How to Reconcile Conflicting Market Data Without Inventing Certainty

Who this guide is for

This guide is for market intelligence teams, manufacturers, exporters, strategy leaders, and B2B decision-makers using official statistics, commercial databases, industry reports, company filings, tenders, and interviews to select markets or accounts.

Contents

1 Treat disagreement as a diagnostic signal 2 Preserve every source in an evidence ledger 3 Classify the source of conflict 4 Apply the SellersGPT Reconciliation Protocol 5 Choose the right output type 6 Decide when the evidence is sufficient 7 Illustrative example 8 Common mistakes 9 What this does not mean 10 Common questions 11 Sources and methodology

1. Treat disagreement as a diagnostic signal

When Source A reports one value and Source B another, do not immediately ask which is correct. Ask: What exactly does each value measure, for which population, geography, time, classification, valuation, and release vintage? Official statistical agencies document that data can be revised, adjusted, sampled, imputed, or compiled under different concepts. The U.S. Census Bureau describes monthly, quarterly, and annual revision procedures for merchandise trade, including late transactions, corrections, reclassification, and seasonal adjustments. U.S. Census Bureau - Merchandise Trade Revision Procedure Statistics Canada's quality policy describes data quality as fitness for use and emphasizes documentation of concepts, methodology, and major quality features. It also notes that comprehensive measurement of every possible error is rarely possible. Statistics Canada - Policy on Data Quality and Methodology Therefore, disagreement is not automatically evidence of bad data. It is a prompt to inspect the measurement system.

2. Preserve every source in an evidence ledger

Do not copy only the headline number. Create one immutable raw record per observation: Field What to record Source Publisher, dataset, table, filing, or interview Source type Official statistic, filing, transaction, survey, estimate, interview Retrieval URL, access date, file name, query export Release vintage Publication and revision date Variable Exact field label and definition Classification HS, NAICS, SIC, product taxonomy, geography level Scope Included and excluded products, firms, flows, and populations Period Reference period, fiscal/calendar basis, monthly/annual Geography Origin, destination, clearance, headquarters, facility, market Valuation Current/constant currency, customs value, revenue, producer/purchaser price Unit Currency, pieces, kilograms, establishments, employees, capacity Field What to record Quality Sampling error, suppression, imputation, revision, disclosed limitation Raw value Unmodified source value Transformation Conversion formula and transformed value, if any The raw value must remain available after currency conversion, aggregation, inflation adjustment, or code mapping. A reviewer should be able to reproduce every transformation.

3. Classify the source of conflict

Use nine conflict classes. Definition conflict One source measures imports, another apparent consumption, and another vendor revenue. The labels may all sound like market demand, but the denominators differ. Scope conflict One series includes parts and service; another includes finished equipment only. One database covers employer establishments; another also includes non-employers. Classification conflict An HS code changed, a commercial taxonomy groups products differently, or an industry code captures establishments rather than merchandise. The United Nations Statistics Division maintains methodological guidance on coverage, time of recording, valuation, quantity, partner country, and commodity classifications for international merchandise trade. United Nations Statistics Division - IMTS Methodology Period conflict Calendar year, fiscal year, trailing twelve months, and year-to-date values cannot be compared without alignment. A monthly annualized rate is not an annual total. Geography conflict Port, customs district, province of clearance, state of origin, headquarters, facility location, and final destination are different concepts. Valuation conflict Customs value, landed cost, wholesale revenue, retail revenue, producer price, and purchaser price sit at different points in a chain. Currency and price-date choices add another difference. Unit conflict Pieces may include products of different sizes. Weight may be more stable, or equally misleading. Missing quantity data can distort unit-value calculations. USITC removed automated unit-value functionality after suppressed quantity fields produced inaccurate calculations and now cautions users to inspect suppression flags. USITC - Unit Value Data Caution Revision conflict Two downloads of the same series may represent different vintages. Keep the release date and do not silently replace previously used figures. Measurement conflict A census, sample survey, administrative record, modeled estimate, and expert interview carry different errors. BLS explains that standard errors help users judge the precision of sample estimates and that sampling and nonsampling errors are distinct. U.S. Bureau of Labor Statistics - Standard Errors

4. Apply the SellersGPT Reconciliation Protocol

The Reconciliation Protocol is a SellersGPT recommendation. Step 1 - Restate the decision question Specify the measure needed. "How big is the market?" is too broad. A usable question is: "What was the 2025 quantity of covered product imported for consumption into the United States under the defined code set?" Step 2 - Create a comparability matrix Place sources in rows and the nine conflict classes in columns. Mark fields as matched, convertible, incompatible, or unknown. Step 3 - Prefer direct fit over brand authority An official dataset that measures a different variable is not the answer to the current question. Choose the source that best matches the defined measure, while retaining other sources as context. Step 4 - Make only defensible transformations Convert currency using a documented rate and date. Aggregate product codes only when coverage is understood. Rebase indexes only when methodology permits. Do not infer missing quantities from values unless a defensible price assumption is disclosed. Step 5 - Decompose the residual After alignment, calculate: Residual difference = aligned Source A - aligned Source B Then attribute what can be explained by coverage, timing, valuation, revision, or other documented factors. Leave the remainder unresolved rather than inventing a cause. Step 6 - Triangulate with an independent measure Use a source with a different error structure: trade quantity, establishment count, production, company filing, installed-base evidence, tender volume, or buyer interviews. Triangulation asks whether the decision conclusion survives, not whether every number converges. BEA notes that it adjusts source data to align concepts, definitions, and scope and to fill coverage gaps in national accounts. U.S. Bureau of Economic Analysis - Source Data Adjustments This is an official national-accounting practice, not permission for analysts to make undocumented adjustments. Every transformation in a market study needs its own rationale.

5. Choose the right output type

Use one of five outputs: Output Use when Point estimate Sources and method align; uncertainty is acceptable for the decision Output Use when Bounded range Multiple compatible estimates define defensible lower and upper cases Scenario set Differences reflect explicit assumptions about scope, price, or adoption Directional conclusion Evidence supports relative movement or ranking, not magnitude Unresolved The missing definition or data can change the decision materially Do not publish a midpoint automatically. A range of 40 to 100 does not imply 70 is the most likely value. Name the evidence behind each bound. The World Bank's Statistical Performance Indicators framework explicitly recognizes dimensions of statistical-system performance and remaining gaps in measurable indicators. World Bank - Statistical Performance Indicators Framework The practical lesson is modest: data availability and system quality vary, so the output must match the evidence.

6. Decide when the evidence is sufficient

Apply a decision-change test: 1 What action is being considered? 2 What value or range would reverse that action? 3 Does the unresolved conflict cross that boundary? 4 Can additional research reduce the uncertainty at reasonable cost and time? If every supported value leads to the same Test decision, the team may proceed with a bounded pilot and monitor the unknown. If one plausible value supports Go and another supports Stop, reconcile further before committing. Use confidence labels:

• High: direct measure, matched definition, current vintage, documented quality.

• Moderate: compatible proxy or transparent transformation with limited uncertainty.

• Low: incomplete scope, indirect mapping, old release, or material unknown. These labels are SellersGPT editorial conventions, not official statistical grades.

7. Illustrative example

Illustrative example. All products, values, calculations, and conclusions below are fictional. They do not represent a SellersGPT customer or a market estimate. An industrial supplier finds three estimates for a North American replacement-component market:

• Source A: USD 84 million of imports under a broad code set.

• Source B: USD 61 million of manufacturer revenue in a narrower product category.

• Source C: USD 96 million from a commercial report that includes service and accessories. The team does not average them to USD 80.3 million. It builds the matrix: Source Flow Scope Valuation Geography Fit A Imports Broad parts Customs value One country Useful ceiling, not market size B Sales Core product Manufacturer revenue Two countries Narrow product proxy C Estimated market Product, service, accessories Undisclosed North America Context; method incomplete After code review, USD 22 million of Source A belongs to unrelated applications, leaving a USD 62 million one-country customs-value import proxy. Source B remains a separate USD 61 million two-country manufacturer-revenue proxy. Their proximity is a convergence clue, not a USD 61-62 million range: geography, valuation, and flow still differ. The team carries both indicators separately and leaves service, accessories, domestic output, exports, and channel inventory unresolved. The calculation is illustrative only. The important result is the traceable definition, not the apparent precision.

8. Common mistakes

Choosing the number closest to the plan Define the question and source hierarchy before seeing the result. Averaging incompatible values A mean does not repair different scope, period, geography, or valuation. Treating the newest value as automatically best A newer release may be more current but less detailed, preliminary, or based on a changed classification. Converting a missing value to zero Suppressed, unavailable, and true zero are different states. Losing the release vintage Store the exact download and revision date so a changed series can be explained. Using a polished chart as evidence of methodology Inspect definitions, source notes, transformations, and uncertainty behind the visual.

9. What this does not mean

• Official sources are not error-free or immune to revision.

• A commercial source is not invalid merely because its method differs; its fitness depends on transparency and fit.

• Reconciliation does not require all values to become identical.

• A confidence label is not a statistical confidence interval.

• A bounded decision is not permission to hide material unknowns.

• This guide does not replace professional statistical, accounting, customs, or legal review.

10. Common questions

Which source should win when both are official? The source whose variable, scope, geography, period, and method best match the decision question. Authority is necessary context, not a substitute for fit. Can we use a midpoint of a high and low estimate? Only if a justified probability model makes the midpoint meaningful. Otherwise report the bounds and their definitions. How should interviews be reconciled with statistics? Use interviews to explain mechanisms, validate coverage, and identify missing actors. Do not upscale a few opinions into a population estimate without a defensible design. When should research stop? Stop when additional precision is unlikely to change the decision or when a bounded pilot can resolve the uncertainty more efficiently. Continue when plausible values cross a material decision threshold.

11. Sources and methodology

Sources 1 U.S. Census Bureau, Revision Procedure for Merchandise Goods Trade. Official source. Accessed August 24, 2026. 2 Statistics Canada, Policy on Informing Users of Data Quality and Methodology. Official source. Accessed August 24, 2026. 3 U.S. Bureau of Economic Analysis, Does BEA Adjust Source Data Used to Estimate GDP? Official source. Accessed August 24, 2026. 4 U.S. Bureau of Labor Statistics, Measures of Reliability: Standard Errors. Official source. Accessed August 24, 2026. 5 U.S. International Trade Commission, Why Is There No Unit Value Data Anymore? Official source. Accessed August 24, 2026. 6 United Nations Statistics Division, International Merchandise Trade Statistics Methodology. Official source. Accessed August 24, 2026. 7 World Bank, Statistical Performance Indicators Framework. Official source. Accessed August 24, 2026. SellersGPT analysis and framework The evidence ledger, nine conflict classes, Reconciliation Protocol, output types, decision-change test, and confidence labels are SellersGPT editorial recommendations. The official sources document quality, revisions, methodology, and limitations; they do not endorse any specific market estimate produced with this framework. Update policy Review whenever a cited dataset is revised, classifications change, a new benchmark release appears, a source changes methodology, or a decision uses a different scope. Preserve old vintages and the audit trail.