DeltaRates — Six things you can check yourself

0 checks run · every step below is live; nothing on this page is pre-recorded.

The pack that has to go out on Monday

The situation

It is Tuesday 8 September 2026. You are the NAV accountant at the fund's administrator. The yield accrual pack for the 7 September NAV is on your desk: eight positions, $17.2m, $47,330 of investment income. It goes to investors on Monday 14 September 2026.

Investors subscribe and redeem at this NAV. An error in it is a restatement, not a footnote. Materiality on the investment-income caption is $2,367.

What you have to do

What normally happens

about 4 days · you, the manager, and their ops team

With DeltaRates

under 4 seconds · you alone

1Get the pack, and the detail behind each line

You email the manager asking for the support behind all eight lines, and wait. A spreadsheet comes back that they built from their own records.

You send the same email, and the manager still prepares the pack. What arrives is different: each line names the exact dates it accrued over and cites a signed on-chain measurement at each end, so the support is attached rather than promised.

2Check the arithmetic

Re-add all eight lines in your own workbook, using the numbers they gave you. You are testing their sum, not their source.

One button re-derives all eight from the signed measurements themselves, here in your browser.

3Confirm the source, not just the sum

Hand-check a line or two against a block explorer, and take the manager's word on which block and which dates each figure refers to.

Each line's two endpoints are named by hash and re-read from the chain, so the figures are checked against Ethereum rather than against the manager.

4Resolve what does not tie out

Query the lines you cannot tie. Wait again. Accept management's representation for whatever is still open when the deadline arrives.

The exceptions come back already identified, each naming the bytes in dispute, so the manager receives a fact instead of a question.

5Sign it off

You sign the pack with the manager's own numbers as your evidence.

You sign it with evidence the auditor will re-perform against at year end, rather than with a representation they will have to re-test.

The pack, as it arrived

Illustrative pack. The signed series behind every line is live, and so is everything the sweep does with it.

Yield accrual pack for the NAV struck at 7 September 2026. The last two columns are filled in by the sweep.
#	Position	Accrual window	Stated size USD	Booked rate / income	Re-derived rate / income	Verdict
1	USDC lent via Aave v3 Aave v3 · USDC held throughout the period	11 Aug to 7 Sep	4,250,000	3.67% $11,398	not yet	not yet checked
2	USDC lent via Compound v3 Compound v3 · USDC held throughout the period	11 Aug to 7 Sep	1,800,000	4.64% $6,049	not yet	not yet checked
3	Sky savings USDS Sky · sUSDS entered 13 Aug 2026	13 Aug to 7 Sep	2,400,000	6.40% $10,179	not yet	not yet checked
4	Lido staked ETH Lido · wstETH entered 21 Aug 2026	21 Aug to 7 Sep	3,100,000	2.26% $3,225	not yet	not yet checked
5	Frax staked ETH Frax · sfrxETH entered 21 Aug 2026	21 Aug to 7 Sep	950,000	2.64% $1,155	not yet	not yet checked
6	Ethena staked USDe Ethena · sUSDe entered 1 Aug 2026, before the fund onboarded	1 Aug to 7 Sep	1,500,000	4.47% $3,241	not yet	not yet checked
7	ETH/USDC liquidity Uniswap v2 · UNI-V2 entered 21 Aug 2026, booked as LP return	21 Aug to 7 Sep	1,200,000	3.11% $1,711	not yet	not yet checked
8	OTC yield note, bilateral held off-platform · OTC entered 1 Aug 2026	1 Aug to 7 Sep	2,000,000	5.10% $10,372	not yet	not yet checked
	Total		17,200,000	$47,330

What that is worth

The misstatement is caught before publication. Line 3 is out by $4,487, nearly twice materiality. Found on Tuesday it is an email; found in next year’s audit it is a prior-period adjustment.
Each exception comes with its next action. One line to send back, one to relabel, one needing a different procedure, one unchanged.
The auditor re-performs against the same evidence. Year-end starts from agreed facts instead of your working papers.

Year-end, and investment income is a significant account

The situation

You are the senior on the digital-assets section of this fund's 31 December 2026 audit. Investment income is a significant account and the yield positions are all of it. Fieldwork closes in three weeks, the file goes to partner and EQR review, and the firm's digital-asset files were selected for regulatory inspection last cycle.

PCAOB AS 2501 and ISA 500 both rank independent third-party confirmation of observable data above management representation. That is the category you are trying to get this evidence into.

What you have to do

What normally happens

a sample of 5 of 8 · conclusion: “management represented…”

With DeltaRates

all 8 lines · conclusion: “re-performed against independent evidence”

1Request the schedule and its support

You request the client's schedule plus their supporting exports, then tie those exports back to the client's own ledger.

The same request. You still ask the client for the schedule. It arrives with signed evidence cited per line, so there is no separate export to tie and no reconciliation between two of the client's own files.

2Decide how much of it to test

You sample five of the eight lines, because testing all eight costs more time than the extra coverage is worth.

You test all eight. At one click per line the sample stops being a budget decision.

3Establish the evidence is genuine

You cannot, independently. The export came from the client, and an honest export and an edited one are the same file.

You confirm who signed it. A forgery is manufactured in your browser and fails on the one thing that cannot be imitated.

4Re-perform the measurement

You query whichever data provider your firm happens to have, at whichever block the client told you to use.

You re-read it at whichever archive node your firm already trusts, at the block the evidence itself names.

5Establish what the figure covers

You ask for the basis of the return and receive a spreadsheet whose gaps, if any, are invisible.

The figure states the period it actually describes and the count of signed measurements behind it.

6Put it in the file

Your conclusion reads that management represented the rate applied was correct.

Your conclusion reads that you re-performed it against independent signed evidence, and the file still verifies years later with the vendor switched off.

What that is worth

Re-performance, not a representation. The workpaper reads “re-performed against independent signed evidence at a node of our choosing; agreed to $0.31.”
Your reviewer can re-run it from the file. Every endpoint is named by hash, so partner, EQR and inspector reproduce the test without contacting the client.
The population becomes the sample. At one click per line there is no reason to test five of eight.
It still works in year three. The evidence does not depend on us having an API, a price, or a business.

An allocator's due-diligence questionnaire, due on the 18th

The situation

You run the fund. A $40m allocation is in late-stage diligence and their ODD questionnaire arrived Monday 7 September 2026. Section 7.3: describe how reported returns are independently substantiated, and provide supporting evidence. It is due on Friday 18 September 2026.

The last two ODDs you answered ended the same way: a NAV letter, some screenshots, then a request for read-only access to your systems and six more weeks. The allocation stalls while the question is open, and the last one went elsewhere.

What you have to do

What normally happens

about 6 weeks · ends in a systems review

With DeltaRates

one attachment · nothing of yours exposed

1Evidence the returns you reported

You ask the administrator for a NAV letter and screenshot each position from the protocol's own interface.

There is nothing to request, because you are the one who reported them. The evidence was created at the moment each measurement was taken, months before this questionnaire existed, and your figures already cite it.

2State the period each figure covers

You draft a paragraph describing your methodology, and hope the description is accepted in place of the evidence.

The period is on the face of the bundle, with the measurements inside it counted.

3Let them satisfy themselves

A call where you explain why your numbers should be believed, followed by a request for read-only access to your systems.

They verify it on their own machine, at nodes they choose, with you not in the room and nothing of yours exposed.

4Answer the next allocator

Start again from scratch, because the last answer was assembled by hand for one reader.

Send the same artifact. It is not addressed to anybody in particular, which is what makes it re-usable.

What that is worth

You answer 7.3 with an attachment, not a paragraph. Nothing for them to inspect and nothing for you to expose.
Weeks off the item that normally stalls diligence. Performance substantiation generates the follow-ups, because every answer to it is currently your own word.
One artifact serves every audience. The allocator, the auditor and the administrator all read the same evidence.

Your IC meets on Thursday and the manager reported 3.67%

The situation

You allocate to external managers. One runs a stablecoin yield sleeve you are in for $25m, and their September report says 3.67% annualised on the USDC lending book for the 28 days from 11 August to 7 September 2026. Doubling the allocation is on the IC agenda for Thursday 10 September 2026.

Your mandate requires independent verification before an increase. You have two days, no site visit scheduled, and if the figure is overstated the finding lands at your next mandate review, on you rather than on the manager.

What you have to do

What normally happens

about 2 days · and it is not independent

With DeltaRates

an afternoon · the manager is not involved

1Get support for the reported figure

You email the manager asking for supporting detail, and receive it from the party whose performance is the question.

You do not have to ask. The evidence behind the figure is signed and public, so it is not the manager's to give or withhold. This is the one journey here where the request disappears entirely.

2Check the arithmetic

Recompute their return from the numbers they supplied.

Re-derive it from the signed measurements at each end of the period, at public nodes with no relationship to us or to the manager.

3Rule out a period chosen to flatter

You cannot tell. Nothing in what you were sent shows whether the start and end dates were picked after the outcome was known.

Both ends were signed on the days they describe, 28 days apart, each before the other existed.

4Take it to the investment committee

You present it with a caveat about where the support came from.

You present it as verified, with both endpoints named by hash so any member can repeat the check.

What that is worth

You verify it without asking the manager for anything. Which is the whole difference: requesting support from the party being assessed is not independent verification.
A flattering window is not available to anybody. Both ends of the 28 days were signed days apart, each before the other existed.
Cheap enough for every manager, every quarter. Monitoring cost is why most allocators verify on exception; at an afternoon a manager that stops being the constraint.

Your index review, and an auditor asking about input data

The situation

You administer a published DeFi reference rate that structured products settle against. Your annual methodology review is Friday 9 October 2026, and your external auditor's request list arrived ahead of it: evidence of the input data behind every determination, and the controls stopping it being altered afterwards.

Structured products settle on this rate. A determination you cannot evidence is a finding against the benchmark itself, and subscribers can dispute settlements struck on it.

What you have to do

What normally happens

your own table is the audit trail

With DeltaRates

an audit trail nobody has to trust you on

1Collect the input data

You pull each input from a vendor API into your warehouse, on your own schedule.

You pull it the same way, on the same schedule. The difference is what each row carries: a signature over its own exact bytes, made when the value was measured.

2Date each input correctly

You record when you fetched it, which is not when it was produced. On an event-accrued source those differ by hours.

Each input carries the second its own source reports it accrued to, separately from the block it was read at.

3Store it so it can be evidenced later

It goes in a table you control, which is precisely the thing your auditor is asking you to prove was not altered.

The signature travels with the row, so the evidence does not depend on your warehouse or on anyone trusting it.

4Publish the determination

Your methodology is applied to the inputs you hold. Unchanged.

Your methodology is applied to the inputs you hold. Unchanged — every input behind it is now independently checkable.

5Evidence it at review

You produce that same table and explain to the auditor why they should trust a record you can edit.

You hand over evidence they check themselves, for any determination in the last twelve months, with us switched off.

6Survive a vendor change

You re-plumb the feed, and the historical evidence for determinations you already published is at risk.

Past determinations stay evidenceable whatever happens to the vendor, because the proof is in the rows you already hold.

What that is worth

Your audit trail stops being a table you control. Auditor, regulator and subscribers can each check an input without trusting your warehouse.
An input survives its vendor. The three vendor changes you absorbed this year would each have cost you nothing: an API change or an acquisition no longer puts a series’ evidence at risk.
Determinations become reproducible by a third party. “Here is our methodology” becomes “here is the arithmetic, run it yourself” — for any determination in the last twelve months, on 9 October or three years later.

The six checks, without the framing

The situation

This is going into the accounting pipeline and you are the one who has to make it true. What you need is which guarantee attaches to which artifact.

Each check below states what it proves. They are the same checks the role journeys run, with the narrative removed.

One call for the whole answer

Every other check on this page proves a property. This one is the product surface: a single HTTP request that returns the realized return over a period and both signed measurements it was derived from. No key, no account, no library.

curl "…"

What this proves the derived figure and the evidence for it arrive together, so a consumer never has to assemble a period from raw rows and never has to be told which two measurements a number came from.

Forge a measurement and watch it caught

A keypair is generated in this tab and used to re-sign a real measurement, then to reuse our genuine signature over a fabricated one. Both are put through the same checks.

What this proves the signature binds these exact bytes to our key, and the measurement to the round that key signed.

A figure fixed before the outcome existed

A measurement from weeks ago is re-checked here, then compared with what followed it.

What this proves the starting figure was signed on the day it describes, before the ending figure existed.

A return that states its own coverage

A 90-day return is requested, and the answer states the span it actually describes.

What this proves the figure carries the period it covers and the count of signed measurements behind it.

Integrity with the network switched off

The browser's own fetch is replaced with a function that refuses every call, and the integrity checks are re-run.

What this proves hash and signature verification need no network and no vendor at all.

When a value was produced, not when it was fetched

The gap between an oracle's accrual time and the block it was read at, measured on the live wstETH series.

What this proves each index carries the second its own source reports it accrued to.

The same value from unrelated nodes

The signed index is re-read at its own block from two endpoints with no relationship to us.

What this proves the signed value is the value the chain returns, at a node of your choosing.

What this looks like in your job

The pack that has to go out on Monday

What normally happens

With DeltaRates

The pack, as it arrived

What that is worth

Year-end, and investment income is a significant account

What normally happens

With DeltaRates

What that is worth

An allocator's due-diligence questionnaire, due on the 18th

What normally happens

With DeltaRates

What that is worth

Your IC meets on Thursday and the manager reported 3.67%

What normally happens

With DeltaRates

What that is worth

Your index review, and an auditor asking about input data

What normally happens

With DeltaRates

What that is worth

The six checks, without the framing

One call for the whole answer

Forge a measurement and watch it caught

A figure fixed before the outcome existed

A return that states its own coverage

Integrity with the network switched off

When a value was produced, not when it was fetched

The same value from unrelated nodes