Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 37 additions & 0 deletions .claude/board/PR_ARC_INVENTORY.md
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,43 @@
> census §8.3 trap 10: read the body FIRST, then open for write — never
> inline both in one expression.

## 2026-09-01 — lance-graph #1130 (MERGED c630ab9) — loco band mirror + armed parity; two palette rulings recorded

**Added.** `epistemic_bassin::loco_band::EPISTEMIC_CALLS` (zero-dep mirror of ogar-loco's epistemic core band `0x86..0x8B`) + `lance-graph-ogar::parity::assert_epistemic_band_parity()` in the excluded armed tier — written against loco's shared-core TABLE by raw index so it compiles against any revision; the assertions carry the parity. Verified GREEN against OGAR main after OGAR #296 merged (both directions + count pin 6).

**Locked.** The 256:256 ruling: the classid swaps the whole 256-entry palette (r2il and dismech already share `0x90..` indices under different classids); the domain-shaped rest (24-axis catalogue as a `ValueCodebook`, per-axis macros) belongs in an epistemic vocabulary under its own classid; compound macros take r2il's byte-addresses-a-360-byte-script microcode shape. Core placement of the six universal calls is what makes them readable in EVERY vocabulary (`compose` cannot be forged).

**Deferred.** The rung-per-classid extension commit (`b175502`) STRANDED at merge — cherry-picked forward in the next PR. Hambly-Lyons stays bandless (jc Pillar 11 red).

**Docs.** Plan fifth-round block; companion OGAR #296 (mint, census re-pin 95→101).

- **Confidence (2026-09-01):** working — contract 1390/0, clippy/fmt clean; armed parity red-then-green across the OGAR merge boundary exactly as designed.

## 2026-09-01 — lance-graph #1129 (MERGED a16e89d) — EpistemicBassin24: the pair supersedes the falsified signed net

**Added.** `epistemic_bassin::{EpistemicBassin24, AxisState, info_gain_u4, sigma_tension_u4, sweep_ternlog, eval_ternlog}` — `agree_u4[24] + disagree_u4[24]`; net/contest/entropy DERIVED; the REVERSED falsifier (contested ≠ silence, survives accumulation); Belnap-mask projections with the knowledge-join theorem pinned (join = OR-OR = accumulate's state layer, incl. a Both cell that exists only at the join); the sweep queries as ternlog truth tables pinned exhaustively against the scalar oracle (`ASKED_SILENT 0x02` = the missing link in one instruction).

**Locked.** The co-architect ruling: #1127's loci-never-magnitude law re-scoped to the A9 READING (readings are classid-selected per row); the `EpistemicWitness = 16` tenant mint discarded UNCOMMITTED (no mint until one real row needs both readings); lanes carry proprioception never evidence, each grounded in a shipped certificate (Shannon → `dismech_candidates` counts; EWA → `sigma_propagation::pillar_5plus_bound`; Hambly-Lyons laneless while jc Pillar 11 is red). `basin_lanes` superseded in place, its collapse pin kept green as the record.

**The finding worth keeping.** Two disable-run PASSES were findings: stepwise saturation masked by pack's clamp AND order-dependent for mixed signs (fixed: exact-sum-then-clamp); the sigma-tension calibration vacuous because every arm used `bound = 1.0` (fifth constant-that-does-not-bind of the arc).

**Docs.** EPIPHANIES ×3; STATUS_BOARD superseded-in-place; plan fourth-round block.

- **Confidence (2026-09-01):** working — 1385/0 → 1390/0, clippy/fmt clean, seven disable-runs red-then-green.

## 2026-09-01 — lance-graph #1128 (MERGED 9742e1a) — one-hop field-map slices, rebased + reconciled with the #1127 census

**Added.** `basin_lanes::BasinLanes` (since superseded by #1129 — kept per reserve-don't-reclaim) + `accumulate_children`; `hhtl::{missing_ancestors, direct_children}` (mechanical-hydration address list + the one-hop selector, addresses only — the type-level half of the mechanical/epistemic split).

**Locked.** DN dissolution (every path truncation is an entry; tree-siting questions dissolve); mechanical hydration writes structure only, epistemic output = silence; the one-hop law (a parent expresses its DIRECT children; second hop continues or cancels).

**The finding worth keeping.** Three parallel branches worked one register; the reconciliation audit ran BEFORE re-pushing, so my falsified "not in the shipped set" claim and the conflated siting axes were corrected in the same landing — and the per-lane arithmetic arrived at the identical shape in two independent sessions (convergence kept as evidence, cross-referenced with `w2b_one_node_field.rs`).

**Docs.** EPIPHANIES audit entry; plan §W2b second/third-round blocks; `episodic_basin` module-doc clarification.

- **Confidence (2026-09-01):** working — 1378/0 at merge; the whale falsifier holds at carrier and register level.


## 2026-08-31 — lance-graph #1099 (MERGED ae24f6e5) — D-MAR-1 re-derived from the plan, and a gate that had silently stopped being one

**Added.** `FieldMask::{difference, is_subset_of}` and the `WideFieldMask` pair (plan §2.1). `chunks_view` removed — it cloned both operands per wide op; `zip_fold` now reads in place via `chunk_at`.
Expand Down
14 changes: 14 additions & 0 deletions .claude/plans/dismech-causal-replay-v1.md
Original file line number Diff line number Diff line change
Expand Up @@ -528,6 +528,20 @@ Consequences, recorded before the catalogue is designed:
palette addresses a whole loco body (360 bytes = 180/120/90 calls under
the lane shape) — scripts all the way down, each still replayable.

**⊘ Extension (operator, same day): one classid per RUNG level — the brutal
form.** Each rung of the reasoning ladder gets its own 256-entry palette, so
a program's vocabulary IS its epistemic altitude: rung-2 bytes resolve
against the 144 verb atoms, rung-3 against the 34 NARS tactic recipes (THE
runbooks), rung-4 against StyleFamily macros — and with the microcode shape,
**one byte at rung N addresses a whole script at rung N−1**. Escalation and
de-escalation become ADDRESSING operations (a classid change), not control
flow; every level stays replayable because every level is still loco bodies.
This also mirrors, at the palette level, the tier-graded register ruling at
the node level (granularity down, aggregation up). Consequence for W5
(D-DCR-5, HELD): the operator's rung 5–9 table, when it lands, is now also
the CLASSID MINT LIST for the upper palettes — the hold therefore stays
exactly where it is; no rung classid is minted before the table.

#### Still open — probes, not rulings

- **The named 24-axis catalogue** — WHICH axes, minted as a classid-named
Expand Down
Loading