GOOG/GOOGL multi-class shares overcount — opposite direction to PR #257 allowlist (Alphabet aggregate-only filer)

[dackclup]

Symptom

On the 2026-05-23 production cron (90157482, commit 9015748), both GOOG (Class C) and GOOGL (Class A) tickers store Alphabet's 12.12B total shares (sum across all classes) instead of the per-class breakdown — making each ticker appear to have a $4.6T market cap.

Ticker	Stored shares_outstanding	Stored market_cap	Real (per-class)	Overcount
GOOG (Class C)	12,116,000,000	$4,596.6B	~$1,048B (5.9B shares × ~$178)	4.4×
GOOGL (Class A)	12,116,000,000	$4,640.1B	~$1,057B (5.9B shares × ~$179)	4.4×
Combined	(storing 2 × 12.12B = 24.24B)	$9,236B	~$2,105B (real Alphabet MC)	4.4×

fair_price.median = $85.29 against actual price ~$379 → mos_pct = -344.8% / -349.0%. Value pillar contaminated (15.94 / 11.59 — appears expensive on every ratio because each per-class ticker is reported at ~$4.6T).

Root cause

compute/ingest/fundamentals.py reads shares_outstanding from SEC companyfacts for Alphabet's CIK (0001652044). The companyfacts aggregate API returns the total common-shares outstanding across all classes (~12.12B) — Alphabet files non-dimensionally for the EntityCommonStockSharesOutstanding concept, so both GOOG and GOOGL tickers receive the same aggregate. The same aggregate × the per-class price = a market cap that double-counts the company.

This is the opposite direction of the V/NWS/NWSA/FOX/FOXA/BRK-B/STZ allowlist pattern fixed by PR #257:

• PR feat(ingest): Issue #248 PR2b — multi-class XBRL dimensional override (V/NWS/NWSA/FOX/FOXA/BRK-B/STZ) #257 allowlist tickers: companyfacts UNDERCOUNTS (returns one class only via dimensional filter). PR feat(ingest): Issue #248 PR2b — multi-class XBRL dimensional override (V/NWS/NWSA/FOX/FOXA/BRK-B/STZ) #257 fix: per-filing XBRL dimensional sum > primary wins.
• GOOG/GOOGL: companyfacts OVERCOUNTS for the per-class ticker (returns total). A symmetric fix would need the per-class XBRL value, which would be < primary — so the PR feat(ingest): Issue #248 PR2b — multi-class XBRL dimensional override (V/NWS/NWSA/FOX/FOXA/BRK-B/STZ) #257 elif branch's summed > primary invariant rules out a naive extension.

Defense gap

NONE of the current defense flags catch it:

• data_quality_input_corruption (Defense Investigate NVDA Sloan accruals flag — genuine red flag or growth-stage artifact? #7 / risk_overlay.py:411) — TBVPS doesn't break the $10K/share ceiling because shares are plausible-magnitude (12B is a sensible order-of-magnitude for a $2T company)
• cross_source_disagreement (PR feat(ingest): Issue #248 PR2a — cross-source observability + #246 Rule 18 retrofit (schema 0.10.3) #256 observability) — yfinance.info likely also returns the same total-shares aggregate; both sources agree → no annotate fires
• share_count_extraction_missing (PR feat(defense): add share_count_extraction_missing annotate (closes #176) #181) — shares aren't None
• extreme_estimate_majority (PR feat(defense): add extreme_estimate_majority annotate (issue #177) #183) — fires correctly as a downstream symptom, but doesn't identify the upstream cause

Suggested fix path

Three options (deferred to methodology-scientist Mode B verdict):

1. Reverse-allowlist per-class extraction — for known multi-class non-dimensional filers (GOOG/GOOGL/possibly DISCK-style historic patterns), explicitly extract the per-class share count from per-filing XBRL using a different XBRL fact name or ticker-CIK-class mapping. Adds a second allowlist + extraction path opposite to PR feat(ingest): Issue #248 PR2b — multi-class XBRL dimensional override (V/NWS/NWSA/FOX/FOXA/BRK-B/STZ) #257.
2. Class-aware downscaling — for the GOOG/GOOGL pair, divide the aggregate by the known voting-equivalent class ratio. Risk: data-driven heuristic; needs methodology Mode B sign-off.
3. Defense annotate — fire a new multi_class_aggregate_shares_suspected flag when one ticker's CIK matches another ticker's CIK in the universe (Alphabet GOOG/GOOGL share CIK 0001652044). Annotate-only, no scoring veto, surfaces the pattern at cohort-audit time.

Reproduction

python -c "
import json
for t in ('GOOG', 'GOOGL'):
    with open(f'frontend/public/data/stocks/{t}.json') as f:
        d = json.load(f)
        print(f'{t}: shares={d[\"raw_metrics\"][\"shares_outstanding\"]:,} '
              f'mc=\${d[\"raw_metrics\"][\"market_cap\"]/1e9:.1f}B '
              f'mos={d[\"fair_price\"][\"mos_pct\"]:.1f}%')
"

Expected output (current state):

GOOG:  shares=12,116,000,000  mc=$4596.6B  mos=-344.8%
GOOGL: shares=12,116,000,000  mc=$4640.1B  mos=-349.0%

Surfaced by

15-parallel-agent self-audit on claude/eager-bohr-12bQi (2026-05-26). stock-detail-auditor Step 3 LLM verdict: broken_data (not cron_pre_fix_state because GOOG/GOOGL are NOT in PR #257's MULTI_CLASS_SHARE_ALLOWLIST — they're a distinct error class). Filed as a separate issue (rather than expanding PR #257's allowlist) because the underlying mechanism is opposite to PR #257's.

Cross-references

• PR feat(ingest): Issue #248 PR2b — multi-class XBRL dimensional override (V/NWS/NWSA/FOX/FOXA/BRK-B/STZ) #257 — multi-class XBRL dimensional override (V/NWS/NWSA/FOX/FOXA/BRK-B/STZ undercount fix)
• PR feat(defense): add share_count_extraction_missing annotate (closes #176) #181 / fix(ingest): per-filing XBRL fallback recovers STZ-style dimensional shares_outstanding (closes #176) #182 — share_count_extraction_missing STZ recovery
• PR ERIE shares_outstanding=2,542 — XBRL dimensional-fact extraction failure (STZ-class pattern, sibling of #176) #246 / fix(ingest): extend shares_outstanding fallback to catch implausibly-low primary extraction (#246 ERIE) #253 — ERIE implausibly-low extraction fix
• Issue STZ: market_cap = None on 2026-05-14 cron — shares_outstanding missing from XBRL fact extraction #176 — STZ shares_outstanding extraction missing
• Issue V (Visa) — shares_outstanding=469M (real ~1.85B, ~4× off) — no veto despite obvious mcap understatement #248 — V/Visa cross-source escalation
• Damodaran 2019 Investment Valuation 3rd ed. Ch. 16 — TSO calculation across share classes (per-class market cap should NOT sum to >100% of company)

Severity / priority

MEDIUM — visible to users on /stock/GOOG + /stock/GOOGL (both show absurd $4.6T market cap and -345% MoS). Composite rank impact is real but Top-5 rotation Rule 16 invariant is intact (GOOG / GOOGL are not currently in any top-5 contention because the value pillar tanks them to low ranks). No safety violation. Q3 2026-08-19 cohort audit is the natural decision gate.