The interest-rates record.
One maintained truth object, fully inspectable. This page is assembled from the same live record our machines serve. It keeps the kinds of thing it holds apart — what reporting says, what our own coverage measures, and what markets expect — and labels each claim with the kind of source behind it. Nothing here is retyped for marketing. When the record changes, this page changes with it.
Stated plainly: this is the second subject published in this format — the same section contract as the Hormuz record, in an unrelated domain. Sections below say “we could not look” where they could not, because that is what is true for this subject; the exhibit shows the method holding on a younger record, which is the claim.
What this record is definition
The price of borrowing money; the benchmark rates set by central banks steer economies and asset prices.
Why we keep a record on it. It runs through 11 coverage domains we watch, including AI Investing, Crypto DEFI, Ecomm News. We hold no first-document date for this subject, so the record cannot say how far back its coverage runs. That is unmeasured, not zero.
What it is not. Our attention, not the world's state. Every count here is a count of what our own sweep collected about interest rates.
Tier: full — earned by a claim carried by three or more separate publishers. Not yet met: a first-party issuer document read from the issuer itself; at least one market question bound to this record. written · Written definition, from the curated orientation glossary.
What the record holds now
Separate sections, kept apart on purpose, each addressed by the name a machine fetches it under and each carrying whether we measured it or could not look. They can disagree — and as of 2026-09-26, they do. Reading only one of them is how a machine gets this wrong.
What the issuing authorities publish primaryDocuments
declared · documents from the issuing authority itself · issuers named; no readings yet
The first-party issuers for this record are named below, but no row here has been fetched or hand-verified yet — this layer states its SOURCES, not their current values. Nothing here is a current fact; the automated verification against these issuers is the next build, and when it runs each row will state the date it was confirmed.
What outside reporting says reporting
measured · attributed sources
What outside reporting in our tracked set says. Attributed, not first-party policy data — we link each source; verify at the source. What the issuing authorities publish is the section above; what this record cannot tell you is the one named limits below.
How closely we are watching ourAttention
measured · how closely we are watching — our attention, not any central bank’s policy stance
OUR COVERAGE REGIME
steady
Our coverage of interest rates is 61% below its 3-week baseline, sustained. This measures our attention — how much our tracked set is writing about it — not any central bank's policy stance.
TRACKED (OURS)
131 days
Across 11 intelligence verticals, 50 signals in the current sweep (as of 2026-09-26). Our own count.
EVENT ARC
—
No maintained event arc in the current build.
What prediction markets price marketSensor
we could not look · 0 from the event’s own venue record, 2 market questions we recorded with their prices of our own
We could not look. This record has no bound event, so there is nothing to join market questions against. That is a blank we did not look in, not a search that found nothing. Our public grading of that sensor — those prices scored against how reality resolved — is how we keep it honest. Each question below is one defined outcome with the crowd’s odds attached; none of them is a number of ours.
How our cited sensor has scored refereeGrades
we could not look · the pass that failed: the per-subject accuracy check
We publish no probability of our own, so there is none of ours to score. What gets scored is the prediction-market prices we cite: each is checked against how that question actually resolved, and the checking runs across every subject at once rather than one subject at a time. Nothing specific to interest rates has been scored, so this section can say nothing about it on its own. The scoring anyone can read is at https://sigb.ai/api/calibration-data.
Our read ourRead
our analysis of this record's own dated sections; ours, and labeled ours — never a forecast, and never a probability of ours · as of 2026-09-26
What this read did not use (1) — show them
- single-source reporting — no press-claims ledger exists for this subject; the corroborated feed does not publish uncleared tiers.
No synthesis is written for this record. A synthesis is our own paragraph over corroborated claims with receipts; until it is computed, the clauses above are the whole read and nothing is templated to look like one.
What independent sources agree on claims
measured · window 2026-09-20 to 2026-09-26
2 claims about this subject were carried by three or more separate publishers in the window Sep 20, 2026 – Sep 26, 2026. The count is of publishers, not of truth: many outlets echoing one official statement is a wide echo, not a second check.
The press archive, claim by claim consensus
we could not look · who/what/when/where claims counted across independent reporting
What this subject’s own press archive agrees on, counted honestly: who/what/when/where claims scored by how many reporting paths that did not copy each other assert them, with counting a wire story and its reprints once applied first, so forty echoes of one wire story count once. It measures reporting agreement, not reality.
Honest state: no retroactive ledger has been built for this subject. The machinery exists — it computed the first subject's corpus — but nothing publishes here until it has run for THIS subject with real receipts. We do not preview conclusions.
What changed, and when changeLog
measured · The ledger holds 1 recorded change for interest rates, the most recent on Sep 26, 2026 — each one a change in how closely we are watching, written the moment we observed it, never backfilled and never edited after the fact. The ledger began recording 2026-07-19; states before that date were tracked but not yet ledgered, and we say so rather than backfilling.
The same question, read two ways
What keeping the layers apart changes for a machine reader. The left column is what a one-layer reader concludes; the right column is what the maintained record keeps in view. Each line is derived from the record’s own layers — the same truth object machines fetch — not from a staged chat.
Without the record — one layer at a time
Reading only the coverage line (coverage state — our own measurement): attention is 61% below its baseline and steady — concludes “nothing is moving here.”
Reading only one reporting headline (attributed): “Higher Bond Yields Mark a Return to Normal” — concludes “whatever that single frame implies — with no way to see the layers that disagree with it.”
Reading only one market price (belief, not fact): “Will the Fed increase interest rates by 25 bps after the September 2026 meeting?” at 52% — concludes “the crowd has settled it — mistaking one question's odds for the whole state of the world.”
3 single-layer reads, 3 different verdicts — each brittle, and none aware the others exist.
With the record — every layer in view
No first-party reading is on file for this subject yet: 4 issuers named, none machine-read so far. That is unmeasured, not a clean bill. Independent sources agree on 2 things about interest rates in the current window: “Good news for Modi government as S&P Global raises India's FY27 GDP growth forecast to 7 percent from 6.6 per” (9 separate publishers · 85 articles, earliest Sep 20, 2026); “Federal Reserve official says inflation persistence led her to support interest rate hike” (5 separate publishers · 9 articles · 1 wire (AP), earliest Sep 20, 2026). We cannot count single-source reporting on this subject: no press-claims ledger has been computed for it, and the corroborated feed publishes only the tiers that cleared the bar (strong, moderate), so its silence measures the feed, not the reporting. We could not measure contradiction: this record holds no pair of claims about the same proposition, so there is nothing a polarity test could run on. That is a blank we did not look in, never a clean bill. The ledger holds 1 recorded change for interest rates, the most recent on Sep 26, 2026 — each one a change in how closely we are watching, written the moment we observed it, never backfilled and never edited after the fact. We do not yet diff against a previously published copy of this record; what stands is the ledger's own dated transitions. We cannot show a divergence: the market join was not performed for this record. No first-party document has been read for this subject yet, so there is nothing to hold as provenance. An empty shelf is not a clean one. That is the maintained state of interest rates as this record holds it, with 5 of 7 clauses above stating plainly that we could not look — our analysis of the record's own dated sections, attributed throughout, ours and labeled ours, never a forecast of ours.
Composed from the record’s own dated layers — derived judgment — assembled from the layers above at build time; ours, and labeled ours. The layers it holds apart: observed policy state · observed reporting · the record itself · coverage state · market expectation · our reading · reporting consensus · the recorded history. show the machine’s judgment sentence, as served
We keep the layers apart and do not collapse them. Our attention on interest rates stands 61% below its baseline and is steady. That measures our coverage, not the subject. The reporting layer is attributed press coverage tied to its jurisdictions; the market rows price something narrower — specific committee meetings, each on its own terms — and a priced probability is not a headline's mood. The market layer prices 2 live questions — collective belief about defined outcomes, never proof. A system that read any single layer alone could answer wrongly in either direction; the maintained record keeps every layer in view, each claim tied to its jurisdiction and its kind of source. This is our judgment over the layers above, labeled as such.
Provenance of this panel: each line above is derived from the record’s own layers, not from a staged chat. The coverage numbers, the headline, and the market price are fields in the same truth object served at GET /api/truth/interest-rates; the right column is composed from the record’s dated layers, with the served judgment sentence one tap away, verbatim. No transcripts were invented — the disagreement is in the data.
What this record does not tell you limits
7 sections on this record came back "we could not look". Each is named below with the pass that failed, generated from the section's own state rather than typed here.
- How our cited sensor has scored: we could not measure it — the pass that failed is the per-subject accuracy check. We publish no probability of our own, so there is none of ours to score. What gets scored is the prediction-market prices we cite: each is checked against how that question actually resolved, and the checking runs across every subject at once rather than one subject at a time. Nothing specific to interest rates has bee
- What we have collected: we could not measure it.
- What the issuing authorities say: we could not measure it — the pass that failed is the first-party read. No first-party reading is on file for this subject yet: 4 issuers named, none machine-read so far. That is unmeasured, not a clean bill.
- What rests on one source: we could not measure it — the pass that failed is the single-source tier. We cannot count single-source reporting on this subject: no press-claims ledger has been computed for it, and the corroborated feed publishes only the tiers that cleared the bar (strong, moderate), so its silence measures the feed, not the reporting.
- What contradicts: we could not measure it — the pass that failed is no polarity-comparable pair. We could not measure contradiction: this record holds no pair of claims about the same proposition, so there is nothing a polarity test could run on. That is a blank we did not look in, never a clean bill.
- The outside sensor, beside our record: we could not measure it — the pass that failed is the market sensor. We cannot show a divergence: the market join was not performed for this record.
- Held as provenance: we could not measure it. No first-party document has been read for this subject yet, so there is nothing to hold as provenance. An empty shelf is not a clean one.
Standing limits of the method, true of every run:
- This is the second published record in this format, and the machinery has not reached every section for this subject. Each section says which pass it could not run, and the generated limits above name them; nothing is previewed to look finished.
- The coverage regime measures our declared source universe (139 actively-scraped verticals — definitions on the API page), not the whole internet, and it is our attention — not any central bank's stance.
- Cadence, stated plainly: the record refreshes on a daily sweep with intraday market updates. It is maintained, not instantaneous.
Verify it yourself verify
This page, GET /api/truth/interest-rates and the get_truth_object tool are assembled by one builder, so what you fetch is what you are reading. One honest distinction: get_entity("interest rates") returns the broader live entity dossier — a different, complementary object, not this record.
curl -s https://sigb.ai/api/truth/interest-rates
curl -s https://sigb.ai/api/mcp -X POST -H 'content-type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_truth_object","arguments":{"topic":"interest-rates"}}}'Served to machines as the resource sb://exhibit/interest-rates, and through the API and MCP.
Where this record comes from provenance
This record reads as of 2026-09-26, assembled 2026-09-26. Version sb.tr-7e0807ad — a hash of the sections above, so a rebuild that changed nothing is distinguishable from a record that moved. Release unstamped. No freshness clock is registered for this record's own inputs yet; the sections above each carry the date they speak for.
Informational only, never financial, legal, or investment advice. This exhibit is rebuilt from the live record on every data deploy.