other_case`, read by continuing the per-line token walk (the bbox+stats CSV is one whitespace token → fixed offsets across all 5 column tiers; no bespoke tier detector). `direction` = ICU `UCharDirection` code, load default `U_LEFT_TO_RIGHT` 0, out-of-range → `U_OTHER_NEUTRAL` 10 (`unicharset.h:712`). `mirror` clamped like other_case, out-of-range → -1 (`unicharset.h:721`). **Byte-identical 112/112 each** vs tesseract's own `get_direction`/`get_mirror` on real `eng.lstm-unicharset` (self-validating oracle; direction 6 distinct values, mirror 10 bracket pairs). Additive, zero-dep. +3 tests (26 unicharset total). Consumed by `tesseract-core::CharSet::{get_direction,get_mirror}`. EPIPHANIES `E-CPP-PARITY-6`; sixth leaf of `PROBE-OGAR-ADAPTER-UNICHARSET`; first to read past the bbox CSV. Remaining sub-leaf: the float stats inside the CSV. Branch `claude/happy-hamilton-0azlw4`.
+
 > **2026-06-20 — ADDED (D-UNICHARSET-OTHERCASE, the case-pair leaf)**: `lance_graph_contract::unicharset::UniCharSet` gained `get_other_case(id) -> i32` + `dump_other_case()`, backed by `other_cases: Vec<i32>`. The case-paired unichar id (`'C'`→`'c'`), parsed as the token after the script and clamped at load (`unicharset.cpp:901`: a value `>= size`, and the absent default = size, fold to the id itself). Out-of-range id → `INVALID_UNICHAR_ID` -1 (`unicharset.h:703`). **Byte-identical 112/112** vs tesseract's own `get_other_case` on real `eng.lstm-unicharset` (self-validating oracle `other_case` mode; 60 self / 52 pairs). Additive, zero-dep. +4 tests (23 unicharset total). Consumed by `tesseract-core::CharSet::get_other_case`. EPIPHANIES `E-CPP-PARITY-5`; fifth leaf of `PROBE-OGAR-ADAPTER-UNICHARSET`; the last field reachable by token-offset (direction/mirror/bbox need the multi-tier parser). Branch `claude/happy-hamilton-0azlw4`.
 
 > **2026-06-20 — ADDED (D-UNICHARSET-SCRIPT, the script-table leaf)**: `lance_graph_contract::unicharset::UniCharSet` gained `get_script(id) -> i32` / `get_script_table_size()` / `script_from_script_id(sid) -> Option<&str>` / `script_of(id) -> Option<&str>` / `dump_script()`, backed by new `script_ids: Vec<i32>` + an interned `scripts: Vec<String>`. The first leaf to transcode an **interning side-table** (`add_script`, `unicharset.cpp:1063`): `null_script` "NULL" seeded at sid 0 (the `unichar_insert` set_script, `unicharset.cpp:680` → `null_sid_ == 0`), real scripts intern from 1 in id order. Script name = token after the optional bbox/stats CSV (mixed-tier safe). Out-of-range → `null_sid_` 0 (`unicharset.h:681`). **Byte-identical 112/112** vs tesseract's own `get_script` on real `eng.lstm-unicharset` (self-validating oracle `script` mode; table `["NULL","Common","Latin"]`). Additive, zero-dep, behaviour-preserving on the bijection. +4 tests (19 unicharset total). Consumed by `tesseract-core::CharSet::{get_script,script_of}`. EPIPHANIES `E-CPP-PARITY-4`; fourth leaf of `PROBE-OGAR-ADAPTER-UNICHARSET`. Branch `claude/happy-hamilton-0azlw4`.

.claude/board/TECH_DEBT.md

@@ -52,6 +52,22 @@ enum, not a local rename. Canonical = the contract. Surfaced while grounding
 Deferred: cross-crate dep addition, out of scope for the convergence-probe
 increment. Same class as the resolved `CausalEdge64` shadow.
 
+### TD-CONTRACT-NOT-FMT-GATED — `lance-graph-contract` is not fmt-checked in CI (2026-06-20)
+
+**Open.** `.github/workflows/style.yml` runs `cargo fmt --check` only on
+`crates/lance-graph/` and `crates/deepnsm/` — NOT on `crates/lance-graph-contract/`.
+Consequence: merged PRs (symbiont / SoA work) have left rustfmt-1.9.0 drift in
+contract files — observed in `hhtl.rs:682`, `nan_projection.rs:125`, and
+`soa_graph.rs:{178,248,326,413}` (all whitespace/wrapping, behaviour-preserving).
+A local `cargo fmt -p lance-graph-contract -- --check` is therefore red on `main`
+even when a given PR's own files are clean. This has been re-discovered 3×
+(class_view.rs, nan_projection.rs, now hhtl/soa_graph) — recording it so the next
+session doesn't a 4th time. **Pay by** either adding a contract-crate `cargo fmt
+--check` step to `style.yml` (and a one-shot `cargo fmt -p lance-graph-contract`
+normalization commit), OR a deliberate decision to leave the contract crate
+ungated. Until then: leaf PRs keep their OWN files fmt-clean and do not reformat
+others' merged files (avoids muddied diffs + conflicts with in-flight PRs).
+
 ### TD-ONTOLOGY-LINT — `lance-graph-ontology` pre-existing clippy (12) on toolchain 1.95 (2026-06-18)
 
 `cargo clippy -p lance-graph-ontology -- -D warnings` exits 101 with 12 errors on the pinned 1.95 toolchain — all **pre-existing on `main`** (e.g. `odoo_blueprint/op_emitter.rs:182` is byte-identical on `origin/main`), in `hydrators/owl.rs` (2), `odoo_blueprint/op_emitter.rs` (1), `ttl_parse.rs` (3), + others. Mostly mechanical (`iter_cloned_collect` → `.to_vec()`, etc.). The crate is not in the CI clippy sweep ("CI tests 4 of ~30 crates"), so the debt accumulated un-gated. Surfaced while wiring `class_id_for_guid` (E-OGAR-ONTOLOGY-WIRED-1; `registry.rs` itself is clippy-clean + fmt-clean). Fix is a focused lint pass, out of scope for the wiring increment. Same class as `TD-CAUSAL-EDGE-LINT`.

crates/lance-graph-contract/examples/unicharset_dump.rs

@@ -1,7 +1,7 @@
-//! Dump a `.unicharset`'s id→unichar table (default), its per-id property bits
-//! (`properties` mode), its per-id script ids (`script` mode), or its per-id
-//! case-pair ids (`other_case` mode) — the Rust side of the byte-parity probe
-//! `PROBE-OGAR-ADAPTER-UNICHARSET`.
+//! Dump a `.unicharset`'s id→unichar table (default) or a per-id column:
+//! `properties` (category bits), `script` (script ids), `other_case` (case-pair
+//! ids), `direction` (bidi codes), `mirror` (mirror ids) — the Rust side of the
+//! byte-parity probe `PROBE-OGAR-ADAPTER-UNICHARSET`.
 //!
 //! ```sh
 //! # on a box with libtesseract + libleptonica installed:
@@ -33,7 +33,9 @@ use lance_graph_contract::unicharset::UniCharSet;
 
 fn main() -> ExitCode {
     let Some(path) = std::env::args().nth(1) else {
-        eprintln!("usage: unicharset_dump <path/to/eng.unicharset> [properties|script|other_case]");
+        eprintln!(
+            "usage: unicharset_dump <path/to/eng.unicharset> [properties|script|other_case|direction|mirror]"
+        );
         return ExitCode::FAILURE;
     };
     let mode = std::env::args().nth(2).unwrap_or_default();
@@ -43,6 +45,8 @@ fn main() -> ExitCode {
                 "properties" => print!("{}", unicharset.dump_properties()),
                 "script" => print!("{}", unicharset.dump_script()),
                 "other_case" => print!("{}", unicharset.dump_other_case()),
+                "direction" => print!("{}", unicharset.dump_direction()),
+                "mirror" => print!("{}", unicharset.dump_mirror()),
                 _ => print!("{}", unicharset.dump()),
             }
             ExitCode::SUCCESS