This site checks a video's claims against official records. The rules below are the whole method — if a page ever violates one, that page is wrong.
§ M.1 The verdict scale
| Badge | Meaning |
|---|---|
| True | The claim matches the record on its own terms. |
| Mostly true | Accurate core, wrong or exaggerated at the edges. |
| Misleading | Contains true facts arranged to support a conclusion the record does not. |
| False | The claim contradicts the record. |
| Opinion | Not fact-checkable as stated; evaluated only if converted into measurable indicators. |
| Not yet verified | Checkable, but this site has not yet committed the data to check it. |
§ M.2 The three registers
Every claim page separates its prose into three typographically distinct registers, so a reader always knows which epistemic ground a sentence stands on:
Finding — what the record shows
Statements a stranger could re-derive from the committed dataset or a cited document.
Interpretation — what can be inferred
Economic reasoning on top of findings. Plausible, argued, and labelled as inference.
Opinion — not fact-checkable
Value judgments and rhetoric — the video's and anyone else's — quarantined and tagged.
§ M.3 Source hierarchy
Ranked. A claim page cites the highest-ranked source available; a lower rank never overrules a higher one on the same fact.
- Deutsche Finanzagentur — the issuer's own auction results and procedures (primary for all auction facts).
- Deutsche Bundesbank — tender result notices, statistics, monthly reports.
- ECB, Bundesministerium der Finanzen, US Treasury, Destatis, Eurostat — official European, federal and foreign-government statistics.
- OECD, IMF — international official data.
- Bloomberg, Reuters, Financial Times, TradingView — market data and reporting, used only where no official series exists, and marked as such.
§ M.4 Working rules
- No verdict without data. A claim stays Not yet verified until its dataset is committed — however confident anyone feels. Claims 04–10 carried that badge until their series (Bundesbank term structure and lending rates, ECB peer yields, US Treasury, BMF interest expenditure, the issuance and rating registers) landed in the repository.
- Numbers live in the dataset, not in prose. Summary statistics render from
the generated
_data/*.jsonartifacts (built bytools/build_data.pyandtools/build_market_data.pyfrom the committed raw files), so a data refresh refreshes every figure and chart; individually cited auctions (a date, a quota, an ISIN) are verified against the committed CSV. - German sources speak German first. Definitions are quoted verbatim in the original with a translation alongside — never translation-only.
- Metrics are named. “Subscribed” is ambiguous; pages say bids ÷ offered, bids ÷ allotted, or allotment share explicitly (the distinction that decides claim 01).
- Reproducibility is the audit.
just fetch && just datarebuilds everything; the dataset page documents each artifact's origin.
§ M.5 Known limits
- The record starts in 1999 — “for decades” claims older than that are checked against 27 years, and the pages say so.
- Auction demand is one window on demand for Bunds; the yield and lending-rate series added for claims 04–10 open others. Still absent, and named as gaps where they matter: CDS spreads (commercial terminals only), Bund–swap spreads, and bid-level/foreign-participation data (unpublished).
- Two committed registers are authored, not fetched:
data/issuance_plan.csvanddata/ratings.csvtranscribe figures from archived official documents, each row carrying its source URL. The BMF interest-expenditure CSV was retrieved manually (the portal blocks scripted downloads); its checksum is recorded like every other raw file. Interest-burden comparisons are in nominal euros — no budget-share series is committed yet. - Register quotes are English renderings of the video's German wording, verified against a
committed verbatim transcript (
specs/transcript.mdin the repository, with timestamps); the German original sits beside each register entry in_data/claims.yml.