A 3-D rendering of the event spaces and patterns of the p8v2 P-program, run over a fixed 128-byte prefix of enwik9, with k = 8 and N = 128. Rendered from p8v2-viz-spec.md; every figure below is reproducible from that file alone.
PROVISIONAL — f and ω are not specified
Five things in these figures are scaffolding, not design. They exist so the program runs. Everything derived from them is drawn dashed, in warning colour, and named here and on every frame.
strengths
log-scaled occurrence counts
stands in for LSA
keep/prune score
occurrence count, budget never binds
stands in for omega
composite predict
longest context wins
rejected by the design
trace membership
composite mispredictions
belongs to omega, unspecified
settling
iterative relaxation, 24 iters
stands in for f
byte events — the byte ES and every mem_cell_ik-models — the k0 profile and the k1 edgestokens — nodes, AND-gate inputs, consequentsmemchain — inert scaffolding, not a seriesPROVISIONAL — dashed
The fixture
Exactly 128 bytes, all printable US-ASCII, verified byte-for-byte against the first 128 bytes of a real enwik9 before these figures were rendered.
sha256 b492bc32a8260bc10e8f1278f7f362fdece48d097ff9954cca681b86aa3193f5. Colons are written as character references so this page contains no literal http scheme text, which acceptance criterion 11 forbids.
Stills
1. t = 8 — the first full k = 8 window — Eight bytes are in the chain, so every level from k = 2 to k = 8 has an antecedent for the first time. All seven token nodes are present, each with its k AND-gate inputs reaching down to the byte events they gate on, and each with a consequent edge onto the live input at x = 127.2. t = 22 — the http: example being learned — The design document's worked example. B[18..22] is 'http:'. This is the step whose learned patterns Phase C later settles against.3. t = 127 — the end of the scan — The chain is full. Cells past lag 32 have faded to gridline ink: they are the inert tail, carrying the memory-chain pattern and nothing else. Trace rings have accumulated over the whole run.4. Phase B — the learned model at rest, camera iso — Every kept pattern at every level, each drawn at the offset where it was first learned. The iso camera crops x to [88, 142], so this shows the settling window and the k0 profile; the inert tail is in the scene but outside the crop (§7).5. Phase C iteration 0 — PROVISIONAL — Cells for indices 18, 19 and 22 are clamped; 20 and 21 are open and start at the k = 0 marginal — which is what 'no information yet' actually means here.6. Phase C iteration 4 — PROVISIONAL — Token instances above AND_THETA fan out across offsets: not just the alignment starting at 'ht' but every overlapping one.7. Phase C iteration 23 — PROVISIONAL — The final relaxed state. Whether it reaches 'http:' is not an acceptance criterion (§12.7); the caption reports what happened.8. Phase B, camera levels — down the chain axis — Looking along x. The levels stack up the z axis and the fan-out of token nodes over byte values is visible. z = 6 is empty by design.9. Phase B, camera top — plan view — Looking down from above: lexicon occupancy across the chain, per level.10. Phase D — trace comparison, camera chain — Pure side-on: the value axis collapses and the chain reads as a strip of 128 cells, so trace membership reads as a barcode of surprises. Class-2 rings are trace_k1, class-3 rings are trace_composite.
Animations
p8v2-scan.gif
Phase A + Phase B held — 128 frames of the learning scan, then the model at rest. A byte enters at the right edge and ages leftward; learned predictions run left to right onto it.
p8v2-settle.gif
Phase C — settling — 24 relaxation iterations on the http: example. Entirely PROVISIONAL: this is a plain iterative relaxation standing in for a process the design specifies as continuous, bidirectional and frequency-domain.
