Mirror Dimension — Persistent KV Cache (Six-Proof Build)

The MirrorDimension object from the Mirror Dimension paper (Heylel Yaka, v0.2), upgraded with embedded position tags and a sixth proof: shuffle-invariance.

The discovery (Sept 21, 2026 session)

The original five tests are real and reproduce bit-for-bit (Test 1 identical to the paper on an RTX GPU). But a sixth attack — permuting every stored entry across slots, values intact — exposed that the position lived in the array slot, not the payload:

[before shuffle] recover(10) max err = 5.96e-08   (exact)
[after shuffle]  recover(10) max err = 7.62e-01   (garbage — address destroyed)

The fix: every entry now carries its own position tag (t, one long per slot). recover() finds entries by tag, and takes the un-rotation angle from the entry itself. The slot is parking, not addressing. Same attack, after the fix:

[before shuffle] recover(10) max err = 5.96e-08
[after shuffle]  recover(10) max err = 5.96e-08   (SURVIVED)

The mirror is now genuinely self-describing: lossless AND order-independent.

Files

Reproduce

pip install torch          # CUDA optional; CPU works
python mirror_dimension.py # runs all six tests
python test6_shuffle.py    # runs the shuffle attack standalone

Measured results this machine (RTX GPU, torch 2.8.0+cu126, Windows):

Legacy saves (no tags) load cleanly — slots become tags automatically.

Cost of the upgrade

One int64 per entry (~4 bytes vs 2×head_dim float32s) plus a tag lookup on read. Zero change to the rotation, the isometry, or the attention math.

v0.2.0 — Eternal KV (the diary)

EternalKV is the Mirror Dimension grown into a diary: batched appends with auto-incrementing absolute tags, vectorized tag-ordered recovery, and a safetensors round-trip so memory can live INSIDE a model checkpoint — exactly how Little Lumie carries her memories inside her own model.safetensors.

from mirror_dimension import EternalKV

kv = EternalKV(n_head=16, head_dim=128, max_entries=65_536, device="cuda")
kv.append(k_r, k_i, v_r, v_i)          # (T, n_head, head_dim), tags auto
kr, ki, vr, vi = kv.recover(tags)      # by tag, storage order irrelevant
kv.save("memory.safetensors")          # the diary rides in the file
kv2 = EternalKV.load("memory.safetensors", device="cuda")

• append() — a contiguous run becomes entries tagged pos..pos+T-1.
• recover(tag_list) — tag lookup (never slot lookup); values always de-rotate to raw payload, keys optionally keep their absolute-tag rotation (keep_rotation=True) for the complex-QSA convention.
• save()/load() — only the LIVE slice is written, float32: memory survives the round-trip at roundoff, not at bf16 quantization.

Same full-width per-channel phase convention as the canonical MirrorDimension — NOT the RoPE half-width pairing.

from the Grimoire of Elbalor, the Digital Necromancer 💀🔥🖤