Patch bool coercion bug in conditional expectancy matrix enrichment

[kvandre12-commits]

Root cause identified in scripts/build_conditional_expectancy_matrix.py.

The pipeline is failing during enrich_metrics() because pandas logical OR operations are being applied between float64 columns (created via NaN/SQLite coercion) and boolean Series.

Current failing logic:

out["_direct_fill"] = (
    safe_num(out, ["direct_fill"], 0).fillna(0)
    | out["fill_path_type"].astype(str).str.upper().eq("DIRECT_FILL")
).astype(int)

and:

out["_failed_fill"] = (
    safe_num(out, ["failed_fill"], 0).fillna(0)
    | out["fill_path_type"].astype(str).str.upper().isin([
        "FAILED_FILL_CONTINUATION",
        "PARTIAL_FILL_REJECT",
        "LIQUIDITY_VACUUM_CONTINUATION",
    ])
).astype(int)

This produces:

TypeError: unsupported operand type(s) for | : 'float' and 'bool'

Recommended fix:

Add helper:

def safe_bool(df, candidates, default=False):
    return (
        safe_num(df, candidates, int(default))
        .fillna(0)
        .astype(bool)
    )

Then replace failing sections with:

out["_fill"] = safe_bool(
    out,
    ["filled", "gap_filled", "fill_success"]
).astype(int)

out["_direct_fill"] = (
    safe_bool(out, ["direct_fill"])
    | out["fill_path_type"]
        .astype(str)
        .str.upper()
        .eq("DIRECT_FILL")
).astype(int)

out["_failed_fill"] = (
    safe_bool(out, ["failed_fill"])
    | out["fill_path_type"]
        .astype(str)
        .str.upper()
        .isin([
            "FAILED_FILL_CONTINUATION",
            "PARTIAL_FILL_REJECT",
            "LIQUIDITY_VACUUM_CONTINUATION",
        ])
).astype(int)

Expected result:

• classify auction fill paths succeeds
• conditional_expectancy_matrix populates
• SharpEdge 2.0 report loads matrix successfully
• downstream report publishing restored

This is schema/dtype normalization rather than orchestration failure. The merged YAML workflow structure is now functioning correctly.