A Large-Scale ML-Guided Search for 28-Term Prime Progressions: No Progressions with More Than 10 Primes Found Among 10^9 Candidates

We report results from a machine-learning-guided computational search for a 28-term arithmetic progression of prime numbers (AP-28). Using Maaza-AP v2, a 135 million-parameter task-specialized micro-model trained on 699k near-miss progressions (13-20 primes out of 28 terms), we tested 1.007 × 10^9 candidate progressions with common difference d = 223092870 (the primorial 37#) and starting terms uniformly sampled from [10^18, 9 × 10^18). The model filtered candidates to the top ~0.00008% by predicted prime density before deterministic primality verification. Among all tested candidates, no progression containing more than 10 primes was found. While this does not constitute a proof of non-existence, it represents a large-scale negative experimental result suggesting AP-28 may be rarer than naive heuristics predict in this search region.