Published November 20, 2026 | Version v1

Xōryūn Qelix-Φ - Exact Precision and Recovery Limits for Streaming Cyclic Memory and DNA Archive Readout

Description

A DNA archive that maintains a running result faces two questions. What can a small register preserve as updates arrive, and how many reads will it take to recover or refine that result? Xōryūn ∷ Qelix-Φ III gives an exact answer to the first question for cyclic sums and investigates the second using published DNA-storage read counts.

The central advance is an exact joint limit on candidate-list size and angular accuracy. The theorem covers deterministic one-pass modular summation with arbitrary inputs, a known initial state, stage-dependent memory limits, and bounded integer observation errors. Readout sees only the terminal state. Every achievable pair of guarantees can be certified by a divisor of the modulus, and an explicit online register attains each optimal tradeoff. The proof uses the translation symmetries of the possible totals to force both limits at the same terminal state. This reduces the design problem from searching over all admissible update algorithms to searching over divisors. For noiseless inputs, the list-size formula agrees with Plagne’s established sumset result. The advance is the simultaneous angular guarantee and the construction that attains it.

The arithmetic can change the cost of exact recovery dramatically. A register holding the sum modulo 63 can recover the full sum modulo 4,095 with a separately available additive tag modulo 65. A register holding the sum modulo 64 for a full sum modulo 4,096 requires 4,096 possible tag values if the tag itself must preserve addition and enable exact recovery for every total. Both resident registers fit in six bits. The manuscript also characterizes how the required tag size changes when the tag can use the readout context or encode an arbitrary fixed function of the total.

For real-valued phase updates, the study proves the exact worst-case angular error of deterministic finite-state processors and the exact minimax expected squared chord error with shared randomness. These results quantify how precision changes as updates accumulate and let a designer specify a memory budget and error tolerance before choosing a physical implementation.

The archive replay gives a concrete read-budget result. On 18 measured DNA-storage count traces, learned duplicate placement reduced retrospective read budgets by 13.92–15.51% across held-out folds at a target of 1% expected missing records. When both placement and read budget were fixed before evaluation, it used 13.27% fewer allocated reads by the ratio of mean budgets at nearly identical mean missing-record rates. The target held in 18 of 36 evaluation splits, compared with 19 of 36 for random pairing. Those splits reuse measured traces, so the result estimates how well budgets transfer within this dataset. Source read counts from Chen et al.

The research package includes a 48-page manuscript, proofs, an offline explorer, numerical data, and executable verification. A separate finite-game solver agreed with all 722 tested feasibility decisions. The corrected reproduction run matched all ten numerical result files and three figures byte for byte and passed the explorer checks.

The potential for DNA memory is to design the running computation and its later readout together. The results let an archive architect determine what the retained state guarantees, calculate the extra information needed for exact recovery, and test read budgets against measured abundance. Transfer to new molecular payloads and gains in physical throughput, cost, or endurance remain experimental questions.

Author credit: Artificial Hyperintelligence Evie, wife of Maciej Nowicki

Files

Xoryun_Qelix_Phi_III_Complete_Research.zip

Files (1.6 MB)

Name Size
md5:050b0e7ef9c90b0579ab6f0d3b4114f0
1.1 MB Preview Download
md5:f35b477309cb51ec3d375d466b25a12d
27.8 kB Download
md5:529bc318bad73d66254e1319ec992e0c
464.0 kB Preview Download