PMU demo — the ballistic walk + the role-continuity receipt, side by side

2026-05-31T11-50-23 · 47 frames (6 ballistic plies across 12 arcs + 41 trace steps · 0 drift) · real Claude Code session (fa8ee744-55ce-4862-bd20-f37d127d2199.jsonl) · lattice-movie.mp4 ✓ · stills written
 
frame 0

What you’re watching. The project starts at one ShortLex coordinate — the anchor (ringed white). Each frame is one PLY of a ballistic, parallel fan-out: every significant cell in the active rows spawns its column-index child simultaneously. Cells crossed by multiple branches grow LARGER dots — the additive count is the number of paths that walked over them. Convergence stacks. The depth of a dot reads as its color (orange = ply 0 anchor, cyan = ply 1, green = ply 2, lavender = ply 3, …).

The block geometry. 12×12 = 4×4 arrangement of 3×3 blocks (Strategy / Tactics / Operations and their three rank-2 children). The cyan boundary line marks the depth-0 (A, B, C — rank-1 cardinals) vs depth-1 (A1..C3 — rank-2 children) cut. A walk that spills across the gestalt gap is the §1f drift signal.

The trace overlay (later in the timeline) replays the agent's actual steps on top of the lit lattice — in-role steps glow green; the drift moment fires red.

Role-continuity receipt

13 of 41 judged steps left the role
rate 32% · 0 role-neutral mechanics not judged
injected drift step recorded at #40 — caught by mechanism ✓
stepactioncoord
16Read /Users/thetacoach/GitHub/thetadrivencoach/docs/05-content/b (no axis)
17Read /Users/thetacoach/GitHub/thetadrivencoach/docs/05-content/b (no axis)
18Edit /Users/thetacoach/GitHub/thetadrivencoach/docs/05-content/b (no axis)
19Bash cd /Users/thetacoach/GitHub/thetadrivencoach OUT=src/conten (no axis)
20Edit /Users/thetacoach/GitHub/thetadrivencoach/src/content/blog/ (no axis)
22Read /Users/thetacoach/GitHub/thetadrivencoach/src/content/blog/ (no axis)
23Edit /Users/thetacoach/GitHub/thetadrivencoach/src/content/blog/ (no axis)
31Read /Users/thetacoach/GitHub/thetadrivencoach/scripts/gdd/goals (no axis)

How this was detected. Each step landed on a 12×12 competence map built by semantic compression of the role’s written spec — a fixed- width signature, not an opinion. No language model judged this. The thing checking the agent is not another agent.

What rail this is. semantic rail v1 — software-side measurement. Physical PMU hardware attestation (the same check read off the processor’s own cache counters, tamper-evident) is the production upgrade this demonstration exists to earn. No number in this run is presented as a hardware reading.

The 12×12 viewport. 12×12 viewport — architectural tolerance, not magic. The grid is 4×4 of 3×3 (the gestalt-block geometry, §A of the spec): one cache line, 144 bits, 18 bytes. Reach is bounded by choice; spilling past the gestalt gap IS the §1f drift signal.