dashboard.py

"""Phase 6 - Streamlit dashboard for the SFDR 2.0 classification engine.

Run:  streamlit run dashboard.py

The dashboard reads the FROZEN engine (Phases 2-5) -- no logic lives here.
Three interactive controls wire to the levers identified in Phase 5:
  - PAB-replication toggle  (the dominant structural lever)
  - Year slider             (PAB pathway ratchets -7% p.a.)
  - Taxonomy multiplier     (the deeming-provision tipping point)
"""
import copy
import streamlit as st
import pandas as pd

from sfdr2.data import SAMPLE_PORTFOLIO
from sfdr2.classify import classify
from sfdr2 import sensitivity as S

st.set_page_config(page_title="SFDR 2.0 Classification Engine", layout="wide")

st.title("SFDR 2.0 Portfolio Classification Engine")
st.caption(
    "Interactive implementation of the EU SFDR 2.0 product-category logic "
    "(Sustainable / Transition / ESG Basics) with Paris-Aligned Benchmark "
    "exclusions and the 70% threshold."
)

# ---------------------------------------------------------------------------
# Sidebar -- the three levers from Phase 5
# ---------------------------------------------------------------------------
st.sidebar.header("Product structure")
pab_repl = st.sidebar.checkbox(
    "Product replicates a PAB benchmark",
    value=False,
    help="Activates the PAB-replication deeming provision. Sustainable / "
         "Transition categories are deemed to meet the 70% floor.",
)
year = st.sidebar.slider(
    "Reporting year",
    min_value=2024, max_value=2035, value=2026, step=1,
    help="PAB pathway: 50% intensity cut vs parent, then 7% p.a. self-decarbonisation.",
)

st.sidebar.header("Data assumption")
tax_mult = st.sidebar.slider(
    "Taxonomy-aligned revenue multiplier",
    min_value=0.5, max_value=3.0, value=1.0, step=0.05,
    help="Scales every issuer's Taxonomy-aligned revenue. Use to test the "
         "tipping point of the 15% deeming provision.",
)

# Apply the Taxonomy multiplier to a copy of the portfolio
H = copy.deepcopy(SAMPLE_PORTFOLIO)
for h in H:
    if h.get("taxonomy_aligned_rev_pct") is not None:
        h["taxonomy_aligned_rev_pct"] *= tax_mult

# ---------------------------------------------------------------------------
# Headline classification
# ---------------------------------------------------------------------------
result = classify(H, pab_replication=pab_repl, year=year)

c1, c2, c3 = st.columns([1.2, 1, 1])
with c1:
    color = {"Sustainable": "#1f7a3c", "Transition": "#c47d00",
             "ESG Basics": "#3a5a8c", "Non-classifiable": "#a02020"}.get(result.label, "#444")
    st.markdown(
        f"<div style='padding:18px;border-radius:8px;background:{color};color:white'>"
        f"<div style='font-size:13px;opacity:0.85'>Classification</div>"
        f"<div style='font-size:32px;font-weight:600'>{result.label}</div>"
        f"<div style='font-size:13px;opacity:0.9;margin-top:4px'>"
        f"via {result.carrying_path}</div></div>",
        unsafe_allow_html=True,
    )
with c2:
    st.metric("PAB-replication", "Yes" if pab_repl else "No")
    st.metric("Pathway aligned", "Yes" if result.pathway_aligned else "No")
with c3:
    st.metric("Reporting year", year)
    st.metric("Taxonomy multiplier", f"{tax_mult:.2f}x")

if result.pathway_warning:
    st.warning(result.pathway_warning)

st.divider()

# ---------------------------------------------------------------------------
# Decision-tree audit trail
# ---------------------------------------------------------------------------
st.subheader("Decision-tree audit trail")
st.caption("Each node is one regulatory test. The 'decisive' marker shows the "
           "node that carried the classification.")
audit_rows = [
    {"#": i + 1, "Step": n.step, "Test": n.test, "Value": n.value,
     "vs": n.threshold, "Pass": "Yes" if n.passed else "No",
     "Decisive": "<-- decisive" if n.decisive else ""}
    for i, n in enumerate(result.audit)
]
st.dataframe(pd.DataFrame(audit_rows), use_container_width=True, hide_index=True)

# ---------------------------------------------------------------------------
# Issuer-level screening (Phase 2 detail)
# ---------------------------------------------------------------------------
st.subheader("Issuer-level PAB screening")
st.caption("Synthetic sample. None = data gap (conservative: never qualifying, "
           "but still in the denominator).")
df = pd.DataFrame(H)
display_cols = ["issuer", "sector", "weight_pct",
                "coal_rev_pct", "oil_rev_pct", "gas_rev_pct",
                "power_gen_rev_pct", "power_carbon_intensity_gco2_kwh",
                "scope12_intensity_tco2_per_eur_m",
                "taxonomy_aligned_rev_pct", "taxonomy_aligned_capex_pct",
                "controversial_weapons", "tobacco_producer", "ungc_oecd_violation"]
st.dataframe(df[display_cols], use_container_width=True, hide_index=True)

# ---------------------------------------------------------------------------
# Phase 5 -- cost-of-compliance numbers
# ---------------------------------------------------------------------------
st.divider()
st.subheader("Cost-of-compliance diagnostics")

col_a, col_b = st.columns(2)
with col_a:
    st.markdown("**Exclusion-gate attribution**")
    ea = S.exclusion_attribution(H)
    attr_rows = [{"Screen": k, "Weight lost (%)": round(v["weight"], 1),
                  "Names": ", ".join(v["names"])}
                 for k, v in sorted(ea["by_screen"].items(),
                                    key=lambda kv: -kv[1]["weight"])]
    st.dataframe(pd.DataFrame(attr_rows), use_container_width=True, hide_index=True)
    st.caption(f"Total excluded weight: {ea['total_excluded_weight_pct']:.1f}%  |  "
               f"Data gap: {ea['gap_weight_pct']:.1f}%")

with col_b:
    st.markdown("**Decarbonisation residual**")
    dr = S.decarb_residual(H)
    st.metric("Post-exclusion WACI (tCO2e/EURm)", dr["post_excl_waci"])
    st.metric("1.5C pathway target",
              dr["pathway_target"],
              delta=f"{dr['absolute_gap']:+.1f} gap",
              delta_color="inverse")
    st.caption(f"Further intensity cut required: **{dr['further_cut_required_pct']:.1f}%** "
               "(screening is not decarbonisation)")