Purpose
The workflow supports the paired NLPCC research item by preserving evidence for comparisons and daily decisions.
Research need
Allocation results are difficult to interpret without frozen evaluation, simple baselines, and traceable decision records.
Architecture
The recorded workflow connects baseline runs, held-out evaluation, trace records, and portfolio diagnostics.
Inputs
Candidate allocation outputs and evaluation-period records are described only at a class level. No private data source is named.
Implementation decisions
Evidence is organised around comparisons rather than a single score: no-news and no-LLM baselines, data-visibility checks, and risk or turnover diagnostics are part of the stated protocol.
Constraints
The current site does not contain the pipeline repository or execution logs, so the page avoids claiming official parity, completed test coverage, or production readiness.
Outputs
Inspectable trace records and comparable evaluation summaries are the intended public-safe outputs.
Limitations
Traceability can expose a decision path but cannot by itself establish causal validity or robustness.
Status
Research build associated with the ongoing agent study.