Card 60 — the time-arrow detector runs in Rust; both booby traps defused S3 DONE · ≤1e-9 GREEN

The Rust engine version of the time-arrow detector was built — walking the exact path the two traps from iteration 10 marked out.

Trap 1 was avoided by design. The graph builder that connects bars which can "see" each other uses the literal-rule scan (proven tie-safe last iteration), not the clever stack that hallucinated connections through equal-price walls. And the trap became a permanent tripwire: the test suite now carries the exact price patterns that fooled the stack version — [2,1,1,2] and friends — checked forever against a dumb-but-obvious reference implementation that lives inside the tests. If anyone ever "optimizes" this code and reintroduces the bug, the suite fails within seconds.

Trap 2 shaped how the tests assert. Mathematically, playing the window backwards flips this feature's sign perfectly. In floating-point it flips to within a hair (the order you add numbers matters in the 16th digit), so the reversal test demands agreement to 10⁻¹² rather than perfect bits — strict enough to catch any real asymmetry bug, honest enough to pass on correct code.

The verdict: the Rust engine reproduced all 19,602 frozen reference answers on Bitcoin and Ethereum within the sacred one-billionth tolerance (actual disagreement: ~10⁻¹⁵, from the summation-order effect alone). A hand-worked symmetric zig-zag [2,1,2,1,2] scores exactly 0.0 — perfect time-symmetry — and an always-rising market honestly answers "undefined" (its visibility network is a bare chain with nothing to cluster).

The shared machinery grew once more: the standard 200-bar record now carries 14 features (this batch: 12 → 13 → 14). The one red light is the same deliberate watchdog as always — the database blueprint doesn't know about column 14 yet. That wiring is next.