Certificate adversarial corpus
28 canonical positive and negative cases, including malformed packages, revoked keys, stale evidence, replay and valid signatures over invalid semantics.
Evidence ceiling E3Proof boundary
ENGINEERING ASSURANCE
Challenge the ending.
The assurance program concentrates on whether hostile evidence, replay and interrupted finalization can produce an unsupported FINAL.
Current reproducible work
Seller-controlled evidence. Precisely bounded.
These campaigns run in the build gate. They do not constitute independent validation or customer production evidence.
Delivery schedules
720 reordered and duplicate delivery schedules preserve the decision and evidence commitment for the defined fixture.
Durable fault lab
152 SQLite reference experiments cover competing CAS proposals, transaction interruption, process termination, recovery and idempotent retries. This is not production D1 or real network-partition validation.
Bounded decision campaign
V121 established agreement across 4,800 finite cases under the declared seller-controlled profile. That does not establish estate-wide zero false FINAL.
Release controls
Corpus disagreements, skipped scoped checks, timeouts and surviving selected mutations block the scoped build.
External assurance
Independent-team implementations, buyer-authored executed cases and real customer production outcome evidence remain NOT ESTABLISHED.
Independent implementation
A separate implementation must be independently authored.
An outside team receives the public profile and canonical vectors. It selects its language, retains authorship and source/build provenance, and consumes the same corpus. A seller-coded second language is not counted as independent external validation.
Buyer-controlled falsification
The buyer holds the truth and unseen cases.
The local challenge runner passes case inputs to a verifier without passing the buyer's private oracle file. The buyer controls authorship, authoritative reality and truth custody. The runner does not create a process sandbox or certify those facts.
Private challenge runner
A local runner for language-neutral verifier adapters. Keep oracle files and initial results on buyer-controlled infrastructure.
Explore →Institutional evaluation
SCOPE → Formal SCOPE → a buyer-controlled institutional evaluation with agreed workflow and success criteria.
Explore →Production evidence
Measure real outcomes when they exist.
Future reports should distinguish evaluated consequential outcomes, contradiction and REOPEN events, independently authored attacks, and known false FINAL. Every count must carry its source, observation period, authority, verification boundary and signed external execution evidence.
Consequential customer outcomes
NOT ESTABLISHED in this assurance program.
Externally authored executed cases
NOT ESTABLISHED in this assurance program.
Real production contradiction / REOPEN history
NOT ESTABLISHED in this assurance program.