The Audit Evidence Problem Public Ledgers Were Supposed to Solve
Why the useful accounting question is not “can we put records on-chain?” but “can an auditor inspect one genuine field, verify its prior commitment, and avoid exposing everything else?”
Keywords
Audit evidence; tax assurance; accounting information systems; selective disclosure; public ledger; documentary evidence; Merkle commitment; field-level verification; workpapers; control dependency; confidentiality; digital audit; tax examination; verifiable records; design science.
The post…
The most common mistake in blockchain accounting is to begin with the ledger. That mistake usually leads to one of two dead ends. In the first, someone proposes putting accounting records on a public chain, as though commercial confidentiality, tax privacy, supplier relationships, pricing data, payroll detail, and internal control evidence could simply be made public because immutability sounds attractive. In the second, someone proposes hiding the records completely and proving only abstract facts about them, as though audit evidence were normally a predicate about values no one has inspected.
Both approaches miss the accounting problem.
Audit and tax examination are not usually exercises in admiring ledgers. They are exercises in evidence. An auditor, tax authority, forensic accountant, or internal-control reviewer does not normally ask, in the abstract, whether a database exists. The examiner asks for a particular field. What was the tax amount on this invoice? What date was used? What counterparty was recorded? What amount was posted? What code was applied? What movement explains the change in the reconciliation? What figure was present when the transaction was recorded, and is the figure being shown now the same figure that existed then?
Those questions are granular. They are field-level questions. They do not require the whole accounting record to become public. They do not require the examiner to receive every line of an invoice when the tax amount alone is under examination. They also do not require the examiner to accept a cryptographic proof about hidden values where the audit objective is to inspect the actual value. They require something more precise: selective production of the genuine field under examination, together with independent verification that the field belongs to a record previously committed at a known time.
That is the accounting evidence problem public ledgers can help solve.
The useful claim is not that a public ledger proves a sale occurred. It does not. A ledger commitment does not show that goods were delivered, that a receivable is collectible, that a counterparty is independent, that management had the right intent, that a valuation is fair, or that a transaction has commercial substance. It does not prevent fraud. It does not settle legal admissibility. It does not transform a false record into a true one. A committed falsehood remains false.
The useful claim is narrower and stronger: a disclosed accounting field can be checked against a prior public commitment, so that its integrity and timing become independently verifiable while other fields remain confidential.
That is a different kind of accounting evidence. It is not a new cryptographic primitive. Merkle commitments, inclusion proofs, hashing, and public anchoring are established techniques. The novelty lies in the accounting architecture: making the unit of commitment and disclosure correspond to the unit of audit interest. The field the examiner wants to inspect becomes the field the system can produce, prove, and document in the workpaper.
The difference matters because accounting evidence is not merely data. Evidence is data used in a conclusion. The evidential value of an accounting field depends on its relation to a record, a time, a control environment, a source system, and a professional judgement. A number copied into a spreadsheet may be useful, but it is weak if its history is mutable and unverifiable. A full invoice may be more informative, but it can disclose more than the examiner needs. A zero-knowledge proof may establish a mathematical statement, but it does not produce the value where inspection of the value is the audit objective. A selectively disclosed field with an inclusion proof does something different: it gives the examiner the actual field and a way to verify that the field is the one previously committed.
That is the shift.
The system begins with an ordinary accounting transaction. An invoice, journal entry, payment record, tax calculation, or reconciliation movement is represented as an ordered set of fields. Each field has a tag and a value. A field might be “invoice date,” “counterparty identifier,” “tax code,” “net amount,” “tax amount,” “gross amount,” “payment reference,” or “write-off amount.” Each field is encoded in a canonical way, combined with a per-field nonce, and hashed. The per-field nonce is important because many accounting fields are low entropy. Dates, tax codes, common amounts, and repeated business identifiers may be guessable. Without a nonce, an observer might try dictionary attacks against public commitments. With a nonce withheld until disclosure, the commitment does not allow simple confirmation of guessed values.
The field hashes are then combined into a Merkle tree. The root of that tree commits to the entire field set. Only the root is anchored on a public commitment medium. The underlying field values remain off the public medium. Later, when an examiner asks for one field or a small group of fields, the record-holder discloses the field value, the nonce, and a compact inclusion proof. The examiner recomputes the field hash, folds the proof up to the committed root, and checks that the root was anchored on the public medium. If the verification succeeds, the examiner knows that the disclosed field is the same field committed earlier.
The rest of the transaction remains undisclosed.
This is not privacy by pretending the public ledger is private. It is privacy by committing publicly and disclosing selectively. That distinction is essential. A public ledger is useful as a timestamped, independently verifiable commitment medium. It is not an appropriate place to publish ordinary commercial records. The contribution is to use the public medium for what it can do well: preserve evidence of prior commitment and ordering. The accounting records themselves remain within the entity and are released only to the extent required by the engagement, audit request, tax examination, or control procedure.
This also avoids a second error: treating hidden-value cryptography as a substitute for audit evidence. Zero-knowledge proofs and homomorphic methods can be powerful where the audit objective is a predicate. If the question is whether a total equals the sum of concealed components, or whether a value falls in a permitted range, a hidden-value proof may be appropriate. But many audit and tax questions require inspection of the actual value. The examiner must see the amount, the code, the date, or the counterparty field and then decide what further procedures are necessary. A proof about a concealed value is not the same as producing the value.
Selective disclosure therefore occupies an important middle ground. It avoids full-record disclosure while still producing the actual record element. It gives the examiner something inspectable and independently verifiable.
Consider a tax authority examining a sales invoice. The authority does not necessarily need the entire commercial record. It may need the tax amount, tax code, date, and gross amount. Under a conventional system, the taxpayer produces an invoice extract or source document, and the authority relies on ordinary document inspection, system controls, and substantive procedures. Under a public-ledger commitment architecture, the taxpayer can produce the relevant fields and a proof that those fields are part of a record previously committed. The tax authority can verify that the fields were not invented after the request, while still not receiving unrelated line items, pricing detail, or other commercially sensitive data.
That does not prove the sale occurred. It does not prove delivery. It does not prove the tax treatment is correct. It does not prove that the customer exists or that the invoice is enforceable. It proves something narrower: the field being inspected is the field that was committed, and the commitment existed at or before the anchoring time. In tax and audit work, that can be valuable because many disputes begin with the integrity and timing of records. The examiner still performs residual procedures for substance, authorisation, collectability, completeness, and legal character.
This is why the architecture should be understood as an evidential layer, not an audit replacement. It strengthens the record presented to the examiner. It does not replace professional judgement. It does not decide whether the figure should be accepted. It helps answer whether the figure now being shown is the same figure previously committed.
The same point applies to audit workpapers. A workpaper does not simply record that a number was seen. It records what was tested, what evidence was obtained, what assertion was addressed, what limitations remained, and what further procedures were performed. A selective-disclosure proof bundle is useful only if it can be documented in that form. A sufficient workpaper should identify the field, the disclosed value, the nonce, the inclusion proof, the committed root, the public-medium reference, the verifier output, the timestamp or ordering evidence, the assertion supported, and the residual procedures that remain necessary. Without those elements, the proof risks becoming a technical artefact detached from audit reasoning. With them, it becomes part of an accountable evidential chain.
The workpaper framing also prevents overclaiming. The most dangerous misuse of this system would be treating a successful proof as evidence that the underlying transaction occurred. It is not. A committed invoice field proves fidelity to the committed record, not fidelity to economic reality. That boundary must be central. It is not a weakness to state it. It is the condition that makes the contribution credible.
Accounting already distinguishes between different assertions. Existence, occurrence, accuracy, completeness, valuation, rights and obligations, cut-off, presentation, and disclosure do not collapse into one another. A field-level commitment can support some of these more directly than others. It is strongest for integrity and timing of committed fields. It can support arithmetic accuracy when disclosed fields are recomputed. It can contribute to existence or occurrence of the committed record, meaning the record existed in committed form at a relevant time. It does not establish occurrence in commercial substance. It contributes to completeness only if the committed population is itself shown complete through controls, reconciliation, or source-system assurance. It does not solve valuation, collectability, related-party status, delivery, management intent, or legal enforceability.
This disciplined mapping is what makes the architecture an accounting contribution rather than a technology demonstration.
A public commitment medium also introduces a control dependency. That dependency must be assessed, not assumed. The medium must provide public verifiability, durable commitment carriage, timestamping or ordering, settlement finality, reorganisation resistance, and accessible verification. Its governance, availability, fees, and operational risks matter. In the reference implementation, a Bitcoin BSV environment is used because it supports low-cost script-carried commitments at scale. But the theoretical contribution is not “use BSV for accounting.” The theoretical contribution is medium-independent: any public commitment medium satisfying the relevant control requirements could be assessed for the same evidential role. BSV is an implementation case, not the theory.
That distinction matters because ledger choice can distract reviewers and practitioners from the evidence architecture. A sceptical reader may have strong views about particular networks, governance histories, reputational issues, or technical assumptions. The correct response is not to deny that these concerns exist. The correct response is to treat the public medium as a control dependency. If the medium fails the engagement’s requirements for finality, availability, governance, cost, or verifiability, then the deployment fails for that engagement. The accounting contribution remains the same: field-level verifiable evidence requires a medium that can carry and preserve public commitments. Which medium satisfies that requirement is an implementation and assurance question.
The architecture also clarifies the difference between an internal system assertion and an independently verifiable record property. Conventional accounting systems can produce reports and extracts. Those may be reliable when system controls are effective. But the record’s integrity remains heavily dependent on the entity’s mutable systems. A public commitment provides an external anchor. It does not eliminate the need for controls over the source system, encoding process, access management, change management, or reconciliation. It adds a layer of evidence outside the entity’s unilateral control. That is why it is best thought of as a sidecar: it sits beside the ERP or accounting system and creates an evidential commitment at the point of record creation, payment, or reconciliation.
This sidecar model is important for adoption. The system does not require replacing the ledger, accounting software, or audit methodology. It does not require every trading partner to adopt the same infrastructure before any value is realised. A single entity can commit its own records and later produce selective disclosures to auditors, tax authorities, internal reviewers, or counterparties. Bilateral adoption adds further value because both parties can independently commit corresponding fields and an examiner can compare them. But the first use case is simpler: an entity creates independently verifiable evidence over its own records.
The most persuasive example is a VAT, GST, or sales-tax workflow. A supplier issues an invoice. The invoice contains customer details, product lines, quantities, prices, discounts, tax codes, net amounts, tax amounts, and gross totals. The system commits the fields at issue. Later, a tax authority requests the tax amount and tax code for a set of invoices. The entity discloses only those fields and their inclusion proofs. The authority verifies that the disclosed fields match the prior commitments. It can then test tax arithmetic and perform whatever residual procedures are needed for classification, exemption, business purpose, and compliance. The authority obtains stronger evidence of field integrity without receiving the full commercial record.
This is proportionality in tax examination. The state gets the field it needs. The taxpayer avoids unnecessary disclosure. Both parties can verify that the field was not altered after the fact. That is a practical contribution.
The same structure applies in external audit. Suppose an auditor is testing accounts receivable. The auditor may request invoice totals, payment references, credit notes, write-offs, and dates. The system can disclose selected fields across a population and allow recomputation of a roll-forward. The auditor can verify that each disclosed field belongs to a prior commitment. This can strengthen evidence over the integrity of the committed movement data. It does not prove collectability. It does not prove the debtor’s confirmation. It does not prove subsequent cash receipt unless those fields are separately evidenced. The auditor still performs confirmation, subsequent receipt testing, ageing analysis, dispute review, and collectability procedures where required.
The value is not that audit work disappears. The value is that some evidence becomes harder to alter and easier to verify independently.
This also matters in forensic accounting. A forensic accountant often cares about whether a record was altered, backdated, substituted, or selectively reconstructed after a dispute began. Field-level commitments can help identify whether the field presented now was present in committed form earlier. They can also support exception investigation by showing which fields changed or which records cannot be verified against prior commitments. But forensic work is especially sensitive to overclaiming. A record can be committed falsely at origin. A fraudster can commit a false invoice just as easily as a truthful one if upstream controls fail. Therefore, the system provides integrity and timing evidence, not fraud exclusion.
This is why the “garbage in” boundary must remain explicit. No cryptographic commitment proves the world. It proves consistency with a committed input. Accounting and audit are concerned with the relationship between records and economic reality. Cryptography can strengthen the record side of that relationship. It cannot, by itself, establish the reality side.
A good design-science accounting paper must therefore do more than show that the code works. It must show how the artefact maps to accounting concepts. It must state design requirements, provide implementation evidence, explain control dependencies, identify failure modes, and translate technical outputs into audit implications. A useful evaluation must connect technical tests to accounting assertions. If a tampered field fails verification, what assertion does that support? If a proof-store withholds a fragment, what is the audit consequence? If the public medium reorganises before settlement depth, what should the auditor do? If the committed population is incomplete, what controls or reconciliations are necessary? If the verifier output is placed in the workpaper, what residual procedure remains?
These are not side questions. They are the accounting contribution.
The structured expert benchmark reinforces this point. Six expert profiles — audit partner, tax compliance reviewer, AIS academic, IT audit manager, forensic accountant, and internal controls specialist — were used to test expected interpretation of the evidence bundle. The benchmark asks whether qualified readers should correctly identify the supported assertion, reject unsupported claims, identify residual procedures, judge workpaper sufficiency, recognise the confidentiality benefit, and flag overclaim risk. The pattern is instructive. The core evidential contribution is straightforward: experts should understand that the proof supports field fidelity, timing, and integrity. The confidentiality benefit is also straightforward: selective disclosure reduces unnecessary exposure. The main risk is over-reliance, especially around completeness and economic substance. That is exactly where the manuscript must remain explicit.
This kind of benchmark does not turn the paper into a large behavioural study. It is not a claim of statistical generalisation. It is a structured validation of interpretability: can knowledgeable accounting readers understand what the evidence does and does not prove? The answer matters because the system is only useful if auditors and tax examiners can use it without overclaiming. A technically correct proof can still be dangerous if junior staff treat it as substantive assurance over the underlying transaction. The manuscript’s response is to keep the “can/cannot verify” boundary central and to require residual-procedure documentation in the workpaper.
The most important disagreement pattern concerns completeness. If an entity commits selected records, the proof can show that a disclosed field belongs to one committed record. It cannot show that all relevant records were committed. Completeness requires population controls. The auditor must understand how records enter the commitment process, whether every relevant transaction is included, whether exceptions are logged, whether omitted records can be detected, and whether the committed population reconciles to source systems and financial statement balances. Without that, field-level proof may be strong for individual records and weak for population completeness.
Economic substance is the second major boundary. A committed invoice field does not prove delivery, business purpose, enforceability, or arm’s-length status. The auditor must still inspect contracts, delivery evidence, counterparty confirmations, subsequent receipts, related-party indicators, and management representations where relevant. The field commitment tells the auditor that the disclosed field is the committed field. It does not tell the auditor that the transaction should be recognised.
The public-medium dependency is the third boundary. IT audit specialists will properly ask how the medium is assessed. What finality depth is required? What happens if there is a reorganisation? How are verifier outputs retained? How are public-medium references documented? What governance risks exist? What fees apply at volume? What if the medium changes or ceases to satisfy requirements? These questions are not objections to the architecture. They are part of the architecture. A deployment must document them in the same spirit as service-organisation assessment.
The fourth boundary is workpaper sufficiency. A proof result alone is not enough. The workpaper must include field identity, disclosed value, nonce, inclusion path, committed root, public-medium reference, verifier version, timestamp or ordering evidence, and residual-procedure note. Without this, a future reviewer cannot understand what was tested or why the conclusion was limited. A useful verifier must therefore produce audit-friendly output, not merely a technical pass/fail result.
The fifth boundary is over-reliance. Any technology that produces a clean green tick risks being misunderstood. This is a problem in audit automation generally, not only in public-ledger evidence. The more elegant the proof, the more explicit the boundary must be. The proof is corroborative documentary evidence. It is not a substitute for professional judgement.
The architecture also faces real-world edge cases. Partial disclosures may occur across multiple transactions. Consolidated groups may commit records through different systems. Corrections and reversals must be retained rather than overwritten. Assigned or factored receivables may require linking an original commitment to a later transfer. A tax authority may request bulk fields across a population. A medium may need to be migrated if it no longer satisfies the control requirements. These cases do not invalidate the architecture, but they require deployment procedures. The important point is to treat them as accounting and control-design problems, not merely technical exceptions.
Corrections provide a useful example. If a field was committed incorrectly, the correct response is not to erase the old commitment. The old commitment remains evidence of what was recorded. A correcting commitment should be made and linked to the prior one. The audit trail then contains both the error and the correction. This is consistent with accounting practice: errors are corrected through entries and documentation, not by pretending the original record never existed.
Assignment and factoring provide another example. If a receivable is transferred, the assignee may need to rely on fields from the original invoice and assignment document. A selective-disclosure architecture can produce the relevant fields and link them to commitments. But it does not prove the legal validity of the assignment. That remains a legal and substantive question. Again, the architecture supports evidence, not judgement.
Medium migration provides a third example. If the reference public medium no longer satisfies the engagement’s control requirements, future commitments may move to another medium. Prior commitments remain verifiable if the old medium and verification data remain available. A migration procedure should document the old roots, new roots, cross-references, and verification continuity. The theory does not require loyalty to one chain. It requires verifiable public commitment.
These edge cases show why the public-ledger framing must remain medium-independent. The commitment medium is infrastructure. The accounting contribution is the evidential relationship between field, commitment, disclosure, verification, workpaper, and residual procedure.
The paper’s broader claim is therefore modest in scope but significant in practice. It does not promise a new audit world in which cryptography replaces judgement. It promises a better evidence object: a field that can be inspected, verified against a prior commitment, and documented with its limitations. That object sits between ordinary internal records and full external confirmations. It is stronger than a mutable extract because integrity and timing are independently checkable. It is narrower than a confirmation because it does not involve a counterparty attesting to substance. It is less intrusive than full-record disclosure because unrelated fields remain private. It is more audit-relevant than a hidden-value proof when the examiner needs to inspect the actual value.
That is the contribution.
There is also a governance lesson. Accounting systems increasingly produce digital evidence, but not all digital evidence is equal. Some evidence is merely digital because it is stored electronically. Some is system-generated but mutable. Some is externally corroborated. Some is cryptographically committed. The profession needs better categories for these differences. A field-level public commitment creates a distinct category: independently verifiable documentary evidence of a committed record field. It does not dominate other evidence categories. It complements them.
This category can matter in tax administration because tax authorities often face a proportionality problem. They need evidence but should avoid unnecessary collection of commercial detail. Selective disclosure allows a narrower request. It also creates a cleaner dispute record: the authority can say exactly which field was requested, what was disclosed, how it verified, and what additional issues remain. That is useful both for compliance and for taxpayer protection.
It can matter in audit because auditors increasingly rely on system outputs, APIs, automated controls, and digital workpapers. A verifiable field commitment can strengthen reliance on selected records, especially where audit procedures involve large populations of similar fields. It can also support exception-based testing: records that fail verification, lack proofs, or fall outside settlement requirements become exceptions for further investigation.
It can matter in internal audit because control teams can use commitments to test whether records have been altered after key process points. For example, invoice fields committed at issuance can later be compared against disclosed fields during exception review. Payment fields committed at settlement can be reconciled against bank and ledger data. Control testing can focus on discrepancies, missing proofs, late commitments, and unauthorised corrections.
It can matter in forensic work because timing and integrity are often contested. A public commitment does not prove truth, but it can narrow the factual dispute. Was this field present in this form at that time? Was the record later substituted? Does the disclosed value match the prior commitment? These questions are not the whole forensic case, but they are often important.
The objection that the cryptography is not new is correct but incomplete. Accounting research often contributes by integrating established techniques into accounting settings in a way that changes evidence, control, measurement, or decision-usefulness. The novelty is not the hash function. It is the accounting object created by field-level commitment and selective disclosure. The question is not whether Merkle proofs existed before. They did. The question is whether accounting evidence has been designed around the field as the unit of commitment, inspection, confidentiality, verification, and workpaper documentation. That is the contribution.
The objection that practitioner validation is limited is also fair. A full behavioural study would be valuable. It should test whether auditors and tax examiners correctly interpret proof bundles, whether they over-rely, whether workpaper quality improves, and whether confidentiality benefits affect evidence requests. But the absence of a full behavioural study does not erase the design-science contribution. It defines the next stage. The current artefact establishes construct validity and technical validity: the architecture is specified, implemented, tested, reproduced, mapped to assertions, and translated into workpaper logic. Practitioner reliance is the next empirical question.
The objection that the reference implementation uses a particular public ledger is also fair. That is why the theory must remain medium-independent. A medium is acceptable only if it satisfies the public-commitment requirements for the engagement. A reviewer may dislike BSV, Ethereum, a permissioned notarisation log, or any other medium. That is not the point. The point is to specify what the medium must provide and how an auditor assesses it. The implementation demonstrates feasibility; it does not define the boundary of the theory.
The objection that real-world deployment is messier is also fair. Real accounting systems involve partial records, corrections, consolidation, tax adjustments, multiple entities, access controls, system migrations, and organisational failure. But that is not a reason to avoid the architecture. It is a reason to design the deployment around controls and residual procedures. Field-level verifiable evidence is one layer in an assurance environment. It is not the environment itself.
The best way to summarise the idea is this: the public ledger should not contain the accounting record. It should contain a commitment to the record. The auditor should not receive everything. The auditor should receive the field needed for the audit question. The proof should not be treated as proof of the world. It should be treated as proof that the field produced is the field committed. The workpaper should not say “blockchain verified the transaction.” It should say “the disclosed field was verified against a prior public commitment; residual procedures were performed for the assertions not supported by that verification.”
That wording is less dramatic than most blockchain claims. It is also far more useful.
A mature accounting treatment of public ledgers should be boring in the right way. It should specify the evidence object. It should map to assertions. It should identify controls. It should state residual work. It should document limitations. It should avoid technological overclaiming. It should preserve confidentiality. It should give auditors and tax authorities something they can actually use.
Field-level selective disclosure does that. It does not make accounting trustless. Accounting cannot be trustless because records must still represent economic events, controls must still operate, and professionals must still judge. But it can make parts of the record independently verifiable. It can reduce unnecessary disclosure. It can make tampering easier to detect. It can make workpapers more precise. It can make tax requests more proportional. It can make the evidential boundary clearer.
The future of digital audit evidence will not be a single universal ledger containing all records. That would be commercially naïve and legally reckless. It will be layered. Private systems will hold records. Public or independently controlled media will hold commitments. Examiners will receive selective disclosures. Verification tools will produce workpaper-ready outputs. Controls will govern the encoding and commitment process. Substantive procedures will continue to test reality.
That is the sensible direction.
The central accounting question is not whether a transaction is “on-chain.” The central question is what evidence an examiner can rely on, what remains undisclosed, what residual procedures are necessary, and how the conclusion is documented. A public commitment is valuable only when it improves that evidence chain.
The field-level architecture offers a practical answer. It makes the specific accounting field — the amount, date, tax code, counterparty identifier, or reconciliation movement — the object of verifiable evidence. It allows the examiner to inspect the genuine field. It allows the entity to preserve confidentiality over everything else. It allows the workpaper to record both the verification and the limits of the verification.
That is not a revolution in audit judgement. It is a disciplined improvement in audit evidence.
And that is exactly what public-ledger accounting should have been aiming at all along.