Credit Brief — BYD Company Limited (比亚迪股份有限公司)

002594.SZ · Shenzhen (A-share; H-shares: 1211.HK) · Filing window: FY2023–FY2025 annual reports + 2026 interim · Sources: official periodic reports filed via cninfo (巨潮资讯网)

⚠️ EXTRACTION DEFECT NOTICE — THE FINANCIAL SECTIONS (§4–§6) OF THIS SAMPLE ARE NOT USABLE. This file is the first cold-start run of the name → public-fetch → mechanical-extraction → English-brief chain. The fetch and pipeline mechanics worked end-to-end (zero LLM spend), but BYD's financial-statement layout breaks the mechanical extractor in two compounding ways: (1) statements are presented in RMB thousands, which the extractor failed to detect and normalized as if RMB yuan — every extracted value is understated 1,000×; (2) BYD's primary statements carry a note-reference column between the line item and the period columns, which consumed the first value slot — annual-period values landed one year early, and the true current-period values were dropped (the 2026 H1 column contains only junk). The tables in §4–§6 are retained as extracted, verbatim, solely to document the defect and validate the rendering format — do not cite any number in them. Ground-truth anchors quoted directly from the filing text are in the notice block at the top of §4. Remediation belongs to the extraction engine (upstream repo); this brief must be rebuilt after the fix.

All figures are extracted mechanically from official filings and are traceable to source page (see §9). Narrative sections are analyst commentary. Not investment advice. Not a credit rating.

1. Identity & legitimacy

Note: these two sections are QCC-sourced and unaffected by the extraction defect (which is confined to the financial sections §4–§6). They are empty in this sample because the QCC (企查查) API returned only empty responses throughout the 2026-10-04 backfill window — a provider-side outage/throttle, confirmed by a same-day control query on a known-good listed issuer also returning empty. No registry facts are asserted here rather than invented; §1–§2 will be populated by a mechanical rebuild once qcc_data.json can be produced.

• (registration profile not populated in this run — QCC phase skipped)

• Legal representative: 王传福 (Wang Chuanfu) — per FY2025 annual report, company-profile section

• Auditor: Ernst & Young Hua Ming LLP (安永华明会计师事务所（特殊普通合伙）) — standard unqualified (clean) opinions for FY2023, FY2024 and FY2025 (opinion-type tables in each annual report; FY2025 report signed 2026-03-27, report no. 安永华明（2026）审字第70013328_H01号)

2. Compliance & legal

• (compliance & legal screen not populated in this run — QCC phase skipped)

Same QCC API outage as §1 — the four-family risk screen (失信被执行人 / 被执行人 / 行政处罚 / 经营异常名录) could not be run. An absent screen is not a clean screen.

3. Business profile

BYD Company Limited is a Shenzhen-headquartered, vertically integrated Chinese manufacturing group spanning automobiles (including new energy vehicles), rechargeable batteries, and electronics assembly. Founded in 1995 as a rechargeable-battery maker, it entered the automobile business in 2003 and has since become the world's largest new-energy-vehicle manufacturer by volume, selling across most price segments through multiple brands and exporting at growing scale. Its automotive operations rest on a deeply vertically integrated parts empire (batteries, power semiconductors, and other components produced in-house), and it separately runs one of China's largest handset-electronics assembly businesses. The company is dual-listed in Shenzhen (002594) and Hong Kong (1211.HK); founder and chairman Wang Chuanfu remains legal representative. Berkshire Hathaway, a famous shareholder since 2008, has been progressively reducing its stake since 2023 — a widely followed ownership story, though by the latest filings it is no longer a top-tier holder.

Top customer/supplier tables not generated for this issuer.

4. Financial dashboard (CNY mn)

⚠️ DEFECTIVE EXTRACTION — DO NOT CITE. Values below are rendered verbatim from the extraction output (outputs/extracted/财务数据_摘要.json) to document the defect. Manifest failure modes, visible in the tables: current-period columns (FY2025, 2026 H1) are 0.0 for every note-referenced line (junk 1.0-style tokens); FY2022–FY2024 values actually belong to FY2023–FY2025 respectively; and all values are 1,000× too small.

Ground-truth anchors, quoted directly from the FY2025 annual report text (「二、主要会计数据和财务指标」； balance-sheet values from the 合并资产负债表, p.123, presented in RMB thousands) — for verifying the defect, not for reuse:

Assets (defective — see notice)

Liabilities (defective — see notice)

Equity (defective — see notice)

Income statement (defective — see notice)

5. Ratios

⚠️ DEFECTIVE EXTRACTION — DO NOT CITE. Ratios are computed from the same defective inputs. Rows whose numerator and denominator share the unit error still render at roughly the right magnitude (e.g. debt-to-asset ratio), but rows mixing shifted and dropped values produce absurdities (net margin, interest coverage, receivable days) — these absurdities are themselves evidence of the defect.

Mechanical risk flags (defective — see notice)

• ⚠ Debt-to-asset ratio: 0.71 (level: medium, reference threshold: <0.75)
• ⚠ Gross margin: -10.25 (level: high, reference threshold: <0.05)

6. Cross-check flags (mechanical reconciliation)

⚠️ DEFECTIVE EXTRACTION — DO NOT CITE the individual values. The reconciliation layer itself worked as designed and is the one mechanical tripwire that caught the corruption: it raised 93 cross-document conflicts, all severity high, every one of the pattern "newer report's value vs 0.0 in the older report" — i.e. the note-column shift and drop described in the notice. The sample's §6 renderer currently dumps all 93 bullets (noise); a future build should collapse identical-pattern conflicts into a summary with counts per family/period. Two representative conflicts, with corrected citations:

• FY2023 · Cash and bank balances — FY2024 annual report, consolidated balance sheet p.123 comparative column 「货币资金」 = 102,738,734 (RMB thousands) vs FY2023 annual report extraction slot = 0.0 (junk). Deviation 100%, severity high.
• FY2023 · Inventories — same pattern: newer-report value vs 0.0, deviation 100%, severity high.

5 check(s) not run for missing inputs: 前五大应收 vs BS 应收账款; 前五大其他应收 vs BS 其他应收款; 前五大预付 vs BS 预付款项; 现金及现金等价物净增加额 vs 货币资金变动; 授信表内总合计 vs Σ机构行. (The first three need note-level top-5 tables, which the pdf-text channel did not extract from this layout; the cash-movement check needs the CF-statement net-increase line, also not captured for this layout.)

7. Red flags

WITHHELD IN THIS SAMPLE. The product discipline requires every red flag to anchor mechanically to §4–§6 data (value + period). That data is defective (see the extraction defect notice), and fabricating anchors from hand-corrected numbers would violate the traceability rule the product stands on. Red flags for BYD will be written when the extraction engine handles the RMB-thousands + note-column layout and this brief is rebuilt from a clean extraction. For format reference, see the completed sample eve-energy-300014.md.

8. Industry context

Automobiles (new energy vehicles), vertically integrated into batteries and electronics manufacturing. Chinese auto demand is strongly cyclical and policy-driven — purchase-tax incentives, emission-standard switches and the NEV transition arrive in waves, and makers without the capacity to iterate exit first (the 2018–2022 wave of insolvencies and restructurings among second-tier domestic brands is the reference case).

Risk anchors for a creditor, per the knowledge-base risk card (风险卡-汽车, entities/风险卡-汽车.md):

1. Two-way float is the real credit variable. Automakers occupy cash from both ends — dealers pay before delivery upstream, suppliers are paid after — so the good years silently stack float and short-term debt. In a downturn the float reverses (wholesale volumes sag → dealer collections slow → acceptance bills fall due → suppliers sue), and the reversal shows up in the channel and payables long before the income statement. A maker with BYD's scale of accounts-payable financing is structurally long this float; the sign of the float, not the margin, is what turns first.
2. Regulation and standards, not just demand, set the cycle — each standards switch (emissions, NEV quota, safety rules) re-prices the whole fleet and disproportionately hurts the tail. Watch for policy inflections (e.g. tighter subsidy/tax treatment, mandatory supplier-payment terms) that compress the float itself.
3. Licences are worth more than capacity. Production licences and brand qualifications retain value in distress while stamping plants and idle lines do not — heavy assets on an automaker's balance sheet are not evidence of recoverable collateral.
4. Parent and listed company are one credit. The listed maker is typically the group's cash chest; irregular guarantees, fund occupation or debt shuffling between parent and listco mean the group's distress lands on the listed entity, and a parent bond default almost always drags the listco.
5. Headline leverage understates true debt. Operating liabilities (advances from dealers, supplier payables, acceptance bills) inflate the debt-to-asset ratio while hiding the float's reversal risk; rating practice uses total-debt-to-capitalization instead. A steepening leverage slope — not the level — is the hard signal.

For BYD specifically, the 2023–2025 domestic price war and the group's vertical integration are the two structural facts a reader should hold: the first compresses margins across the sector, the second gives BYD unusual cost room and supplier-float capacity relative to peers — which cuts both ways for creditors, because the same float becomes the fastest-moving liability in a downturn.

9. Method & provenance

Pipeline ran fully mechanical in this run: cninfo fetch (4 filings, HTTP, no LLM) → inventory/processing → pdftotext-based financial extraction (MinerU OCR not invoked — text-layer PDFs and no token configured) → reconciliation sidecar → ratio computation. Zero LLM calls at any phase: qualitative extraction, material digest and QCC were skipped by flags, and the material-routing planner degraded to its static table by design when no API key is configured.

All figures in §4–§6 are rendered verbatim from the mechanical extraction output (outputs/extracted/), and the extraction is defective for this issuer's layout — see the notice at the top. Cell-level provenance below shows the defect at the raw-token level (the "raw values" captured for FY2025 are note-reference column numbers, not financial amounts):

Defect mechanism, precisely: the FY2025 consolidated balance sheet (p.123) declares 「财务附注中报表的单位为：千元」 and its rows read 「货币资金 | 1 | 75,424,747 | 102,738,734」 — the extractor (a) missed the thousands-unit declaration (its detector looks for 「单位：X」 phrasing; BYD writes 「单位为：X」), scaling every value as if RMB yuan, and (b) aligned the note-reference column (1) to the first period column, so the true FY2025 value landed in the FY2024 slot, the FY2024 comparative was dropped, and the FY2025 slot kept the note-reference 1. The interim report (2026 H1, single period column) lost its values entirely. The audit-opinion facts in §1 were read mechanically from each annual report's audit-opinion section (FY2025 report p.117).

This brief is built solely on officially filed public disclosures. It is not investment advice, not a credit rating, and not an offer of any service. Figures may contain extraction or caliber effects; verify against the original filings before relying on them.