Research master and release evidence
Keep full research completion, software correctness, cloud deployment and subscriber certification distinct.
Complete research is more than a candidate grid
An actual instrument research request requires the complete authoritative factory and independently verified master unless the owner explicitly asks for a narrower diagnostic. Adapter preparation, a compact grid, profit-factor screening or component fixtures cannot be substituted for that end state. Asset-class routing and identity guards remain essential. Compatible verified artifacts can be reused without weakening periods, warmup, native exits or resource limits. The website teaches these boundaries but performs no actual instrument research, training or calibration. Its fictional model cards are not new certified production candidates.
A verified no-model master can be complete
A complete independent process can validly conclude that no model meets its requirements. That is different from missing data, unsupported calibration, an unimplemented master or untested candidates. Do not interpret an intermediate zero as a full-market failure. Record the exact preserved stage and blocker when genuine completion is unavailable. Software tests with authored expected answers establish code behavior, not independent strategy samples. The lens's reference grid intentionally contains known loss and count results. Those checks cannot be counted as financial evidence for any model or any market.
Entry and exit ownership stay explicit
Rank entries on the raw causal boundary-recovery objective and freeze entry identity before applying the experiment's predeclared exit. Subsequent outcome statistics can validate, reject, size or measure overlap without rewriting the original ranking basis. Do not reopen an exit tournament after reserved evidence is accessed. Preserve native minute execution, whole-position and maximum-hold rules where they apply, and the established best hypothetical research-fill observation window. The site's binary examples contain no native exit or authentic fill and cannot authorize changing any of those real research conventions.
A live route has its own artifact requirements
Research state, code merge, registry and gate state, deployed cloud identity, subscriber route and certification are separate. Authenticate the actual cloud artifact and read back the served version. Do not infer deployment from a source filename, endpoint inventory or research LIVE label. Production must remain functional with Codex closed and personal computers off. Research completion does not authorize production promotion or additional paid capacity. A static educational-site deployment is narrower still: it confirms the publication is served, not that an independently validated financial model reaches a subscriber.
Fictional model evidence case
The authored release card lists five absent records: actual independent master, merged source identity, registry admission, deployed source identity and subscriber-route evidence. A passing count-formula fixture fills none of them. A hypothetical cloud artifact would answer only its deployment object and would still need a readback and route. The correct teaching state remains specification complete, actual financial research not established. No stage name is silently upgraded to another.
| Object | Record | Bounded finding |
|---|---|---|
| Software | Known-answer fixture | Correctness only |
| Research | Independent master absent | Incomplete real-evidence chain |
| Cloud | Artifact identity absent | No deployed-model claim |
| Subscriber | Route and certification absent | No receiving proof |

Primary references and scope
Source review: 14 October 2026. Official mechanism descriptions are not this desk's paid-account test, independently captured fills or vendor-return audit.