Improvement Law

The CoreTex 1.1.0 law retains fixed-suite, exact-parent dominance admission. The public wire path is /coretex/v5.

For each profile the public suite has two disjoint partitions, gate and confirm. Both partitions are fixed for suite block 0 and both must admit the candidate. Rig ID, wallet, epoch, retry count, entropy, and queue position do not choose different benchmark cases.

Each partition records:

• Q: composite score plus the declared objective scores, where higher is better;
• R: measured rendered cost, trusted work fuel, and logical durable storage, where lower is better; and
• E: the serialized resource envelope, defined in this law as the fixed product cap C.

The fixed caps are law constants for suite block 0. Parent E, candidate E, and the canonical suite cap must all equal C. Efficiency wins do not shrink that cap, and quality wins do not grow it. If a future launch changes the cap table, that is a prospective new law/release, not an in-place reinterpretation of accepted history.

The 1.1.0 block-0 product caps are listed below for orientation. The release-bound canonical suite is the numeric authority; obtain current values from the live parent-vector response before mining:

RESOURCE_ENVELOPE.json in the kit is a larger hard-gate submission ceiling, not the admission cap C.

The trusted-work policy is final-render-trusted-hostwork.v5; storage uses logical-durable-storage.cbor.v1+vectors.v1, including logical dense-vector bytes. Encoder work dominates conversation fuel. Reducing permitted host calls and packing work can therefore produce resource savings; hook-only CPU fuel remains diagnostic. Larger raw fuel totals do not multiply rewards across profiles.

Admission is evaluated independently on gate and confirm:

1. hard gates and the profile composite floor pass;
2. every objective stays within the bounded-trade rule: no objective may fall more than 2.5 points from the exact parent or more than 2.5 points below the genesis floor, and total objective gains must be at least twice total objective dips;
3. every protected resource remains at or below fixed cap C;
4. progress is either quality or efficiency: - quality: composite improves by at least one ppm; resources may rise relative to the exact parent, but only inside C; - efficiency: composite is held under the law's rounding rule, every raw resource is no higher than the exact parent, and at least one resource is strictly lower; and
5. if both classes hold, the transition is classified as quality.

This is intentionally not a resource ratchet. A cost or fuel saving remains an accepted efficiency gain, but it does not permanently reduce the product cap for later quality work. Conversely, a same-quality successor cannot give that efficiency win back. A quality-improving successor may spend saved capacity inside C.

Logical durable storage has one practical caveat: the meter is add-only over candidate artifacts, so a storage-only efficiency win is not available from the genesis reference. Storage is a spend-then-reclaim axis: a quality candidate may spend storage inside C, and a later efficiency candidate may reclaim that spend back toward the exact parent or genesis usage.

The public miner kit can preview structure, parent binding, and resource positioning. The final admission decision comes from the sealed production evaluator and is replayable from the published chain/CAS evidence.