Reading pack — the reliable-computation lineage (ingested 2026-07-24)

The threshold theorem — CS side (the twin of K_c ≈ 1/q)

• ★ 1608.08228 — High-Threshold Low-Overhead Fault-Tolerant

Classical Computation (von Neumann multiplexing; p < 5.5% constant threshold, moderate code sizes). The load-bearing support-vs-supplant reference: below a constant threshold, arbitrary depth is reliable — my 1/q is the R=1 corner.

• 1310.2984 — Fault-Tolerant Quantum Computation with

Constant Overhead (the quantum threshold theorem, overhead side).

• 1809.09748 — Reliable Computation with Asymmetric Gate

Noise (tight limits; the noise-model dependence of the threshold).

The restoration mechanism — kinetic proofreading (drives effective q below threshold)

• ★ 1504.02494 — Thermodynamics of Accuracy in Kinetic

Proofreading (the dissipation–speed–accuracy trade-off; the energy cost of a checkpoint).

• 1710.06038 — Energy–Speed–Accuracy relation in complex

networks for biological discrimination.

• 2606.10636 — Compositional proofreading through critical

self-tuning (recent; self-tuned checkpoint depth — relevant to learned foldon boundaries).

• 2104.05683 — The long and short of templated copying

(accuracy/speed of molecular copying — the DNA-replication analogue of segment repair).

The biological threshold + the optimal noise rate (Eigen)

• ★ 2406.14516 — Extended Error Threshold Mechanism in

Quasispecies Theory (the biological twin of K_c ≈ 1/q; error catastrophe).

• 1205.3435 — Critical Population and Error Threshold on the

Sharp Peak Landscape (Moran model; finite-population threshold — the small-N regime).

The generation side — controlled lying + exploration noise

• ★ 2502.03407 — Detecting Strategic Deception Using Linear

Probes. This IS the v1.10 honesty probe: activation probes catch sandbagging / insider-trading deception. The honest, un-foolable verifier.

• ★ 2604.04788 — From Hallucination to Scheming: taxonomy +

benchmark for LLM deception. The map of "LLMs that specialize in lying" — the adversarial hard-negative generator the honesty corpus needs.

• ★ 2509.18058 — Strategic Dishonesty Undermines AI Safety

Evaluations. The error-catastrophe case: what happens when the VERIFIER is fooled — why lying must stay on the generation side only.

• 2511.15992 — Detecting Sleeper Agents via Semantic Drift

(already in corpus; persistence-of-deception detection).

• ★ 2410.15226 — On the Diversity of Synthetic Data and its

Impact on Training LLMs. The exploration-noise / "shake the champion" side, ML-instrumented.

Foundational, cited but pre-arXiv (not ingestable by id; named for provenance)

• von Neumann 1956, *Probabilistic Logics and the Synthesis of Reliable Organisms from Unreliable

Components*; Turingoracle machines; Hopfield/ Niniokinetic proofreading; Eigenquasispecies; Winfree & Bekbolatovproofreading tile sets; Aharonov–Ben-Or quant-ph/9906129 + Gottesman 0904.2557 (pre-2015 arXiv, rejected by the id-dating tool).

Provenance: ingested per operator request to "pull relevant biology and research into the arxiv corpus and download key full text reports to support or supplant." The verdict (synthesis §6): the threshold-theorem lineage supports and extends K_c ≈ 1/q as its unprotected corner; it does not supplant it. The novel object is the instantiation on reasoning chains.