A model score, an executable order and realized PnL are not interchangeable observations. ELARVEN's Strategy Decay work began by asking whether the existing research archives were complete enough to trace that chain.
The intended waterfall
The target schema was:
What the archive contained
A data-readiness scan found candidate evidence across both the ELARVEN execution-intelligence work and older DRAGON archives. The first readiness pass located 77 ELARVEN candidates and 1,918 DRAGON candidates across the searched schema categories.
What was actually structured
| Layer | Evidence status |
|---|---|
| Live decision | Structured schema match found |
| Executable entry | Partial structured match |
| Actual fill | No usable structured match |
| Spread cost | Structured schema match found |
| Bid/ask asymmetry | Partial structured match |
| Rejected / missed | Partial structured match |
| Exit / realized PnL | Partial structured match |
| Risk / account context | Structured schema match found |
The resulting mapping verdict was PARTIAL_CORE_STRUCTURED_EVIDENCE__GAPS_REMAIN.
Why the missing fill matters
Without a reliable actual-fill layer, it is difficult to attribute decay precisely. A model can be correct at decision time and still lose value through entry delay, spread, partial execution or a changed exit path. If those states are merged, model error and execution error become indistinguishable.
Research consequence
The work therefore stopped short of claiming a complete execution waterfall. The useful result was the schema itself and a precise description of which instrumentation must exist before strategy decay can be measured without guessing.
Limitations
- The scan identifies structured evidence availability, not causal execution decay by itself.
- Candidate counts are search/mapping counts and should not be interpreted as trade counts.
- A complete future implementation requires systematic live fill capture and consistent order identifiers across the chain.