B64→B72 cross-word backward capability
The target ladder can cross from B64 into digest word 5 by reconstructing the exact e63/feedforward relation and evaluating the next target byte backward from the final digest.
Diagnostic work, exact algebra and local capabilities are not treated as end-to-end mining advantage.
What was tested?
The target ladder can cross from B64 into digest word 5 by reconstructing the exact e63/feedforward relation and evaluating the next target byte backward from the final digest.
Why the test is meaningful
A byte boundary crossing changes which final state word supplies target bits. SHA-256 state shifting gives a deterministic relation between late-round e63 and digest word 5; preserving that identity is necessary for a safe incremental decision.
B72 = B64 ∧ next8(word5)digest_word5 = e63 + IV5 (mod 2³²)incremental decision = first differing bit of word5 byte0How it was tested
Condition on exact B64 records, reconstruct e63 and digest word 5, evaluate the next byte in lazy bit order and repeat two independent 20,000-record campaigns against the full suffix.
What happened
Both replicates reported zero B64-condition, e63-relation and full-suffix errors. Mean incremental B72 work was 45.799 and 45.938 units, with median 46 in both. Most rejections occurred in the first few tested bits.
Exactness and statistical controls
Two independent replicates exercised the cross-word identity and every final target decision. The contract explicitly classifies this as a capability, not a mining submotor.
What the result means
V49 validates the second cross-word bridge. Exact backward comparison can continue into word 5, but the measured work has not been shown cheaper than a native complete implementation.
Limitations
- The experiment begins from B64-conditioned records.
- Diagnostic work units are not ASIC cycles.
- The capability tests target compatibility; it does not generate candidates.
Evidence trail
Recompute e63/feedforward from each anchored state, verify B64, test the next eight bits lazily and compare with the complete 256-bit target decision in both replicates.
Canonical variants
SUBENGINE-V49ASUBENGINE-V49BSUBENGINE-V49CSource: internally audited canonical reports. Local filesystem structure, private headers and operational identifiers are excluded from publication.