Upload engine/predict_yearly.py with huggingface_hub
Browse files- engine/predict_yearly.py +91 -9
engine/predict_yearly.py
CHANGED
|
@@ -154,6 +154,11 @@ WRITING RULES (strict — these govern style, not your view):
|
|
| 154 |
- ARITHMETIC: every product, ratio or percentage you state must be correct to the figures you cite.
|
| 155 |
The market composite is a BLEND of several methods; it is never "EPS × one multiple". If you
|
| 156 |
want to cite a multiple, cite the forward P/E at the current price and say so.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 157 |
- Use ONLY these canonical names for the valuation anchors, exactly:
|
| 158 |
"story value" (the Damodaran story-driven intrinsic value), "mechanical DCF" (the formula-based
|
| 159 |
cross-check), "market composite" (the multiples + analyst-target pricing blend), and
|
|
@@ -4838,7 +4843,7 @@ EPS Forecast Detail:
|
|
| 4838 |
Anchor flag: {pred_fund.get('eps_anchor_flag', 'NO_ANCHOR')}
|
| 4839 |
Cyclical adjustment: {pred_fund.get('eps_cyclical_adjustment') or 'None'}
|
| 4840 |
EPS-FCF inconsistency: {pred_fund.get('eps_fcf_inconsistency') or 'None'}
|
| 4841 |
-
Historical beat rate:
|
| 4842 |
Avg surprise: {pred_fund.get('eps_avg_surprise_pct', 0):+.1f}%
|
| 4843 |
"""
|
| 4844 |
else:
|
|
@@ -4852,7 +4857,7 @@ EPS Forecast Detail:
|
|
| 4852 |
Anchor flag: {pred_fund.get('eps_anchor_flag', 'NO_ANCHOR')}
|
| 4853 |
Cyclical adjustment: {pred_fund.get('eps_cyclical_adjustment') or 'None'}
|
| 4854 |
EPS-FCF inconsistency: {pred_fund.get('eps_fcf_inconsistency') or 'None'}
|
| 4855 |
-
Historical beat rate:
|
| 4856 |
Avg surprise: {pred_fund.get('eps_avg_surprise_pct', 0):+.1f}%
|
| 4857 |
"""
|
| 4858 |
lines.append(eps_block.strip())
|
|
@@ -4881,7 +4886,7 @@ EPS Forecast Detail:
|
|
| 4881 |
avg_surp = pred_fund.get("eps_avg_surprise_pct")
|
| 4882 |
if beat_rate is not None and avg_surp is not None:
|
| 4883 |
beat_sym = "▲" if beat_rate >= 0.65 else "▼" if beat_rate <= 0.35 else "→"
|
| 4884 |
-
lines.append(f" EPS surprise history: beat rate {beat_rate*100:.0f}% {beat_sym} "
|
| 4885 |
f"avg surprise {avg_surp:+.1f}% (last 8 qtrs)")
|
| 4886 |
lines.append(f" NOTE: Predictions based on quarterly trend ML — treat as directional, not precise.")
|
| 4887 |
parts.append("\n".join(lines))
|
|
@@ -5041,7 +5046,7 @@ EPS Forecast Detail:
|
|
| 5041 |
parts.append(f"""SENTIMENT & INSIDER CONTEXT (qualitative — not in blend):
|
| 5042 |
News sentiment: {s.get('news_sentiment',0):+.4f} ({s.get('news_n_articles','?')} articles, trend:{mom_lbl})
|
| 5043 |
Insider score: {ins_str} ({s.get('insider_detail','no detail')})
|
| 5044 |
-
Put/Call ratio: {s.get('options_pc_ratio','N/A')} | Earnings beat rate: {s.get('earnings_score','N/A')}""")
|
| 5045 |
|
| 5046 |
# Detailed insider transactions
|
| 5047 |
if insider:
|
|
@@ -5205,7 +5210,12 @@ _PM_SYNTHESIS_USER_TEMPLATE_YEARLY = _STYLE_RULES + """STOCK: {ticker} | CURRENT
|
|
| 5205 |
|
| 5206 |
CROSS-SECTIONAL QUANT RANK (12-month panel-ML — Crucible's most BACKTESTED signal; weight it
|
| 5207 |
heavily as an independent prior, above either advocate's narrative): {panel_ml}
|
| 5208 |
-
VALUE vs PRICE
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5209 |
|
| 5210 |
NEUTRAL ANALYST (objective baseline — heard before the advocates):
|
| 5211 |
Return: {neutral_return:+.2f}% | Recommendation: {neutral_rec} | Confidence: {neutral_conf:.0f}%
|
|
@@ -5254,6 +5264,31 @@ Respond with ONLY this JSON:
|
|
| 5254 |
}}"""
|
| 5255 |
|
| 5256 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5257 |
def _emit_role(role, phase, entry, *, ret_key="expected_return"):
|
| 5258 |
"""Stream one analyst's FULL result (verdict + reasoning) to stdout the instant it
|
| 5259 |
resolves, as a machine-parseable sentinel line. The SSE wrapper turns this into a
|
|
@@ -5636,6 +5671,7 @@ def analyze_llm_debate_yearly(
|
|
| 5636 |
_PM_SYNTHESIS_USER_TEMPLATE_YEARLY.format(
|
| 5637 |
ticker=ticker, current_price=current_price, months=12,
|
| 5638 |
panel_ml=(_panel_ml_summary(ticker) or "not in the top-1000 universe (no rank)"),
|
|
|
|
| 5639 |
neutral_return=_nr_ret, neutral_rec=_nr_rec,
|
| 5640 |
neutral_conf=_nr_conf, neutral_reasoning=_nr_rsn,
|
| 5641 |
bear_return=_br["expected_return"] * 100,
|
|
@@ -7646,6 +7682,37 @@ def _restatement_from_actions(actions, statements_through):
|
|
| 7646 |
return None
|
| 7647 |
|
| 7648 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7649 |
def _nominate_peers(ticker, name, industry, description, call_json):
|
| 7650 |
"""Business-model comparables nominated by a model and later verified in the panel. Sector +
|
| 7651 |
market-cap matching put banks and insurers next to a ratings/index company (SPGI)."""
|
|
@@ -8075,6 +8142,15 @@ Examples:
|
|
| 8075 |
all_data["corp_actions"] = _corporate_actions_brief(ticker, _ovs.get("Name"), _events_call)
|
| 8076 |
_stq0 = (((all_data.get("income") or {}).get("quarterlyReports") or [{}])[0] or {}).get("fiscalDateEnding")
|
| 8077 |
all_data["restatement"] = _restatement_from_actions(all_data["corp_actions"], _stq0)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8078 |
all_data["pending_deals"] = [a for a in all_data["corp_actions"] if a.get("status") == "pending"]
|
| 8079 |
if all_data.get("corp_actions"):
|
| 8080 |
print(" [actions] " + " | ".join(f"{a['kind']} {a['business']} ({a['status']}{', ' + a['effective'] if a.get('effective') else ''})"
|
|
@@ -8240,15 +8316,21 @@ Examples:
|
|
| 8240 |
# don't set the consensus (GPT-first and Claude-first; Gemini as the shared fallback).
|
| 8241 |
def _story_call(system, user):
|
| 8242 |
out = _dp._call_frontier_json(user, system=system)
|
| 8243 |
-
if out is None: out = _dp._call_claude_json(system, user)
|
| 8244 |
if out is None: out = _dp._call_gemini_json(system, user)
|
| 8245 |
return out
|
| 8246 |
def _story_call2(system, user):
|
| 8247 |
-
out = _dp._call_claude_json(system, user)
|
| 8248 |
if out is None: out = _dp._call_gemini_json(system, user)
|
| 8249 |
if out is None: out = _dp._call_frontier_json(user, system=system)
|
| 8250 |
return out
|
| 8251 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8252 |
if not story_val_result:
|
| 8253 |
print(f" Damodaran story-FCFF: n/a — {getattr(_sd, 'LAST_REASON', None) or 'unknown reason'}")
|
| 8254 |
if story_val_result:
|
|
@@ -8608,7 +8690,7 @@ Examples:
|
|
| 8608 |
"sma50": safe_float(_ov.get("50DayMovingAverage")), "sma200": safe_float(_ov.get("200DayMovingAverage"))},
|
| 8609 |
"styles": (quant_ctx or {}).get("styles"), "quant_rank_pct": None,
|
| 8610 |
"summary": _rd.get("ai_summary"),
|
| 8611 |
-
"text_audit": (_rd.get("llm_debate") or {}).get("text_audit") or [],
|
| 8612 |
}
|
| 8613 |
try:
|
| 8614 |
_rd["diagnostics"]["quant_rank_pct"] = round(100 * (1 - (quant_ctx["comp_rank"] - 1) / quant_ctx["universe_n"]))
|
|
|
|
| 154 |
- ARITHMETIC: every product, ratio or percentage you state must be correct to the figures you cite.
|
| 155 |
The market composite is a BLEND of several methods; it is never "EPS × one multiple". If you
|
| 156 |
want to cite a multiple, cite the forward P/E at the current price and say so.
|
| 157 |
+
- ATTRIBUTION: the valuation is Crucible's own application of a published framework. Write "the
|
| 158 |
+
story value" or "the Damodaran-style value" — never "Damodaran's value/DCF/target", which implies
|
| 159 |
+
he produced or endorsed the number.
|
| 160 |
+
- The intrinsic-value gap you cite is the STORY VALUE vs price. The mechanical DCF is a cross-check:
|
| 161 |
+
name it as such and never present its gap as "the intrinsic gap" or "the premium to fair value".
|
| 162 |
- Use ONLY these canonical names for the valuation anchors, exactly:
|
| 163 |
"story value" (the Damodaran story-driven intrinsic value), "mechanical DCF" (the formula-based
|
| 164 |
cross-check), "market composite" (the multiples + analyst-target pricing blend), and
|
|
|
|
| 4843 |
Anchor flag: {pred_fund.get('eps_anchor_flag', 'NO_ANCHOR')}
|
| 4844 |
Cyclical adjustment: {pred_fund.get('eps_cyclical_adjustment') or 'None'}
|
| 4845 |
EPS-FCF inconsistency: {pred_fund.get('eps_fcf_inconsistency') or 'None'}
|
| 4846 |
+
Historical beat rate (last 8 quarters): {pred_fund.get('eps_beat_rate', 0.5)*100:.0f}%
|
| 4847 |
Avg surprise: {pred_fund.get('eps_avg_surprise_pct', 0):+.1f}%
|
| 4848 |
"""
|
| 4849 |
else:
|
|
|
|
| 4857 |
Anchor flag: {pred_fund.get('eps_anchor_flag', 'NO_ANCHOR')}
|
| 4858 |
Cyclical adjustment: {pred_fund.get('eps_cyclical_adjustment') or 'None'}
|
| 4859 |
EPS-FCF inconsistency: {pred_fund.get('eps_fcf_inconsistency') or 'None'}
|
| 4860 |
+
Historical beat rate (last 8 quarters): {pred_fund.get('eps_beat_rate', 0.5)*100:.0f}%
|
| 4861 |
Avg surprise: {pred_fund.get('eps_avg_surprise_pct', 0):+.1f}%
|
| 4862 |
"""
|
| 4863 |
lines.append(eps_block.strip())
|
|
|
|
| 4886 |
avg_surp = pred_fund.get("eps_avg_surprise_pct")
|
| 4887 |
if beat_rate is not None and avg_surp is not None:
|
| 4888 |
beat_sym = "▲" if beat_rate >= 0.65 else "▼" if beat_rate <= 0.35 else "→"
|
| 4889 |
+
lines.append(f" EPS surprise history (last 8 quarters): beat rate {beat_rate*100:.0f}% {beat_sym} "
|
| 4890 |
f"avg surprise {avg_surp:+.1f}% (last 8 qtrs)")
|
| 4891 |
lines.append(f" NOTE: Predictions based on quarterly trend ML — treat as directional, not precise.")
|
| 4892 |
parts.append("\n".join(lines))
|
|
|
|
| 5046 |
parts.append(f"""SENTIMENT & INSIDER CONTEXT (qualitative — not in blend):
|
| 5047 |
News sentiment: {s.get('news_sentiment',0):+.4f} ({s.get('news_n_articles','?')} articles, trend:{mom_lbl})
|
| 5048 |
Insider score: {ins_str} ({s.get('insider_detail','no detail')})
|
| 5049 |
+
Put/Call ratio: {s.get('options_pc_ratio','N/A')} | Earnings-quality score (0-1 composite, NOT a beat rate): {s.get('earnings_score','N/A')}""")
|
| 5050 |
|
| 5051 |
# Detailed insider transactions
|
| 5052 |
if insider:
|
|
|
|
| 5210 |
|
| 5211 |
CROSS-SECTIONAL QUANT RANK (12-month panel-ML — Crucible's most BACKTESTED signal; weight it
|
| 5212 |
heavily as an independent prior, above either advocate's narrative): {panel_ml}
|
| 5213 |
+
VALUE vs PRICE — canonical anchors (use these numbers and names; the advocates may have mislabelled them):
|
| 5214 |
+
{anchors}
|
| 5215 |
+
The STORY VALUE is the intrinsic value. The mechanical DCF is only a formula cross-check of it — never quote the
|
| 5216 |
+
mechanical-DCF gap as "the intrinsic gap" or "the premium to fair value". The market composite is PRICING (what the
|
| 5217 |
+
market's own yardsticks imply), not value. Treat the story-value-vs-price gap as a first-class input — your synthesis
|
| 5218 |
+
MUST say explicitly whether you expect that gap to close toward value, toward price, or persist, and why.
|
| 5219 |
|
| 5220 |
NEUTRAL ANALYST (objective baseline — heard before the advocates):
|
| 5221 |
Return: {neutral_return:+.2f}% | Recommendation: {neutral_rec} | Confidence: {neutral_conf:.0f}%
|
|
|
|
| 5264 |
}}"""
|
| 5265 |
|
| 5266 |
|
| 5267 |
+
def _pm_anchor_lines(current_price, fair_val, pred_fund, data):
|
| 5268 |
+
"""Canonical valuation anchors for the PM (and the text audit): the PM previously took the
|
| 5269 |
+
bear's framing ('45% above mechanical DCF') as the intrinsic gap when the story value sat 8%
|
| 5270 |
+
below price (SPGI)."""
|
| 5271 |
+
fv = fair_val or {}
|
| 5272 |
+
story = fv.get("story") or globals().get("_CRUCIBLE_STORY") or {}
|
| 5273 |
+
out = [f" • current price ${current_price:,.2f}"]
|
| 5274 |
+
if story.get("value"):
|
| 5275 |
+
out.append(f" • story value (INTRINSIC, Damodaran-style story-driven FCFF): ${story['value']:,.2f} ({(story['value']/current_price-1)*100:+.1f}% vs price)")
|
| 5276 |
+
_dcf = ((fv.get("methods") or {}).get("DCF") or {}).get("price")
|
| 5277 |
+
if _dcf:
|
| 5278 |
+
out.append(f" • mechanical DCF (formula cross-check only): ${_dcf:,.2f} ({(_dcf/current_price-1)*100:+.1f}%)")
|
| 5279 |
+
if fv.get("fair_price"):
|
| 5280 |
+
out.append(f" • market composite (pricing: multiples + analyst targets): ${fv['fair_price']:,.2f} ({(fv['fair_price']/current_price-1)*100:+.1f}%)")
|
| 5281 |
+
_eps = (pred_fund or {}).get("eps_annual")
|
| 5282 |
+
if _eps:
|
| 5283 |
+
out.append(f" • engine forward EPS ${_eps:,.2f} → forward P/E {current_price/_eps:.1f}x at the current price")
|
| 5284 |
+
_rst = (data or {}).get("restatement")
|
| 5285 |
+
if _rst:
|
| 5286 |
+
out.append(f" • RESTATEMENT: {_rst['note']}")
|
| 5287 |
+
for a in (data or {}).get("pending_deals") or []:
|
| 5288 |
+
out.append(f" • PENDING (not closed): {a['kind'].replace('_', ' ')} of {a['business']}")
|
| 5289 |
+
return "\n".join(out)
|
| 5290 |
+
|
| 5291 |
+
|
| 5292 |
def _emit_role(role, phase, entry, *, ret_key="expected_return"):
|
| 5293 |
"""Stream one analyst's FULL result (verdict + reasoning) to stdout the instant it
|
| 5294 |
resolves, as a machine-parseable sentinel line. The SSE wrapper turns this into a
|
|
|
|
| 5671 |
_PM_SYNTHESIS_USER_TEMPLATE_YEARLY.format(
|
| 5672 |
ticker=ticker, current_price=current_price, months=12,
|
| 5673 |
panel_ml=(_panel_ml_summary(ticker) or "not in the top-1000 universe (no rank)"),
|
| 5674 |
+
anchors=_pm_anchor_lines(current_price, fair_val, pred_fund, data),
|
| 5675 |
neutral_return=_nr_ret, neutral_rec=_nr_rec,
|
| 5676 |
neutral_conf=_nr_conf, neutral_reasoning=_nr_rsn,
|
| 5677 |
bear_return=_br["expected_return"] * 100,
|
|
|
|
| 7682 |
return None
|
| 7683 |
|
| 7684 |
|
| 7685 |
+
def _restatement_from_consensus(eec, rev_ttm, rev_growth_yoy, actions, statements_through):
|
| 7686 |
+
"""Consensus-vs-TTM restatement signal (see call site). Requires >=8 analysts, current-FY
|
| 7687 |
+
consensus revenue <95% of reported TTM, a next-FY consensus that is ALSO below TTM, and
|
| 7688 |
+
positive reported growth (a shrinking company legitimately has consensus below TTM)."""
|
| 7689 |
+
try:
|
| 7690 |
+
cur, nxt = eec.get("current_year_revenue"), eec.get("next_year_revenue")
|
| 7691 |
+
n = int(eec.get("analyst_count") or 0)
|
| 7692 |
+
g = rev_growth_yoy if rev_growth_yoy is not None else None
|
| 7693 |
+
if g is not None and abs(g) > 1.0: g = g / 100.0
|
| 7694 |
+
if not (cur and rev_ttm and rev_ttm > 0 and n >= 8):
|
| 7695 |
+
return None
|
| 7696 |
+
if not (cur < 0.95 * rev_ttm and (nxt is None or nxt < rev_ttm) and (g is None or g > 0.02)):
|
| 7697 |
+
return None
|
| 7698 |
+
unit = None
|
| 7699 |
+
for a in sorted(actions or [], key=lambda x: -(x.get("revenue_share") or 0)):
|
| 7700 |
+
if a.get("kind") in ("spin_off", "divestiture"):
|
| 7701 |
+
unit = a; break
|
| 7702 |
+
biz = unit["business"] if unit else "a divested business"
|
| 7703 |
+
gap = 1.0 - cur / rev_ttm
|
| 7704 |
+
return {"kind": (unit or {}).get("kind") or "divestiture", "business": biz, "effective": (unit or {}).get("effective"),
|
| 7705 |
+
"revenue_share": (unit or {}).get("revenue_share") or round(gap, 2), "statements_through": str(statements_through)[:10],
|
| 7706 |
+
"basis": "consensus", "consensus_current_fy_usd_b": round(cur / 1e9, 2), "reported_ttm_usd_b": round(rev_ttm / 1e9, 2),
|
| 7707 |
+
"note": (f"Sell-side consensus revenue for the current fiscal year (${cur/1e9:,.1f}B, {n} analysts) is "
|
| 7708 |
+
f"{gap*100:.0f}% below reported trailing-twelve-month revenue (${rev_ttm/1e9:,.1f}B) although reported revenue "
|
| 7709 |
+
f"grew {g*100:.0f}% year over year — analysts are modelling a smaller continuing company, consistent with "
|
| 7710 |
+
f"the {(unit or {}).get('kind','divestiture').replace('_',' ')} of {biz} having completed; reported TTM revenue, "
|
| 7711 |
+
f"margins and multiples still include it.")}
|
| 7712 |
+
except Exception: # noqa: BLE001
|
| 7713 |
+
return None
|
| 7714 |
+
|
| 7715 |
+
|
| 7716 |
def _nominate_peers(ticker, name, industry, description, call_json):
|
| 7717 |
"""Business-model comparables nominated by a model and later verified in the panel. Sector +
|
| 7718 |
market-cap matching put banks and insurers next to a ratings/index company (SPGI)."""
|
|
|
|
| 8142 |
all_data["corp_actions"] = _corporate_actions_brief(ticker, _ovs.get("Name"), _events_call)
|
| 8143 |
_stq0 = (((all_data.get("income") or {}).get("quarterlyReports") or [{}])[0] or {}).get("fiscalDateEnding")
|
| 8144 |
all_data["restatement"] = _restatement_from_actions(all_data["corp_actions"], _stq0)
|
| 8145 |
+
if not all_data["restatement"]:
|
| 8146 |
+
# Data-driven fallback: model recall lags reality (it still had S&P Global's
|
| 8147 |
+
# July-2026 Mobility spin as "pending"). If sell-side consensus revenue for the
|
| 8148 |
+
# CURRENT fiscal year sits well below reported TTM revenue while the company is
|
| 8149 |
+
# growing, analysts are modelling a smaller continuing company — a completed
|
| 8150 |
+
# spin-off/divestiture is the only ordinary explanation.
|
| 8151 |
+
all_data["restatement"] = _restatement_from_consensus(
|
| 8152 |
+
all_data.get("earnings_estimates_consensus") or {}, safe_float(_ovs.get("RevenueTTM")),
|
| 8153 |
+
safe_float(_ovs.get("QuarterlyRevenueGrowthYOY")), all_data["corp_actions"], _stq0)
|
| 8154 |
all_data["pending_deals"] = [a for a in all_data["corp_actions"] if a.get("status") == "pending"]
|
| 8155 |
if all_data.get("corp_actions"):
|
| 8156 |
print(" [actions] " + " | ".join(f"{a['kind']} {a['business']} ({a['status']}{', ' + a['effective'] if a.get('effective') else ''})"
|
|
|
|
| 8316 |
# don't set the consensus (GPT-first and Claude-first; Gemini as the shared fallback).
|
| 8317 |
def _story_call(system, user):
|
| 8318 |
out = _dp._call_frontier_json(user, system=system)
|
| 8319 |
+
if out is None: out = _dp._call_claude_json(system, user, temperature=0.3)
|
| 8320 |
if out is None: out = _dp._call_gemini_json(system, user)
|
| 8321 |
return out
|
| 8322 |
def _story_call2(system, user):
|
| 8323 |
+
out = _dp._call_claude_json(system, user, temperature=0.3)
|
| 8324 |
if out is None: out = _dp._call_gemini_json(system, user)
|
| 8325 |
if out is None: out = _dp._call_frontier_json(user, system=system)
|
| 8326 |
return out
|
| 8327 |
+
# Third, Gemini-led storyteller: the consensus is the MEDIAN of three sets of
|
| 8328 |
+
# assumptions, so a single drifting story can't move the value (run-to-run stability).
|
| 8329 |
+
def _story_call3(system, user):
|
| 8330 |
+
out = _dp._call_gemini_json(system, user)
|
| 8331 |
+
if out is None: out = _dp._call_claude_json(system, user, temperature=0.3)
|
| 8332 |
+
return out
|
| 8333 |
+
story_val_result = _sd.story_value(_ctx, _story_call, _story_call2, _story_call3, verbose=args.verbose)
|
| 8334 |
if not story_val_result:
|
| 8335 |
print(f" Damodaran story-FCFF: n/a — {getattr(_sd, 'LAST_REASON', None) or 'unknown reason'}")
|
| 8336 |
if story_val_result:
|
|
|
|
| 8690 |
"sma50": safe_float(_ov.get("50DayMovingAverage")), "sma200": safe_float(_ov.get("200DayMovingAverage"))},
|
| 8691 |
"styles": (quant_ctx or {}).get("styles"), "quant_rank_pct": None,
|
| 8692 |
"summary": _rd.get("ai_summary"),
|
| 8693 |
+
"text_audit": ((llm_result or {}).get("text_audit") if isinstance(llm_result, dict) else None) or (_rd.get("llm_debate") or {}).get("text_audit") or [],
|
| 8694 |
}
|
| 8695 |
try:
|
| 8696 |
_rd["diagnostics"]["quant_rank_pct"] = round(100 * (1 - (quant_ctx["comp_rank"] - 1) / quant_ctx["universe_n"]))
|