# Article 03 evidence dossier New Eruditions · Sources checked 1 October 2026 ## Research question and selection decision For one precisely bounded commercial-market claim appearing in selected publications, how many distinct underlying evidence families can be established, which dependencies remain unknown, and what support does each account actually add? **Focal claim:** Electric cars accounted for approximately 45% of new Singapore car registrations in calendar 2025. The exact input claims are 23,684 Electric registrations and 52,678 cars. An exact reported 45.1% is a separate claim, not silently equated with the calculated 45.0%. **Case rationale:** A public official make/importer/fuel table permits an independent extraction and calculation. Reporting contains explicit LTA attribution, an actual republication credit, an uncredited matching numerical pair, and a different one-decimal share. Separately attributed contextual comments make mixed support inspectable. The case concerns a commercial-market observation, not a revenue market size or an EV forecast. **Selection and counts:** The initial leads were CNA, HardwareZone and F&L. F&L was inspectable only in a search-index excerpt; a separate full-page check was blocked. The Business Times was added to obtain a third fully inspectable claim-bearing article and an explicit LTA-attributed pair for the access/discrepancy check. Final selected downstream sample: D02 CNA, D03 HardwareZone, D04 F&L (access-qualified), D05 The Business Times. Four downstream appearances, of which three have inspected claim-bearing bodies. D01 LTA is one separately counted primary-origin record. D06 The Straits Times is the credited intermediate/origin article, outside the selected downstream count. D07 Motorist is supplementary fully inspected 45.1% evidence, added to investigate the numerical issue rather than to increase corroboration. D08 Cochrane is methods context and contributes no EV observations. **Supported structure:** D03 and D05 both explicitly attribute the same 2025 numerical pair to LTA, establishing COM family F1 for the selected registration-statistics claim. D01 is its inspected primary representative. D02 is a candidate F1 member, with origin INFERRED rather than established. D04 names LTA in the indexed excerpt, but its exact numerical inputs and observation-family assignment remain UNRESOLVED. No additional independent national observation family is documented. This is one identified family plus unresolved assignments, not a confirmed exact total of one for every account. ## Grouping rule established before the count A source family is a documented shared origin of observations or measurements supporting a particular claim. Group accounts when explicit producer/dataset attribution and the claim's period, population and input values establish common observations. A transformed calculation of those observations remains in that family. A direct republication relation is recorded separately. Preserve unknown exact versions even when common statistical-series dependence is established. Do not group solely because numbers or words match, by publication order, ownership or domain. Do not call a source IND unless separately documented collection origins and relevance to the focal claim are established. Named common inputs plus distinct inputs permit PAR; neither PAR nor IND is evidenced for the national registration pair in this pack. An unresolved account is not automatically a new family or merged into F1. The classifications and certainty are separate: - DER: explicit derivative/republication of an identified upstream report for the claim - COM: documented shared observations; distinct calculation can still be COM - IND: separately evidenced observations, with documented collection provenance and relevant scope - PAR: partly shared, partly distinct inputs; name each component - CTX: contextual evidence for another claim, excluded from the focal observation-family count - UNK: empirical origin/dependency unresolved Edge certainty is CONFIRMED, INFERRED, ACCESS-QUALIFIED or UNRESOLVED. A CONFIRMED attribution records what is credited; it does not certify acquisition, verification, exact historical editions or truth. Evidence types are SOURCE FACT, DERIVED CALCULATION, INTERPRETATION and UNRESOLVED ISSUE. ## D01 official observation source URL: https://www.lta.gov.sg/content/dam/ltagov/who_we_are/statistics_and_publications/statistics/pdf/MVP02-2_New_Cars_by_make.pdf Title: Annual Vehicle Statistics 2025, Total Annual New Registration of Cars by Make. Footer MVP02-2-New cars (2023-2025). The downloaded PDF has ten pages; pages 1–4 contain the relevant table. Pages 5 onward concern earlier tables and are excluded. **SOURCE FACT / CONFIRMED:** Page 4 prints 52,678 as the 2025 total. Footnote 1: “Figures include COE and Tax Exempted Cars but exclude taxis.” This is 10 words. Page headers and the final Total row were inspected visually, not only through extraction. The table distinguishes Electric from Petrol-Electric and Petrol-Electric (Plug-In), and both importer types are retained. **DERIVED CALCULATION / CONFIRMED:** The 2025 Electric row totals aggregate to 23,684; all fuel-category rows aggregate to 52,678. The CSV contains 212 multi-year table rows, 164 with a positive 2025 count. They are table records, not 212 new independent statistical samples. The 2025 categories reconcile: Electric 23,684; Petrol-Electric 20,435; Petrol 7,638; Petrol-Electric (Plug-In) 918; Diesel 3; Diesel-Electric 0. Do not classify all BYD entries as Electric: its plug-in-hybrid row is distinct. **Version:** Downloaded 1 October 2026. SHA-256 f67067ef615027c959a4723724233b2381d956c99e2c9f0181012b37df4b84f2. PDF metadata records ModDate 16 January 2026 at 13:24:08 +08; that is not proof of publisher release or revision date. Exact historical files used by the reports are not established. A rolling URL plus current matching counts supports statistical-family grouping, not an assertion of identical file editions. **Verification:** Poppler fixed-width extraction was independently cross-checked against pdfplumber table-cell extraction. All 212 records match; each row's six body-type cells sum to its Total. Printed body-type totals also sum to 52,678. This verifies processing, not the underlying raw administrative records. ## D02 CNA URL: https://www.channelnewsasia.com/singapore/byd-top-selling-car-brand-ev-sales-5878551 Current title: BYD cements lead as Singapore's top-selling car brand as EV sales surge to 45% of new registrations. Justin Ong Guang-Xi; displayed publication 23 January 2026, 17:10, updated 4 February 2026, 15:30. Current full body inspected 1 October 2026. Numeric passages are extracted lines 313 and 326; named comments 333–360. Source credit CNA/jx. **SOURCE FACT / CONFIRMED:** Reports the same pair and 45%. No LTA or Land Transport Authority attribution was found in the inspected body. Exact origin remains UNK; the common LTA lineage is INFERRED from the matched values and explicitly attributed parallel reporting. No copied-from-HardwareZone or copied-from-BT edge is supported. **CTX:** Attributed comments by Walter Theseira and Benjamin Loo support descriptions of their reported views. They do not independently measure the focal national share. Exclusive interview acquisition or independence of the speakers' factual inputs has not been established. No New Eruditions interview was conducted. Canonical short claim fragment is in claims.csv C01; do not replicate long passages. ## D03 HardwareZone URL: https://www.hardwarezone.com.sg/lifestyle/cars/byd-best-selling-car-singapore-2025 Current H1: BYD was the best-selling car brand in Singapore last year. Displayed byline Team HardwareZone; 28 January 2026. Current full body, credit and links inspected 1 October 2026. Numeric passage line 117 explicitly names LTA and the pair; footer 158–160 credits Aqil Hamzah and The Straits Times on 23 January, and the source link is the exact D06 article. A clarification at line 156 concerns inclusion of parallel imports in 2024 figures; it does not establish a 2025 revision. **COM / CONFIRMED:** Same 2025 LTA-attributed numerical pair as D05 and the current official table. Exact LTA file edition unknown. **DER / CONFIRMED:** Direct republication credit identifies D06 as the reporting origin. The credit is not a new nationwide measurement; comments in this version must not be represented as newly collected HardwareZone interviews. A claim-bearing exact article hyperlink and the footer support the reporting edge. C08 in claims.csv preserves the short credit fragment. ## D04 F&L Asia URL: https://www.fuelsandlubes.com/electric-cars-reach-record-45-of-2025-registrations-in-singapore/ Title: Electric cars reach record 45% of 2025 registrations in Singapore. Displayed/indexed date 13 February 2026. Search-index extraction inspected 1 October 2026 supplies the opening LTA attribution and 45.1%. A separate full-page check encountered robots.txt blocking; the review also encountered 403. Neither was bypassed. Exact byline, full current body and source hyperlinks were not established. **SOURCE FACT / ACCESS-QUALIFIED:** The inspected excerpt names LTA and prints 45.1%. This establishes a reported producer attribution in that excerpt, not complete live-body inspection. Full-page inputs, population exclusions and exact dataset edition remain UNRESOLVED. **UNK:** Do not merge its exact 45.1% with C05 45.0% merely because both approximate 45%. LTA-producer dependence is recorded; exact observation-family assignment remains provisional. No evidence establishes that it obtained the number from Motorist, even though Motorist appeared earlier with 45.1%. ## D05 The Business Times URL: https://www.businesstimes.com.sg/singapore/evs-made-record-45-new-car-registrations-singapore-2025/ Title: EVs made up record 45% of new car registrations in Singapore for 2025. Derryn Wong; 22 January 2026, 19:18. Current claim-bearing body inspected 1 October 2026. Numerical pair and explicit LTA attribution are in paragraph/line 30. No exact annual-table link or intervening media credit was observed for the statistic. **COM / CONFIRMED:** Explicitly LTA-attributed 23,684/52,678, same 2025 observations as D03. Producer/series relation established; exact document edition and acquisition route unresolved. Whole-percent 45% agrees with the reproduced ratio's whole-percent rounding. **INTERPRETATION:** Additional powertrain framing is editorial selection of statistics, not another collection of national registration observations. ## Supplementary and methodological records - D06 https://www.straitstimes.com/singapore/transport/singapores-top-auto-brand-byd-sells-over-a-fifth-of-all-new-cars-in-2025-as-evs-reach-record-high/ — LTA-attributed pair; inspected briefly. Exact HardwareZone source link and credit verified. Outside selected four. Keep this source's tight use allowance; the dossier relies on HardwareZone for the republication fact. - D07 https://www.motorist.sg/article/5482/byd-was-singapore-s-best-seller-in-2025-for-the-second-year-running-with-11-184-units-registered — David Ting, 3 February 2026; full claim-bearing body checked. Explicit LTA attribution; 45.1% derivation unshown. Supplementary discrepancy record; no assumed path to F&L. - D08 https://training.cochrane.org/handbook/current/chapter-05#section-5-2-1 — Li, Higgins and Deeks, section on multiple reports of one study. Supports the established discipline of identifying underlying evidence units without discarding useful secondary information. Analogy only; no market-specific numerical weighting follows. ## Numerical issue and competing explanations **DERIVED CALCULATION:** 23,684 / 52,678 × 100 = 44.9599453282%; nearest one decimal 45.0%; nearest whole percentage 45%. Therefore conventional rounding of that pair does not explain 45.1%. The displayed one-decimal difference is 0.1 percentage point. **Sensitivity, not observations:** At fixed numerator 23,684, 45.1% requires an integer denominator 52,457–52,572 under nearest/half-up one-decimal rounding. At fixed denominator 52,678, it requires numerator 23,732–23,784. Endpoint formulas and output are retained in calculations.csv and calculation-results.json. **UNRESOLVED ISSUE:** Scope restriction, version differences, classification, calculation or reporting error are possible explanations, but none is established. Exact F&L/Motorist input tables or a dated correction would help resolve them. No outreach was conducted or authorized. Do not claim a COE-only denominator, taxi treatment, revision, error or typographical cause without that evidence. The official exact input pair and its rounded share can be used for the stated scope; the unexplained 45.1% cannot be validated for the same scope. Averaging the numbers would conceal the disagreement and shared-input dependencies. ## Evidence-use decision For a business memo, cite D01 for published rows and total, and label the 45.0% as a calculation from those rows. Retain the 2025 annual vintage, current access date and official exclusions. Cite CNA for separately attributed comments if they serve the memo's explanatory question, retaining that they are reported views. Use HardwareZone to illustrate documented republication, not as independent corroboration of the national statistic. The Business Times gives an explicit LTA-attributed numerical cross-reference. If discussing 45.1%, cite what was actually inspected and state its unresolved inputs and F&L access limitation. **Bounded conclusion:** Four selected downstream appearances; one separately counted primary origin; one identifiable observation family F1 with two selected downstream members established; two selected origin assignments unresolved at exact-observation level; no documented second independent national measurement. Two named contextual CNA views and one reproduced calculation are different evidence roles and are not added to an effective-source score.