Cn Ib Source Verifier

CN IB Source Verifier

Audit the evidence chain from disclosed claim to primary source and, where applicable, reproducible calculation.

Principles

1. Verify the claim actually made, not merely whether the source exists.
2. Match source, date, geography, population, definition, period, currency, and unit.
3. Separate direct copy, transformation, calculation, extrapolation, and management inference.
4. Prefer primary and official sources. Use secondary sources only for discovery or clearly labeled context.
5. Preserve the original value and text before normalization.
6. Never update a volatile claim without stating the new as-of date and whether historical comparability changes.

Workflow

1. Build the claim ledger

Use assets/source-ledger-template.csv or equivalent XLSX fields:

textCopy code
claim_id, claim, section, claim_type, source_type, source_title,
source_locator, source_date, access_date, period, geography,
definition, original_value, reported_value, unit,
calculation_type, start_value, end_value, years,
review_status, reviewer, notes

Create separate rows when one sentence contains multiple independently verifiable claims.

2. Classify the source

Read references/source-hierarchy.md. Identify whether the source is:

• regulator or government;
• exchange or statutory filing;
• issuer primary material;
• recognized international organization or industry association;
• licensed financial database;
• professional research;
• media or aggregator;
• management representation;
• unknown.

Do not describe an industry report as government data.

When selecting where and how to search, read references/source-connectors.md and use scripts/plan_source_search.py. Prefer a configured MCP when it is licensed and preserves lineage; otherwise use the official website, browser, or public web search. 见微数据 is a discovery/index layer unless the underlying exchange or CSRC filing is also verified.

3. Verify the locator and context

Open the primary source and confirm:

• exact page, table, paragraph, URL, or dataset series;
• publication and access date;
• whether the source is final, revised, archived, or superseded;
• geographic and product scope;
• current/historical period;
• numerator, denominator, and methodology;
• currency, unit, nominal/real basis, tax basis, and seasonality where relevant.

4. Reperform transformations

Read references/calculation-checks.md. Reperform:

• currency and scale conversion;
• totals and percentages;
• year-on-year changes;
• percentage-point changes;
• CAGR;
• market shares;
• per-unit and per-capita figures;
• weighted averages.

Keep inputs in the evidence ledger. Do not hardcode an unexplained result.

5. Run deterministic ledger checks

For CSV ledgers, run:

bashCopy code
python3 scripts/check_source_ledger.py source-ledger.csv --output issues.csv

The script validates required lineage and reperforms direct-copy and CAGR checks. Review all contextual issues manually.

6. Assess disclosure support

Classify each claim:

• verified_primary;
• verified_calculation;
• verified_secondary;
• partially_supported;
• stale_source;
• definition_mismatch;
• calculation_error;
• source_not_found;
• professional_judgment_required.

For rankings and superlatives, verify the comparison universe and date. A source that names selected companies does not prove an industry-wide ranking.

Output contract

Deliver:

1. verified claim ledger;
2. issues prioritized by filing impact;
3. corrected calculation and formula where relevant;
4. replacement primary source when available;
5. stale or non-comparable claims;
6. limitations and items requiring issuer, sponsor, counsel, or accountant confirmation.

Do not leave downloaded reports, screenshots, OCR files, or conversion artifacts in the user's project directory unless specifically requested as final evidence.