Imagine you're mapping an unfamiliar building by sending ping-pong balls through its hallways and measuring where they come out. If every ball that enters must exit through exactly one door (a tight, almost-unitary channel), you can exploit quantum interference to learn the layout with far fewer balls than classical statistics would require — that's the Heisenberg regime. But as soon as the building gains even one extra exit — a tiny leak — the question becomes: how fast does that quantum advantage evaporate? Chen and Oufkir answer this precisely for the first time. The committed claim: for quantum channels with input dimension d₁, output dimension d₂, and Kraus rank r, the dilation rate τ = rd₂/d₁ governs a phase transition in sample complexity. Prior work by CGO+26 and MB26 nailed the two extremes — τ = 1 (boundary, pure Heisenberg scaling) and τ ≥ 1 + Ω(1) (away from boundary, classical scaling). This paper closes the remaining gap: the near-boundary regime where 1 < τ < 1 + o(1). The problem is now fully settled. The result splits cleanly by error metric. For diamond norm tomography, the transition is abrupt: the moment τ leaves 1, Heisenberg scaling vanishes and you need Ω(rd₁d₂/ε²) queries — matching the known upper bound. There is no gradual interpolation; one extra degree of freedom kills the quantum speedup instantly. For Choi trace norm tomography, the story is richer. The complexity is Θ(d₁²/ε + min{(τ−1)d₁², (τ−1)²d₁²d₂²}/ε²), which smoothly interpolates between Heisenberg and classical scaling as τ drifts from 1. Two qualitatively different phase transitions in the same problem — that's the headline. The ladder here is the sequence of results from CGO+26 (which introduced τ as the governing parameter and solved the boundary and away-from-boundary regimes) and MB26 (which provided matching upper bounds). Chen and Oufkir's contribution is the lower bound machinery that closes the near-boundary gap. This isn't a new algorithm or a new protocol — it's a tight information-theoretic characterization. The paper doesn't beat a baseline; it completes the map. Architecturally, this is pure quantum information theory — no hardware, no circuits, no experiments. The tools are information-theoretic lower bounds (likely Le Cam-style or Fano-method reductions tailored to quantum channels), combined with careful analysis of how the Kraus-rank structure constrains the channel's distance from unitarity. The entire result lives in the query-complexity model where you have black-box access to the channel and count queries. Integrity is high for what this is: a mathematical proof paper. There is no simulation to cherry-pick, no benchmark to game. The result stands or falls on the correctness of the proofs across 40 pages. The matching upper bounds come from independent prior work (MB26, CGO+26), which means the tightness claim isn't circular — the lower bounds proved here meet upper bounds proved elsewhere by different authors. That's the gold standard for complexity characterizations. The practical milestone is further out. Quantum process tomography matters enormously for calibrating and verifying quantum hardware, but the gap between information-theoretic query bounds and actual experimental protocols involves gate errors, measurement noise, and finite coherence times. The immediate value is conceptual: hardware teams now know exactly how many queries are fundamentally necessary, which sets a floor for any practical protocol. The sharp/gradual distinction between diamond and Choi norms gives practitioners a concrete design choice — which error metric to optimize against — with fully characterized consequences.
