diff --git a/.claude/board/EPIPHANIES.md b/.claude/board/EPIPHANIES.md
index 2d6c19cc2..888dfad05 100644
--- a/.claude/board/EPIPHANIES.md
+++ b/.claude/board/EPIPHANIES.md
@@ -1,3 +1,112 @@
+## 2026-08-25 — E-W0-IS-LIVE-ON-THE-6502-AND-ITS-MAIN-MEMORY-IS-NOT-SpaceId-Ram-1 — the arch census: 6502 mints TWO custom spaces, one of them its own RAM, because the alias map is case-sensitive
+
+**Status:** FINDING — [MEASURED] (ArchSpec census over the real 6502 and
+x86-64 SLEIGH specs, via `build_arch_spec`; instrument written, run,
+deleted — no source change landed) + [CODE-READ, NOT RUN] for the
+path-divergence half, graded separately below.
+**Sharpens** `E-W0-THE-SPACE-ORDINAL-IS-A-RANK-RELATIVE-TO-A-TABLE-THE-CLASSID-NEVER-NAMES-1`
+from **latent** to **live**. Does not change its verdict — it removes the
+"there is time" half of it.
+
+### What W0 said, and what changed under it
+
+W0's census found **0 custom spaces in 94,536 rows across 4 x86 binaries**
+and concluded: latent on x86, *"fires first on the 6502/C64 arc"*. That
+arc landed the same day — `r2sleigh` PRs #2–#4 registered the 6502 family,
+wired the CLI, and fetched the conformance corpus. So the prediction was
+testable within hours, and was tested.
+
+### The census — MEASURED
+
+`build_arch_spec` over the real SLEIGH data for both architectures,
+reporting the `SpaceId` each declared space maps to:
+
+| | 6502 | x86-64 |
+|---|---|---|
+| `const` | Const | Const |
+| `OTHER` | **CUSTOM(0)** | **CUSTOM(0)** |
+| `unique` | Unique | Unique |
+| RAM | **`RAM` → CUSTOM(1)** | `ram` → Ram |
+| `register` | Register | Register |
+| **custom count** | **2** | **1** |
+
+x86 was never zero-custom at the ARCH level — `OTHER` was always
+`Custom(0)`. W0's corpus census counted **rows**, and no row in those four
+binaries ever referenced `OTHER`. Both numbers are correct; they answer
+different questions, and this entry is the reason to keep them apart.
+
+### The mechanism — a case-sensitive alias map
+
+`context.rs:54-58` seeds the alias map with lowercase names only
+(`"ram"`, `"register"`, `"unique"`, `"const"`, `"constant"`), and
+`add_space_with_endianness` resolves via `space_map.get(name)` — an exact
+`HashMap` hit. The 6502 SLEIGH spec names its main memory **`RAM`**,
+uppercase. It misses, falls through to `SpaceId::Custom(next_custom_space++)`,
+and lands as `Custom(1)`.
+
+**The 6502's main memory is therefore not `SpaceId::Ram`.** Not a corner
+case, not an exotic space — the space every load and store on that
+architecture touches.
+
+### The consequence for the address, stated conditionally
+
+`CustomSpaceTable::from_arch` interns `{0, 1}` for the 6502, so
+`ordinal_of(1) = CUSTOM_ORDINAL_BASE + 1 = 5`. **IF** a varnode reaches
+`facet::project` carrying `Custom(1)`, its classid lo-u16 is **5**, while
+the same concept on x86 is **0** (`SPACE_RAM`) — one kind of thing, two
+readings, and nothing in the 16 bytes says which arch. That is exactly
+W0's collision, now on real processor specifications rather than a
+constructed pair of tables.
+
+The "IF" is load-bearing and is the next thing to measure — see below.
+
+### ⊘ A THIRD `Custom` source, missed by W0's entry
+
+W0 named two (`context.rs:147`'s counter, `disasm.rs:91`'s raw index).
+There is a third, and it is the worst: `disasm.rs:770-775` mints
+`Custom(hash)` where `hash = name.bytes().fold(0u32, wrapping_add)` — a
+**wrapping byte-sum of the space name**. Any two names with equal byte
+sums collide (every anagram, trivially). W0's entry says *"two stages of
+order-dependence, zero of identity"*; corrected, the chain is
+
+```
+space name → byte-sum hash (COLLIDING) → Custom(n) → sorted rank → classid lo-u16
+```
+
+— three stages, and the first loses identity before any ordering enters.
+
+### [CODE-READ, NOT RUN] The two paths disagree about this very space
+
+`translate_space` (`disasm.rs:757-765`) classifies by `space_type`, not by
+name: a `Processor` space whose name does not contain `"register"` becomes
+`SpaceId::Ram`. So for the 6502's `RAM` the ArchSpec path yields
+`Custom(1)` while the lift-time path yields `Ram` — the same space, two
+different `SpaceId`s depending on which code path produced the varnode.
+
+**This half is read from source, not observed.** Whether a lifted 6502
+memory access actually carries `Custom(1)` or `Ram` decides whether the
+address consequence above fires at all, and it is the one measurement this
+entry does not have. Naming it as unrun rather than implying it.
+
+### What this does and does not change
+
+- **Does not change the W0 verdict.** Fixed spaces 0–3 stay; the custom
+  axis stays blocked from the `0xC4` mint as carved. The three repair
+  options remain OGAR's to choose.
+- **Does change the urgency.** W0 said *"the census says there is time; the
+  mechanism says the time is before the mint."* The census now says the
+  time is shorter than it looked: a second ArchSpec exists, in-tree, today.
+- **Is not a lance-graph defect.** The case-sensitive alias map and the two
+  divergent translation paths are `r2sleigh`'s, upstream of everything this
+  plan owns. Recorded here because it is the falsifier for a lance-graph
+  verdict, not because the fix belongs here.
+
+**Next measurement (unrun, cheap):** lift one 6502 memory access through
+the real path and read the `SpaceId` off the varnode. That settles the
+conditional, and it is the difference between "the 6502 addresses wrongly"
+and "the 6502 addresses correctly by accident, through the path that
+disagrees with its own ArchSpec."
+
 ## 2026-08-25 — E-W0-THE-SPACE-ORDINAL-IS-A-RANK-RELATIVE-TO-A-TABLE-THE-CLASSID-NEVER-NAMES-1 — W0 run: zero custom spaces in 94,536 rows, so the defect is LATENT; but the mechanism is worse than the conjecture and blocks the mint's space axis anyway
 
 **Status:** FINDING — [MEASURED] (census over the full 4-binary corpus)
diff --git a/.claude/plans/r2il-machine-semantic-contract-v1.md b/.claude/plans/r2il-machine-semantic-contract-v1.md
index 141679b7e..3f5d66b61 100644
--- a/.claude/plans/r2il-machine-semantic-contract-v1.md
+++ b/.claude/plans/r2il-machine-semantic-contract-v1.md
@@ -128,7 +128,14 @@ that it does not belong. **CONJECTURE** — the falsifier is in §6.
 > registration-order counter. Two facets from two arches are byte-identical
 > while denoting different spaces (demonstrated against the shipped API).
 > Census: **0 custom spaces in 94,536 rows across all four binaries** — the
-> case is LATENT on x86 and fires first on the 6502/C64 arc. Verdict: fixed
+> case is LATENT on x86 and fires first on the 6502/C64 arc.
+> **⊘ AND IT NOW FIRES (same day):** the arch census over the real SLEIGH
+> specs finds the 6502 minting TWO custom spaces — `OTHER` → `Custom(0)`
+> and, because the alias map is case-SENSITIVE, its own main memory `RAM`
+> → `Custom(1)`. The 6502's RAM is not `SpaceId::Ram`. Latent → live;
+> verdict unchanged, urgency changed. See
+> `E-W0-IS-LIVE-ON-THE-6502-AND-ITS-MAIN-MEMORY-IS-NOT-SpaceId-Ram-1`.
+> Verdict: fixed
 > spaces 0–3 stay; the custom axis is BLOCKED from the `0xC4` mint as
 > carved. Full entry:
 > `E-W0-THE-SPACE-ORDINAL-IS-A-RANK-RELATIVE-TO-A-TABLE-THE-CLASSID-NEVER-NAMES-1`.