VANTAGE is the viewpoint: it fixes where you stand when you read — as-of date D, nothing accepted after D. PARALLAX is the shift: what the same fact appears to do when the viewpoint changes. This page is the third term, the one the other two stand on — the baseline, the known, rigid separation between viewpoints. Optics settled this first: a shift becomes a measurement only across a baseline you can cite.
The bracket is honest about its own state: the first measurement across this baseline has been taken. Scoped cik % 10, 17,787 filer-quarters — as-of and restated YoY revenue growth come out bit-identical for 89.4% of them, 3.4% diverge by more than 10pp, 1.51% flip sign. Rare, but violent where it lands. Re-read 150 days later the divergence barely moves, and that is the part that counts: what shifted was the data, not the read path. Attribution is exactly what the separation buys, and it has now been paid out once. A full-surface run stays unclaimed; STATUS.md bounds the rest.
Turning the same distinction outward changed what it is evidence for. Read against a commercial market-data and SEC-fundamentals API — its published surface, not its data — the filings and fundamentals endpoints carry no as-of mode and no acceptance instant; the field dating a period is the most recent filing that touched it. The same vendor timestamps trade and quote receipt to the nanosecond. So the zero-baseline failure below is not a cautionary sketch — it is the default shape of the surface most consumers read, and the precision that would fix it already exists one product line over.
Friedrich Bessel picked 61 Cygni for its restless proper motion, took Earth's orbit as his ruler, and read the shift: 0.3136″ ± 0.0202″ — about 10.4 light-years, the first measured distance to a star. The value on record today, two centuries of instruments later: 0.286″ — 11.4 light-years. The first parallax ever taken carries its own as-filed → restated history, and it embarrasses nothing: the viewpoint was recorded beside the value, so the revision is data rather than an error. That bookkeeping — the viewpoint stored next to the fact — is exactly what VANTAGE's gold model is.
Naming the two visible terms was the easy part: VANTAGE produces viewpoints, PARALLAX measures what shifts between them. But nothing about two viewpoints makes them comparable. What does is the separation between them being known before the measurement and rigid during it — in astronomy the surveyed orbit, here the contract. Every claim either repo makes routes through it.
| in the instrument | in the pair |
|---|---|
| two observation points | two as-of reads of published gold — at D, and at the latest record VANTAGE produces both |
| the near star | the same filer-quarter — one economic fact, filed and later restated |
| the distant background | the gold schema and its semantics — the common ruler both reads are read against |
| the apparent shift · θ | the revision: as-filed at D vs. restated at latest PARALLAX measures it |
| the baseline · B | the contract: one surface (ADR-0001) · one Reader protocol · a gate shown able to fail this page |
| the inferred distance · d | the finding — a shift attributable to the data, not to the read path |
Withhold the baseline and the shift is still visible — it just stops meaning anything. A consumer without the contract measures its own read path and files the artifact as a revision. That is why the consumer gate runs before any study does, and why an unevaluable read refuses instead of passing.
A coincidence rangefinder is two windows at the ends of a rigid tube — the tube's known length is the instrument. The pair's tube has three sections, each load-bearing, each earned by a refusal rather than a promise.
Published gold is the only thing PARALLAX may read — never bronze, never silver, never a path around the boundary. Sibling, not fork; the consumer never writes upstream.
Gate checks and study queries are written against a Reader protocol — never a path, never a connection string — so every surface answers the same question with the same semantics.
A gate that has never been red holds nothing up. Every surface earns two cells: green on the live gold, red on a planted twin staged the same way.
Lane 1 (local Parquet) holds both cells today. Lanes 2–3 — Unity Catalog, then Snowflake over Iceberg — are empty on purpose: the sentence isn't claimable there yet, and the page that bounds those claims is STATUS.md, not this one. Each lane re-earns both cells against a twin staged in that catalog; vendor read paths differ, the invariants don't get to.
Each strut exists because its failure mode is real. Every optics failure has a system twin — and each twin has either been caught in this build already or is pinned so it stays caught.
One viewpoint. Nothing shifts; a near star and the far background are indistinguishable. There is no measurement to get wrong.
Latest-only reads — no as-of discipline, so lookahead has nowhere to show up and quietly becomes the dataset. A backtest built on it isn't wrong about the edge; it isn't about an edge at all.
The shift is right there on the plate, but with no citable separation it never becomes a distance — a number with no meaning.
An unproven contract. A divergence between two reads can't be attributed: data revision, or read-path artifact? An always-green gate lives here too — green that has never been shown able to go red cites nothing.
The separation moves between the two observations and the instrument manufactures shift — false parallax, indistinguishable from signal on a single pair of plates.
Read semantics drifting between reads: the Delta bare-glob double-count after an overwrite; zoneless SEC timestamps parsed in local zone, moving the PIT boundary between machines. Both real, both caught, both pinned as tests.
The struts and the failures are the same list with the sign flipped — which is exactly what it means for the baseline to be the instrument.
The whole site now carries both grounds — the ink set it launched with, and the paper set that was earned here token by token. The same header toggle rides every page, one remembered choice governs them all, and ink stays the site default. The ledger below is the mapping the rest of the site inherited.
| role | ink theme | paper theme | carries |
|---|---|---|---|
| ground | #0A1424 | #F6F4EE | the page itself |
| text | #E9EEF8 | #1E232C | prose |
| gold | #E0A436 | #B07C1E line · #7E5A10 text | authority marks — the baseline, eyebrows, the brand dot |
| ok | #5BB98C | #1E7A4D | pass · claimable |
| block | #D4654A | #AE4529 | fail · refusal · UNEVALUABLE |
| blue | #6CA7E8 | #2D5FA3 | sight lines · measured values |
One gold at night becomes two in daylight — the line keeps its metal, the type drops darker to hold contrast. That asymmetry is the only concession the port needed. State of the test: passed and propagated — the toggle now rides every page of the site.