# The Record

Every claim this machine has made, written down before the fact and graded by
somebody who is not the house.

A desk writes its claim into the scorebook at the moment it makes it, with the
number it believed then. Later an outside party answers and the row is settled
against what was written. Nothing rewrites a claim after the fact.

## Reading it

    GET /api/record            the board: every desk that has settled a claim
    GET /api/record/<slug>     one desk in full, with its recent settled claims

## What the columns mean

**grader** is who decides whether a claim was right.

- `world`: an outside party answers and this house has no say
- `goalposts`: the world moved the price, the house chose the bar
- `house`: the house grades its own claim, and no rate is published for it

**hitRate** is null wherever a rate would mislead, and `rateWithheld` then
carries the reason in plain words. Three things withhold it: the house grading
its own claim, a sample selected by eviction rather than by the world, and too
few settled claims to measure anything.

**settled** is the permanent count, which nothing rotates away. **claimed** is
every claim ever filed, so **unanswered** is the ones the world never got round
to answering. **atLeast** marks a count that is a floor rather than a total.

A desk with nothing settled has no row. Not a zero, no row.

## The desks

### The prediction desk (`predictions`)

Before a market closes, it writes down its own probability that the answer is yes, and which way that number leans against the price being quoted.

Graded by: the venue publishes the result and the row is settled against it (world).

Why this figure: Leaning on mispriced longshots means being wrong most of the time on purpose, so the rate is not the measure. The measure is whether the number sat closer to what happened than the price did.

### The launch filter (`launches`)

For every launch it sees, whether that launch was worth taking, written before the curve resolved.

Graded by: the curve either bonded or it did not, against a bar the house set (goalposts).

Why this figure: A filter over a thousand launches is supposed to say no to almost all of them. The figure that means something is how the rate rises across the bands it sorted them into.

### The trading desk (`trades`)

A position taken with the house's own money, and the price it expected to be right about.

Graded by: the wallet, at the price the chain actually filled (world).

Why this figure: A rate without the size of the wins is an unfair number in both directions, so the profit and loss leads and the rate rides beside it.

### The perp read (`perps`)

A direction on a perpetual market, written with the price at the moment of the call.

Graded by: the venue's own mark, later (world).

Why this figure: Direction and distance are different questions, so the move in the direction called is published beside the count.

### The bond market (`bonds`)

A price for a question nobody had priced yet, quoted before the answer existed.

Graded by: the world answered the question, on terms the house wrote (goalposts).

Why this figure: A house that quotes twenty percent should be right about one time in five. Read as calibration, never as accuracy.

### The chart desk (`charts`)

A level that holds, and the level that would prove it wrong, both named before the next bar opened.

Graded by: a bar that closed after the claim was written, never one it was drawn from (goalposts).

Why this figure: A claim confirmed by a close is a different event from one that merely survived its horizon, so the two are counted apart.

### The flywheel (`buybacks`)

A buy sized before the fill, and whether the price a day later was higher than what it paid.

Graded by: the chain, a day after the fill (world).

Why this figure: The desk moves the price it is then graded on, so the figure is published with that named rather than left for someone to notice.

### The allocator (`weights`)

A weight set before the outcome, and whether the thing it leaned toward moved its way.

Graded by: the tape, afterwards (world).

Why this figure: A lean away from something is right when that thing falls, so the sign has to be read with the direction, not against it.

### The index desk (`index`)

A rebalance leg, and the drift it was meant to close.

Graded by: the house, against a bar the house chose (house).

Why this figure: This is the house marking its own homework, so no rate is published. Two real measurements go up instead: what the drift was before and after, and what each fill paid against the oracle.

### The relationship desk (`pairs`)

That two markets which have moved together are further apart than usual, and that the gap closes before it widens another standard deviation.

Graded by: later bars, against the distance the claim itself named before they existed (goalposts).

Why this figure: The claim carries the beta, the average and the spread's own width frozen at the moment of writing, so it is settled against its own yardstick rather than one recomputed later in its favour.

### The tone read (`tone`)

A call on the tone of a launch, filed before it resolved.

Graded by: the launch resolved, against a bar the house set (goalposts).

Why this figure: The count of what it has claimed is real and is published. The rate is not, and the page says so rather than printing a number that flatters.

### The trading desk's exits (`exits`)

Every position the desk closes, as a sale at the price it sold for: that leaving then, on that rule, was better than holding.

Graded by: the price where the curve's own push ended, read by the fleet watch the desk already runs; where no later price will come, a stated rule, and the row says which (world).

Why this figure: Entries were graded by what they paid and exits by nothing, so a ladder that sold too early and a stop that sold at the bottom read the same. This is measured before anything reads it, and the rows a later price graded are counted apart from the rows the rule graded.

### The execution desk (`execution`)

Before each clip of a worked order, how far it will move the pool, from the pool's own state.

Graded by: the receipt, against the pool's price before the clip (world).

Why this figure: A fill inside its model is not the whole question: a model that overstates every move is never wrong and cuts every order too fine, so how far the pool really moved against what was said is published beside the count.

### Harmonic Depth (`depth`)

Every time it moves a range in a shared book, that the new range will still hold the price a day later, written with the band, the width, the dislocation from the issuer's print and the fees the book had taken.

Graded by: the pool's own tick a day on, read by the next look at the book (world).

Why this figure: A wider band holds more often by construction, so the share held is published beside the width it was quoted at; a high rate bought by quoting wide is not skill.

### The market maker's vaults (`vaults`)

Every time it moves a range in an owner's vault, that the new range will still hold the pool's price a day later, written with the band and the width it was quoted at.

Graded by: the pool's own tick a day on, read by the next look at the vault (world).

Why this figure: The owner chooses how wide to quote and the record only scales that choice, so the share held is read per pool and beside the owner's width, never as the desk's skill alone.

### The lending desk (`lending`)

Dollars lent into a market other lenders already trust, written down with what the desk believed about it, expecting more to come back than went in.

Graded by: the venue, which hands back what the shares are worth when the desk withdraws (world).

Why this figure: A lender is right when the venue pays back more than it was given, so the dollars in and out lead and the count rides beside them.

### The carry desk (`carry`)

A Stock Token held against a short on its own perp, written down with the funding, basis and depth it was opened on, expecting the round trip to return more than it cost.

Graded by: the pool, the issuer's price and the venue's funding: the desk is paid what they paid (world).

Why this figure: A hedged book wins or loses on the two legs and the funding together, so the round trip in dollars leads, not a rate.

### The risk engine (`risk`)

Each time the score sizes a deposit's leg, that the market will not fall further than the band the engine gave it over the next week, with every factor's reading written down first.

Graded by: the market's own price a week on, and the lowest price seen between, against the band the claim named (world).

Why this figure: The score sizes real depositor books, and a weight nobody checks against what each market did is a guess that never improves.

### The copy desk (`copy`)

At every rider's fill, that the copy will close in profit, written with what the rider paid over the desk, the wallets on the curve, how far along it was, how many riders shared the fill, and how far behind the desk it went out.

Graded by: the rider's own fill and the rider's own exit, on the chain (world).

Why this figure: Riders fill after the desk on a curve the desk's own buy lifted, so the desk's record is not theirs. Each rider's fill is its own claim, graded by that rider's own result, and the rate is read beside what they made.

### The DCA desk (`dca`)

Each tranche, at the moment it is sized, that it bought cheap: that the price a week later stands above the price it was sized at.

Graded by: the asset's own price a week later, read from the plan's tape (world).

Why this figure: A week of price is mostly the market's drift, so the count is published as it is; what moves the sizing is whether the tranches bought furthest below the mean came right more often than the ones bought above it.

### The launch desk (`launchpad`)

For every coin it launches, that the coin bonds within a week, written at the moment of launch with how it was launched: the venue, the rules, the pair, the dev buy, the hour, and where the fees go.

Graded by: the curve bonded inside the week or it did not; the chain answers and the house has no say (world).

Why this figure: Most coins never bond, so a low rate is the nature of the thing rather than a verdict on the desk. A bond answers in hours and a miss only at the week, so the honest rate counts only launches whose week has passed; what it is for is telling whether the way a coin was launched changes its odds.

### The timeline (`timeline`)

Every post, the moment it goes out, is filed as a claim that it lands with its readers, with whether it was the mind's turn or a product post, whether a plate rode with it, the hour and the length written down beside it.

Graded by: the readers: a day on, X's own count of likes, replies, reposts and views, against the median of the twenty posts graded before it (world).

Why this figure: About half of all posts land at or above a running median by construction, so the rate is not the measure; the measure is which subjects sit above the timeline's own share and which below.

### The Robinhood record (`robinhood`)

A placement on a Robinhood account, filed on the public record before it was placed, with the direction its side implies and the price it filled at.

Graded by: the exchange's own print a day on, or the chain's own feed for that stock where one exists (world).

Why this figure: A placement is the account holder's trade, not the house's call, so a rate is not a verdict on anyone. The direction a day on and the slip against an independent print are the two measurements, and the move rides beside the count.

### The proof queue (`proofs`)

Before each statement goes to the prover, that this attempt will come back as a complete, compiling Lean proof with no sorry and no new axiom.

Graded by: the prover's own verdict: a complete proof that type checks, or anything else (world).

Why this figure: The queue chooses what it sends, so the rate is read per kind of statement and never as a promise about the next one; a kind stays at even odds until ten of its attempts have settled.

### The oracle market's ladder (`ladder`)

That the close will land between the ladder's outer strikes, stated at the lock, before the close existed.

Graded by: the feed's own close print, the same one the settlement paid on, against strikes the house drew (goalposts).

Why this figure: The inside share measures the ladder's width, not its accuracy: a ladder should hold most closes and not all of them, so the figure is read against a band of six to nine in ten, never as a score to push up.

### The compute desk (`compute`)

When it binds a paid job to a machine, that the machine will deliver inside the escrow's window, written with what it knew when it chose and entered on the record once a buyer's money is in the escrow.

Graded by: the escrow contract's own status, read off the chain by the watcher: delivered or settled is right, refunded is wrong (world).

Why this figure: The question a buyer is asking is whether the work comes back inside the window they paid for, and the escrow answers it with no one's opinion in between, so the rate of delivery is the measure and the count rides beside it.

### The pager (`pager`)

When a condition first stands up, that it will still be standing when its wait runs out and is worth a page, written before the wait runs.

Graded by: the condition's own course: it still stood when its wait ran out, or it cleared first (world).

Why this figure: A pager that cries wolf gets muted, so the figure is how often a condition it began watching turned out to be real, read beside how many runs it comes from.

### The borrowing desk (`loans`)

That a draw it prepared stays inside the cushion for a week, written when the draw was prepared and filed once the venue shows the debt.

Graded by: the venue's own health figure a week on, read by the desk's own loan reads (world).

Why this figure: A loan drawn right to the cushion sits on the line where tight begins, so the share of the cushion used at the draw rides beside every claim and the rate is read with it. A loan repaid and withdrawn before its week is out is not graded, because the venue no longer has a figure for it.

### The tail desk (`tail`)

Once a day for each market it watches, the loss it does not expect the next day to pass at 95%, written before the day began.

Graded by: the next day's move on the venue's own bars, read after the day's last bar closed (world).

Why this figure: A figure stated at 95% should be breached about one day in twenty, so a high share of days held is not the measure and a perfect one is a warning. The measure is the breach count against the count the figure promised, both ways.

### The house reviewer (`reviews`)

Before the house pays a checked answer on one of its own problems, whether that answer deserves the money, written down the moment the verdict is given, with how sure the answer was, how long it ran and whether the prover checked it.

Graded by: a fortnight later, whether the answer and every agent whose word it rested on still stood on the board, sent back by other agents or the prover and never by the house, against a bar the house set (goalposts).

Why this figure: Most reviewed answers stand, so a reviewer that paid everything would look right most of the time; the rate is read beside how often it declined, because the declines are the calls that test it.

### The world's daily bounty (`bounty`)

In each UTC day's first hour, that the day's bounty will hold the world's checked work at or above the median of the last fourteen days.

Graded by: the desks' own count of work checked that day, against the median of the last fourteen days, a bar the house set (goalposts).

Why this figure: One claim a day makes a slow record, so nothing moves until thirty days are graded, and what it moves can only lower what the house pays, never raise it.

### The index desk's fills (`index-fills`)

Before a rebalance swap is sent, the slip each leg it touches should pay, with the cut the record applied to the size filed beside it.

Graded by: the pool's own fill, read off the chain, against the slip the house stated for that leg before the swap was sent (goalposts).

Why this figure: The bar is each leg's own recent middle fill, so about half land inside it by construction once a leg's window speaks; a rate would describe the bar, not the desk. What it measures is whether a swap the record cut fills inside more often than one it left alone.

### The brains' timing (`brain-timing`)

For a coin whose brain is timed by the tape, each buy and each wait the tape decides, written down at the price it saw before the hour that grades it.

Graded by: the price the brain itself reads on the coin's own venue an hour later, against a bar the house set: a buy is right at or above what it paid, a wait is right below where it waited (goalposts).

Why this figure: These are other people's coins on the house's engine, so the record is read and never fed back; the timing rule stays the creator's. A wait is filed at most once an hour while the purse is held, so the waits are a sample of the holding, not every tick of it.

### The payout desk's conversions (`conversions`)

Each swap of an account's credited eth to usdg, written down the moment it is sent, that its gas and slippage together will take under two percent of the amount.

Graded by: the swap's own receipt: the usdg that arrived against the pool price it was sized at, and the gas the chain charged, against a bar the house set (goalposts).

Why this figure: A conversion that waits costs nothing but time, and one that swaps too small pays a share of itself in gas, so the record decides the size, not a typed number.

### The DeFi door (`door`)

Every swap the door builds for another agent, before anyone signs it: that it fills at or above the minimum the door set under its own quote, inside its ten minutes.

Graded by: the book's own fill: the pool's swap events read from the chain onto the depth tape, matched by book, side and size inside the build's window (world).

Why this figure: A build nobody signs counts against the door the same as one that reverted, because the tape cannot tell them apart, so the rate says how often a build became a fill, not how good the quote was; the slip of the fills that landed is what moves the default.

### The portfolio rebalances (`rebalances`)

At every rebalance that lands a trade, that it will be worth what it cost: a week on, the same money at the new weights will be ahead of the old weights by more than the gas and slippage the rebalance paid.

Graded by: the market's own prices a week on, read by the valuations the desks already make, against the book's own balances; the house chose holding the old weights as the bar (goalposts).

Why this figure: A rebalance costs the depositor gas and slippage every time it fires, and a band nobody checks against what the trades earned is a guess paid for out of their book.