Creditcrest Record

The file a lender keeps. Canonical serialisation, SHA-256, the hash travelling with the content, amendments that append rather than overwrite. It proves the content has not changed since it was sealed. It does not prove the content was true when written, and says so.

An example. Every figure in it is invented.

Every figure in this example is invented. There is no borrower, no bank statement and no lender: the transactions below were generated from the numbers in the form so that the policy engine has a statement to read, which is what it reads. The record below was sealed to answer this request and is not stored anywhere; a lender keeps theirs.

The credit policy replayed here is the one that says this at the top of itself. models/policies/example-au.txt is an example. Its thresholds are round numbers chosen to make the mechanism visible. They are not advice, not a market standard and not anybody’s policy. A real policy is written by a lender's credit risk function against their own book and signed off by people whose names go on it.

Sealed against round numbers — models/policies/example-au.txt. Written to make the mechanism visible: a $300 surplus floor, a 45 per cent debt service ceiling, thresholds a reader can do in their head while watching the replay work. Or run the same application against In the shape of a published guideline.

What a seal proves, and what it does not. The content is serialised in a canonical form — object keys sorted, all the way down — hashed with SHA-256, and the hash travels with the content. Anyone can recompute it and see whether the two agree. That proves the content has not been altered since it was sealed. It does not prove the content was true when it was written, and nothing here pretends otherwise.

Why not a database row. A row proves what the database says now. The question a lender is actually asked is what they had in front of them at the time.

Sealed

Record 1 — sealed 31 May 2026

Sealed at
31 May 2026 at 04:12 UTC
Algorithm
SHA-256
Applicant reference
EXAMPLE-0001
Product
Example unsecured personal loan — invented, not a product anybody sells
Supersedes
nothing — this is the first record in the chain

Hash: 4fa35da820dec1886b50e721aad863a09334bb038020a212aa1b2a28d0538dd8

What was asked, and what was answered

AskedAnsweredWhen
What do you spend each month, excluding rent and existing repayments?$1,400 a month.31 May 2026 at 03:50 UTC
Is there any commitment that does not appear on the statement supplied?No.31 May 2026 at 03:52 UTC

The conclusion, which is the assessor's and not the software's

Verdict
more_information — The assessor did not conclude either way and has asked for more information.
Reasons
The policy in force on 2026-05-31 was 2026.03. All 9 conditions in 2026.03 were met on the evidence supplied. 2 tests send this application to a person to look at. The referral tests that ask for a person are: dti-band, gambling. I have asked the applicant about the spending behind the referral and have not yet had an answer.
By
Dana Okafor
At
31 May 2026 at 04:10 UTC
The evidence this record carries
Statement period
5 December 2025 to 29 May 2026
Transactions read
77
Share identified
100%
Verified monthly income
$4,799.99
Of which from unnamed payers
$0.00
Monthly commitments
$1,200.00
Monthly outgoings
$2,958.47
Reading engine
1.0.0
Assessment rate
15.90%
Assessed income
$4,799.99
Living expenses
$2,700.00
Commitments
$1,200.00
Repayment on the proposed loan
$436.77
Monthly surplus
$463.22
Complete
yes
Serviceability engine
1.0.0
StepWhatAmountThe rule that produced it
income DIRECT CREDIT 049211 ACME LOGISTICS PAYROLL $4,799.99 Counted in full
expenses Living expenses -$2,700.00 The highest of the available figures, which here was the lender’s benchmark for this household
commitment RENT HARCOURTS ELSTERNWICK REF 88210 -$1,200.00 Found on the statement, 6 payments monthly
proposed The proposed personal loan -$436.77 Assessed at 15.90% — the offered rate plus a 3.0% buffer, with a 5.25% floor

The debt-to-income ratio this file was assessed on

Debt counted
$495,000.00
Gross annual income
$75,000.00
Debt to income
6.6 times
Where it came from
the figures typed into the form on this page, which is the lender’s own definition of what counts

Supplied rather than computed, and carried here for the same reason the policy hash is: a file whose debt-to-income rule decided the answer has to hold the ratio that rule read, or the outcome in it cannot be checked. It is inside the hash like everything else.

The declared expenses and the benchmark this file was assessed against

Declared each month
$1,400.00
The lender’s household benchmark
$2,700.00
Declared as a share of the benchmark
51.85%
Where they came from
the two figures typed into the form on this page: what the applicant declared, and the lender’s own household expenditure benchmark

Both supplied, like the ratio above. The assessment uses the HIGHEST of declared, benchmark and observed and reports one number; a policy that waives a minimum surplus where declared expenses reach a multiple of the benchmark is asking about these two separately, so the file carries them separately. This repository holds no benchmark table: the Household Expenditure Measure is licensed from the Melbourne Institute and the benchmark is the lender's.

The policy version this file was assessed under

Policy
example-unsecured-au
Version
2026.03
Effective
1 March 2026
Hash of the rules that ran
8400e4ccd45d6d4696e20f87c41cc6996ab7b9dc25bd429cc5c32316c4727caa
Conditions met
yes
What the run said
All 9 conditions in 2026.03 were met on the evidence supplied. 2 tests send this application to a person to look at.

Creditcrest Record has no named field for the policy version a file was assessed under, so this sits in the record's extra section. It is inside the hash like everything else, which is the part that matters: the hash above is the one decisioning.js computed over that version's own bytes, so which rules produced this file is provable rather than asserted.

Verified

It verifies

The content was re-serialised and re-hashed just now, and the result is compared with the hash the record was sealed with.

Sealed with
4fa35da820dec1886b50e721aad863a09334bb038020a212aa1b2a28d0538dd8
Recomputed now
4fa35da820dec1886b50e721aad863a09334bb038020a212aa1b2a28d0538dd8
Agree
yes
Reason
The content matches the hash it was sealed with.

Now edit it

The verification failing is the feature, so here it is failing. One field of the sealed record above is changed — nothing else — and the same check is run again.

Change something else and watch it again: Raise the monthly surplus by $1Lower the debt-to-income ratio by oneChange the assessor’s conclusionPut a different name on itBackdate the sealDelete the evidence section

Raise the monthly surplus by $1

The figure the whole assessment turns on, moved by the smallest amount that is still a change.

Before
assessment.monthlySurplus = 463.22
After
assessment.monthlySurplus = 464.22
Sealed with
4fa35da820dec1886b50e721aad863a09334bb038020a212aa1b2a28d0538dd8
Recomputed after the edit
0e7280ab499340a5a217a708ee71ed3d9b58de2a66d45fe5db390e10dc633c28
Agree
no
What the check says
The content does not match the hash it was sealed with, so it has changed since.

The edit was detected. A single character anywhere inside the content — a dollar on a surplus, a name, the second the file was sealed — changes the hash completely, which is the property being relied on.

And the chain, checked with the edited record in it

Record 1 (4fa35da820de): The content does not match the hash it was sealed with, so it has changed since.

The amendment below still verifies on its own after this edit, and that is the shape of the mechanism rather than a hole in it: each link proves its own content, and the chain is what proves the links belong together. It is checked as well, and it is the chain that reports the break.

The edited record in full, so the change can be found by eye

The same record, with one field changed

Sealed at
31 May 2026 at 04:12 UTC
Algorithm
SHA-256
Applicant reference
EXAMPLE-0001
Product
Example unsecured personal loan — invented, not a product anybody sells
Supersedes
nothing — this is the first record in the chain

Hash: 4fa35da820dec1886b50e721aad863a09334bb038020a212aa1b2a28d0538dd8

What was asked, and what was answered

AskedAnsweredWhen
What do you spend each month, excluding rent and existing repayments?$1,400 a month.31 May 2026 at 03:50 UTC
Is there any commitment that does not appear on the statement supplied?No.31 May 2026 at 03:52 UTC

The conclusion, which is the assessor's and not the software's

Verdict
more_information — The assessor did not conclude either way and has asked for more information.
Reasons
The policy in force on 2026-05-31 was 2026.03. All 9 conditions in 2026.03 were met on the evidence supplied. 2 tests send this application to a person to look at. The referral tests that ask for a person are: dti-band, gambling. I have asked the applicant about the spending behind the referral and have not yet had an answer.
By
Dana Okafor
At
31 May 2026 at 04:10 UTC
The evidence this record carries
Statement period
5 December 2025 to 29 May 2026
Transactions read
77
Share identified
100%
Verified monthly income
$4,799.99
Of which from unnamed payers
$0.00
Monthly commitments
$1,200.00
Monthly outgoings
$2,958.47
Reading engine
1.0.0
Assessment rate
15.90%
Assessed income
$4,799.99
Living expenses
$2,700.00
Commitments
$1,200.00
Repayment on the proposed loan
$436.77
Monthly surplus
$464.22
Complete
yes
Serviceability engine
1.0.0
StepWhatAmountThe rule that produced it
income DIRECT CREDIT 049211 ACME LOGISTICS PAYROLL $4,799.99 Counted in full
expenses Living expenses -$2,700.00 The highest of the available figures, which here was the lender’s benchmark for this household
commitment RENT HARCOURTS ELSTERNWICK REF 88210 -$1,200.00 Found on the statement, 6 payments monthly
proposed The proposed personal loan -$436.77 Assessed at 15.90% — the offered rate plus a 3.0% buffer, with a 5.25% floor

The debt-to-income ratio this file was assessed on

Debt counted
$495,000.00
Gross annual income
$75,000.00
Debt to income
6.6 times
Where it came from
the figures typed into the form on this page, which is the lender’s own definition of what counts

Supplied rather than computed, and carried here for the same reason the policy hash is: a file whose debt-to-income rule decided the answer has to hold the ratio that rule read, or the outcome in it cannot be checked. It is inside the hash like everything else.

The declared expenses and the benchmark this file was assessed against

Declared each month
$1,400.00
The lender’s household benchmark
$2,700.00
Declared as a share of the benchmark
51.85%
Where they came from
the two figures typed into the form on this page: what the applicant declared, and the lender’s own household expenditure benchmark

Both supplied, like the ratio above. The assessment uses the HIGHEST of declared, benchmark and observed and reports one number; a policy that waives a minimum surplus where declared expenses reach a multiple of the benchmark is asking about these two separately, so the file carries them separately. This repository holds no benchmark table: the Household Expenditure Measure is licensed from the Melbourne Institute and the benchmark is the lender's.

The policy version this file was assessed under

Policy
example-unsecured-au
Version
2026.03
Effective
1 March 2026
Hash of the rules that ran
8400e4ccd45d6d4696e20f87c41cc6996ab7b9dc25bd429cc5c32316c4727caa
Conditions met
yes
What the run said
All 9 conditions in 2026.03 were met on the evidence supplied. 2 tests send this application to a person to look at.

Creditcrest Record has no named field for the policy version a file was assessed under, so this sits in the record's extra section. It is inside the hash like everything else, which is the part that matters: the hash above is the one decisioning.js computed over that version's own bytes, so which rules produced this file is provable rather than asserted.

An amendment appends. It never overwrites.

When something turns out to be wrong — or, as here, when the question a referral test asked finally gets an answer — the fix is a new record naming the hash of the one it supersedes. The original stays exactly as it was. A record that could be edited in place would be worth precisely as much as no record.

  1. Record 1 — sealed 31 May 2026 at 04:12 UTC

    Sealed at
    31 May 2026 at 04:12 UTC
    Algorithm
    SHA-256
    Applicant reference
    EXAMPLE-0001
    Product
    Example unsecured personal loan — invented, not a product anybody sells
    Supersedes
    nothing — this is the first record in the chain

    Hash: 4fa35da820dec1886b50e721aad863a09334bb038020a212aa1b2a28d0538dd8

    What was asked, and what was answered

    AskedAnsweredWhen
    What do you spend each month, excluding rent and existing repayments?$1,400 a month.31 May 2026 at 03:50 UTC
    Is there any commitment that does not appear on the statement supplied?No.31 May 2026 at 03:52 UTC

    The conclusion, which is the assessor's and not the software's

    Verdict
    more_information — The assessor did not conclude either way and has asked for more information.
    Reasons
    The policy in force on 2026-05-31 was 2026.03. All 9 conditions in 2026.03 were met on the evidence supplied. 2 tests send this application to a person to look at. The referral tests that ask for a person are: dti-band, gambling. I have asked the applicant about the spending behind the referral and have not yet had an answer.
    By
    Dana Okafor
    At
    31 May 2026 at 04:10 UTC
    The evidence this record carries
    Statement period
    5 December 2025 to 29 May 2026
    Transactions read
    77
    Share identified
    100%
    Verified monthly income
    $4,799.99
    Of which from unnamed payers
    $0.00
    Monthly commitments
    $1,200.00
    Monthly outgoings
    $2,958.47
    Reading engine
    1.0.0
    Assessment rate
    15.90%
    Assessed income
    $4,799.99
    Living expenses
    $2,700.00
    Commitments
    $1,200.00
    Repayment on the proposed loan
    $436.77
    Monthly surplus
    $463.22
    Complete
    yes
    Serviceability engine
    1.0.0
    StepWhatAmountThe rule that produced it
    income DIRECT CREDIT 049211 ACME LOGISTICS PAYROLL $4,799.99 Counted in full
    expenses Living expenses -$2,700.00 The highest of the available figures, which here was the lender’s benchmark for this household
    commitment RENT HARCOURTS ELSTERNWICK REF 88210 -$1,200.00 Found on the statement, 6 payments monthly
    proposed The proposed personal loan -$436.77 Assessed at 15.90% — the offered rate plus a 3.0% buffer, with a 5.25% floor

    The debt-to-income ratio this file was assessed on

    Debt counted
    $495,000.00
    Gross annual income
    $75,000.00
    Debt to income
    6.6 times
    Where it came from
    the figures typed into the form on this page, which is the lender’s own definition of what counts

    Supplied rather than computed, and carried here for the same reason the policy hash is: a file whose debt-to-income rule decided the answer has to hold the ratio that rule read, or the outcome in it cannot be checked. It is inside the hash like everything else.

    The declared expenses and the benchmark this file was assessed against

    Declared each month
    $1,400.00
    The lender’s household benchmark
    $2,700.00
    Declared as a share of the benchmark
    51.85%
    Where they came from
    the two figures typed into the form on this page: what the applicant declared, and the lender’s own household expenditure benchmark

    Both supplied, like the ratio above. The assessment uses the HIGHEST of declared, benchmark and observed and reports one number; a policy that waives a minimum surplus where declared expenses reach a multiple of the benchmark is asking about these two separately, so the file carries them separately. This repository holds no benchmark table: the Household Expenditure Measure is licensed from the Melbourne Institute and the benchmark is the lender's.

    The policy version this file was assessed under

    Policy
    example-unsecured-au
    Version
    2026.03
    Effective
    1 March 2026
    Hash of the rules that ran
    8400e4ccd45d6d4696e20f87c41cc6996ab7b9dc25bd429cc5c32316c4727caa
    Conditions met
    yes
    What the run said
    All 9 conditions in 2026.03 were met on the evidence supplied. 2 tests send this application to a person to look at.

    Creditcrest Record has no named field for the policy version a file was assessed under, so this sits in the record's extra section. It is inside the hash like everything else, which is the part that matters: the hash above is the one decisioning.js computed over that version's own bytes, so which rules produced this file is provable rather than asserted.

  2. Record 2 — sealed 6 June 2026 at 23:41 UTC

    Sealed at
    6 June 2026 at 23:41 UTC
    Algorithm
    SHA-256
    Applicant reference
    EXAMPLE-0001
    Product
    Example unsecured personal loan — invented, not a product anybody sells
    Supersedes
    4fa35da820dec1886b50e721aad863a09334bb038020a212aa1b2a28d0538dd8

    Hash: 41921db1b96f1a641fb22db835f5b874d2ebd4a65c2859fc685df62484ac3b76

    Why this amendment was made

    The applicant answered the question the referral test asked, and the conclusion follows from the answer.

    What was asked, and what was answered

    AskedAnsweredWhen
    What do you spend each month, excluding rent and existing repayments?$1,400 a month.31 May 2026 at 03:50 UTC
    Is there any commitment that does not appear on the statement supplied?No.31 May 2026 at 03:52 UTC
    What is the regular payment to the wagering operator for?A syndicate the applicant contributes to fortnightly; they say it is capped and have agreed to close it before settlement.6 June 2026 at 23:20 UTC

    The conclusion, which is the assessor's and not the software's

    Verdict
    not_unsuitable — The assessor concluded the credit contract is not unsuitable for this applicant.
    Reasons
    The applicant has answered the question about the wagering payments and has agreed to close the arrangement. On that answer and on the assessment in this file I conclude the contract is not unsuitable. This is my conclusion and not the software’s.
    By
    Dana Okafor
    At
    6 June 2026 at 23:41 UTC
    The evidence this record carries
    Statement period
    5 December 2025 to 29 May 2026
    Transactions read
    77
    Share identified
    100%
    Verified monthly income
    $4,799.99
    Of which from unnamed payers
    $0.00
    Monthly commitments
    $1,200.00
    Monthly outgoings
    $2,958.47
    Reading engine
    1.0.0
    Assessment rate
    15.90%
    Assessed income
    $4,799.99
    Living expenses
    $2,700.00
    Commitments
    $1,200.00
    Repayment on the proposed loan
    $436.77
    Monthly surplus
    $463.22
    Complete
    yes
    Serviceability engine
    1.0.0
    StepWhatAmountThe rule that produced it
    income DIRECT CREDIT 049211 ACME LOGISTICS PAYROLL $4,799.99 Counted in full
    expenses Living expenses -$2,700.00 The highest of the available figures, which here was the lender’s benchmark for this household
    commitment RENT HARCOURTS ELSTERNWICK REF 88210 -$1,200.00 Found on the statement, 6 payments monthly
    proposed The proposed personal loan -$436.77 Assessed at 15.90% — the offered rate plus a 3.0% buffer, with a 5.25% floor

    The debt-to-income ratio this file was assessed on

    Debt counted
    $495,000.00
    Gross annual income
    $75,000.00
    Debt to income
    6.6 times
    Where it came from
    the figures typed into the form on this page, which is the lender’s own definition of what counts

    Supplied rather than computed, and carried here for the same reason the policy hash is: a file whose debt-to-income rule decided the answer has to hold the ratio that rule read, or the outcome in it cannot be checked. It is inside the hash like everything else.

    The declared expenses and the benchmark this file was assessed against

    Declared each month
    $1,400.00
    The lender’s household benchmark
    $2,700.00
    Declared as a share of the benchmark
    51.85%
    Where they came from
    the two figures typed into the form on this page: what the applicant declared, and the lender’s own household expenditure benchmark

    Both supplied, like the ratio above. The assessment uses the HIGHEST of declared, benchmark and observed and reports one number; a policy that waives a minimum surplus where declared expenses reach a multiple of the benchmark is asking about these two separately, so the file carries them separately. This repository holds no benchmark table: the Household Expenditure Measure is licensed from the Melbourne Institute and the benchmark is the lender's.

    The policy version this file was assessed under

    Policy
    example-unsecured-au
    Version
    2026.03
    Effective
    1 March 2026
    Hash of the rules that ran
    8400e4ccd45d6d4696e20f87c41cc6996ab7b9dc25bd429cc5c32316c4727caa
    Conditions met
    yes
    What the run said
    All 9 conditions in 2026.03 were met on the evidence supplied. 2 tests send this application to a person to look at.

    Creditcrest Record has no named field for the policy version a file was assessed under, so this sits in the record's extra section. It is inside the hash like everything else, which is the part that matters: the hash above is the one decisioning.js computed over that version's own bytes, so which rules produced this file is provable rather than asserted.

Read the two conclusions above side by side

Record 1, 31 May 2026 at 04:10 UTC

more_information — The policy in force on 2026-05-31 was 2026.03. All 9 conditions in 2026.03 were met on the evidence supplied. 2 tests send this application to a person to look at. The referral tests that ask for a person are: dti-band, gambling. I have asked the applicant about the spending behind the referral and have not yet had an answer.

Record 2, 6 June 2026 at 23:41 UTC

not_unsuitable — The applicant has answered the question about the wagering payments and has agreed to close the arrangement. On that answer and on the assessment in this file I conclude the contract is not unsuitable. This is my conclusion and not the software’s.

The first one is still there, with its own timestamp, and it says what it said on the day. Nothing was rewritten: the second record carries the hash of the first in its supersedes field, which is what makes the pair a history rather than two files.

The chain, checked

Records in the chain
2
Each still hashes to its own id
yes
Each points at the one before it
yes
Problems
none

What the software will not write

The conclusion. A record has a place for one and holds whatever the assessor put there, attributed to them by name. This engine does not produce it, does not suggest it, and refuses to seal a record whose conclusion claims to have come from the software — engine, system, model, AI assistant, generated and a dozen other spellings are all refused, because a person's name on the conclusion is the one thing keeping this on the right side of a line that a competitor selling auto-generated file notes has already crossed.

An approval. The four verdicts a record may carry are not_unsuitable, unsuitable, more_information and withdrawn, because approval is not the question the law asks.

The assessor named in the example above, Dana Okafor, does not exist. Every figure in this example is invented. There is no borrower, no bank statement and no lender: the transactions below were generated from the numbers in the form so that the policy engine has a statement to read, which is what it reads.