What provenance is: a claim, its source span and the receipt that binds them

Why it matters

The newest FOMC brief on Kitalpha, "Federal Reserve Raises Target Range for Federal Funds Rate to 3-3/4 to 4 Percent", describes a decision dated 2026-09-16. Its sentences carry bracketed numbers, and beneath them sits an entry on the public ledger, the brief's receipt, signed at 2026-10-10T07:14:42.824246+00:00. A reader might take the brackets for footnotes and the receipt for a stamp of approval. Neither reading is right. The brackets are a chain back to a primary document, and the receipt fixes which text that chain belongs to. Following the chain, and knowing where it stops, is what provenance means.

The concept

Provenance is the recorded history of where a piece of information came from. For a sentence in a financial brief it answers three questions: which document the sentence was drawn from, which words in that document it rests on, and whether the published sentence is the one the author actually released. Every Kitalpha brief is built so that a reader can check all three without taking anyone’s word for it.

Start with the difference between a claim and an assertion. A claim is a statement offered as fact; in a policy brief one might read “the committee voted to maintain the target range”. An assertion is a claim with nothing behind it but the speaker’s say-so. What turns an assertion into a supported claim is evidence, and on Kitalpha evidence takes one specific form: an evidence span. An evidence span is a passage copied word for word from a named source document and stored with that document’s address and date. It is never a paraphrase and never a summary. The brief’s text points to spans with bracketed numbers, so a sentence that ends “[E2]” is saying: the words in span two, taken verbatim from the statement, are what this sentence rests on. That pointing is attribution, and it is the opposite of decoration. A bracket that pointed nowhere, or to a passage that said something else, would be an assertion in costume.

The evidence layer is deliberately mechanical. Spans are extracted from the source rather than composed, so the same statement processed twice yields the same spans. Drafting, where sentences are written around the spans, sits above that layer and must cite it. Above drafting sits a check: an automated checker reads each claim against the spans it cites and records the share it found supported. That share is the brief’s groundedness, a ratio between zero and one. It measures the text against its own evidence and is published beside the text rather than folded into it.

Then comes the receipt. When an edition of a brief is published, Kitalpha writes an entry to a public ledger, a list anyone can read. The entry carries a content hash, a fingerprint computed from the exact published text, which a single changed character alters; a digital signature made with Kitalpha’s private key, which the matching public key, published at a fixed address, can check; and the time of signing. This is the receipt. It binds a specific text to a specific publisher at a specific moment. What it attests is origin and integrity: this text came from the holder of that key and has been unchanged since. What it leaves open is everything else. A receipt says nothing about whether the sentences are accurate, whether the checker found them supported, or whether the source document itself is right. Those are separate questions, and a well-built record keeps them separate on purpose.

So the chain runs: primary document, verbatim span, cited claim, published text, receipt. Provenance is the whole chain, and reading a record well means knowing which link each field belongs to. The table below lists the fields of one live receipt. Look for what kind of thing each one is, a pointer to something you can open or a seal on the text; none is a verdict.

Worked example

Take the receipt for the FOMC brief, Signed Brief receipt c3abebf1, which signs the edition "Federal Reserve Raises Target Range for Federal Funds Rate to 3-3/4 to 4 Percent" at 2026-10-10T07:14:42.824246+00:00. Its source line names the statement the evidence was taken from, and the ledger id c3abebf1… identifies the entry. The brief itself, the brief record, carries the counts the steps below read: how many claims the checker evaluated, how many evidence spans the edition holds, the groundedness ratio, and how many citations the text makes. Every figure is taken from that record, and the steps name it above them.

Read the steps as two groups. The first two are counts of things on the record: claims in the text and verbatim spans beneath it. The third and fourth are the checker’s verdict on the text against those spans, first as the ratio the record stores and then as a percentage. The last counts the citations the text makes, a count of the text’s own pointers rather than of the table beneath it. None of these figures appears on the receipt, and none of the receipt’s fields appears in them. The receipt fixes which text was checked; the counts describe how that text stands against its evidence. A reader who wants to go further opens the source address on the receipt and compares a span with the statement, word for word.

Faded example

Now a second record from a different agency: the edition EIA Weekly Petroleum Status Report — Week Ending September 11, 2026 of the Weekly Petroleum Status Report brief, drawn from a U.S. Energy Information Administration release, with its own receipt on the ledger. The first three steps are given. Complete the last one, the edition’s groundedness as a percentage, then reveal and compare.

Retrieval check

Mark your confidence before each answer. Every option carries an explanation; read the ones you rejected too.

Your summary