With a reserved instance (1yr): $10.80/mo. Spot instance for non-critical workloads: $5.40/mo.
| Operation | Avg latency | Compute cost | At 10K req/day |
|---|---|---|---|
| POST /reduce (small, <100 cls) | 3ms | $0.0000001 | $0.001/day |
| POST /reduce (medium, ~1000 cls) | 50ms | $0.000002 | $0.02/day |
| POST /verify (with Kissat) | 200ms | $0.000008 | $0.08/day |
At t3.small capacity (~100 concurrent req/s for /reduce), the server handles 8.6M requests/day before saturation. Far beyond expected usage.
| Tier | Tokens | Clauses | AUMA evals (a=2) | Wall time |
|---|---|---|---|---|
| Light | < 5K | < 120 | n² ≤ 2,500 | < 200ms |
| Medium | 5K–15K | 120–350 | n² ≤ 22,500 | 200ms–2s |
| Heavy | 15K–50K | 350–1190 | n² ≤ 250,000 | 2s–40s |
| Tier | Instance | Capacity | Cost/mo | Cost/1M reductions |
|---|---|---|---|---|
| Dev | t3.small | 8.6M req/day | $18 | $0.06 |
| Prod | c6i.large | 50M req/day | $62 | $0.04 |
| Scale | c6i.xlarge | 100M req/day | $124 | $0.04 |
CPU-bound workload (BCP propagation). Scales linearly with cores. No external dependencies (no DB, no S3, no network calls in hot path).
| Direct SAT solver | Uniform 3-SAT + solver | |
|---|---|---|
| PHP(10,9) solve time | 181ms (Kissat) | 72ms (reduce 10ms + Kissat 62ms) |
| Sudoku solve time | 3ms | 53ms + 6ms = 59ms |
| Isomorphism detection | N/A | O(1) fingerprint match |
| Cross-problem reuse | None | Solved corpus transfer |
The reduction adds overhead for easy instances but pays for itself on hard instances (PHP, dense graph coloring) where the uniform topology is easier for CDCL. The real value is in the corpus: once an isomorphic instance has been solved, every future match resolves in O(1).
| Operation | Time complexity | Token cost | $ cost (t3.small) |
|---|---|---|---|
| Reduce 100-var CNF | O(n·m²) ≈ 10ms | 0 (internal) | $0.0000001 |
| AUMA solve (a=2, n=100) | O(n²·m) ≈ 718ms | 12,429 | $0.000004 |
| AUMA solve (a=3, n=100) | O(n³·m) ≈ 7s | 12,429 | $0.00004 |
| Verify isomorphism | O(n·m²) + 2·T_kissat | 2 × dimacs size | $0.000008 |
| Solution decode | O(n) | 0 | $0 |
The BCP provability gate as a clause generation criterion is novel. The tokenized uniform 3-SAT isomorphism (1 char = 1 token, polynomial bounds proven) is novel. The AUMA Fourier maximization on uniform topology (exploiting structural balance for signal clarity) is novel. No prior art combines polynomial NP reduction + tokenized representation + polynomial Fourier solve + verified solution set preservation.