Visible Is Not Yet Verifiable

The original study asked whether decentralized autonomous organizations offer a more accountable architecture for Canadian public administration. That question presumed more commensurability than the evidence could bear. The three cases are not three implementations of one institutional function. The Canada Council is a statutory Crown corporation administering grants and prizes through program rules, peer assessment, delegated financial authority, and public law. Polkadot OpenGov is a protocol governance system with track-specific referenda and conviction voting (Polkadot n.d.). Gitcoin Grants 6–15 were historical funding rounds operated through a changing platform and mechanism. Their public records reveal different slices of different decision processes.

The reconstructed question is consequently prior to institutional evaluation:

What propositions remain warranted once each public record is tested for semantic validity, population stability, computational reproducibility, source identity, and legal relevance?

A comparison earns its meaning only after each record passes that test. If one column counts award records, another referendum-level totals, and another dashboard bins of uncertain provenance, a shared statistic does not create a shared object.

Bovens (2007) treats accountability as a relationship in which an actor must explain and justify conduct before a forum that can question, judge, and impose consequences. Traceability may support that relationship, but it does not contain it. A record can be durable yet unintelligible, complete at the transaction layer yet incomplete at the decision layer, public yet practically inaccessible, or auditable without identifying anyone who must answer for the result.

The distinction is well established. Fox (2007) separates transparency from accountability, and Meijer (2009) shows that computer-mediated visibility changes rather than dissolves interpretive problems. Ananny and Crawford (2018) warn against equating exposure with knowledge, while Tan, Mahula, and Crompvoets (2022) locate blockchain accountability across technical and institutional levels. The phrase visible is not yet verifiable therefore does not claim novelty as a general theory. Its narrower contribution is an evidence discipline for cross-system empirical work.

Four questions govern that discipline:

The last distinction is essential for code-driven claims. Repository source can establish that a capability exists in a particular commit. It does not, by itself, establish that the code entered a release, that the release was deployed at a particular block, or that a particular extrinsic executed through it. Those propositions require progressively stronger evidence: release metadata and artifact hashes; runtime or contract identity at an anchored block; then transaction, event, state-transition, and receipt evidence.

The Gitcoin branch contains imported Dune CSV outputs and a derived summary. Its own metadata states that the outputs are captured but not query-replay certified. The underlying SQL, Dune execution identifiers and timestamps, contribution-level official export, and preprocessing inputs required by the analysis script are absent. A clean run from raw records to the published statistics is not presently possible.

Several earlier interpretations also crossed units:

These quantities are quarantined rather than reinterpreted. Historical Grants rounds also should not be used as a timeless proxy for Gitcoin’s later protocol architecture. A publishable branch would require versioning each round definition, obtaining official raw contribution records or reconstructing them at chain level, supplying the query text with its execution metadata, mapping chains, contracts, and deployments, replaying results at transaction level, and stating an explicit rule for identity and sybil treatment.

The former TI/C/S instrument assigned decimal scores to traceability, contestability, sanctionability, and adaptive capacity, then averaged them into system rankings. Decimal places implied a precision the instrument’s validity could not support.

The components mixed unlike functions. A follow-up blockchain proposal is not an applicant appeal. Automated execution is not itself a sanction. A pseudonymous address does not necessarily create an identifiable duty to answer. Conversely, the Canada Council publishes criteria and assessor names, provides score information in its current process, permits review where a procedural error may have occurred, and uses reductions, Major Warnings, Concerned Status, cancellation, and other compliance measures. The prior rubric discounted those institutions while awarding full values to on-chain visibility and reversibility.

The scoring procedure had no validated anchors, independent coders, inter-rater reliability, stakeholder testing, convergent or discriminant validation, or evidence that ordinal judgments could be averaged as interval measurements. The purported adaptive-capacity inter-rater exercise was a scripted sensitivity scenario, not independent rating. Its ordering changed under alternative author-assigned values.

The numerical scores, radar chart, weighted variants, coordinate-wise dominance claim, and adaptive-capacity ranking are withdrawn. Traceability, answerability, contestability, and enforceability may remain useful questions. They are not yet a measurement scale.

The Canada Council for the Arts Act (Department of Justice Canada 2026a) establishes the Council’s statutory objects, powers, financial provisions, audit, and reporting framework. The Council’s current decision page describes external peer assessment for most applications, ranking against published criteria, staff determination of amounts, delegated approval authority, written results and scores, and a limited review path where evidence suggests a procedural error (Canada Council for the Arts 2026). A more detailed assessment page describes cancellation, reductions, Major Warnings, and Concerned Status, but the Council marks that page under review and out of date. Those details therefore require current-rule confirmation before operative reliance. Even with that caution, the earlier depiction of nearly absent contestability and sanctionability was materially incomplete.

Nothing in this audit establishes that statutory decision authority may be displaced by token voting or smart-contract execution. Depending on the institution and design, publishing applications, deliberations, identities, or immutable decision records could raise questions under several distinct legal and policy regimes. Those regimes cover privacy, access to information, confidential information, copyright, official languages, accessibility, records management, security, procurement, and procedural fairness. Automated decision support may trigger additional federal policy obligations depending on institutional scope and system design. Qualified review must determine what those regimes permit and require of an institution.

The former recommendation for a blockchain grant pilot is withdrawn. The evidence does not include a legal opinion, privacy impact assessment, algorithmic impact assessment, threat model, independent security audit, procurement analysis, administrative cost model, accessibility assessment, or consultation with applicants and affected communities. Less intrusive measures, including versioned open data, clearer decision metadata, procedural logs, reason-giving, and auditable appeal statistics, must be evaluated on their own merits before a distributed ledger is treated as necessary.

A future comparison remains possible, but it requires a different design.

Every empirical result reported here is bounded to the archived packet as it stood at the July 2026 correction. Within that boundary the audit describes published Canada Council award records for fiscal years 2017–18 through 2023–24, an indexer listing and its vote responses for OpenGov referenda created between 1 June 2023 and 1 April 2026, imported dashboard outputs for historical Gitcoin Grants rounds 6–15, and source files at two named commits. The legal and procedural statements rest on a separate layer: current official Canada Council descriptions of decision making retrieved 15 July 2026 and federal legislation current to 26 May 2026. No empirical result depends on observations outside those surfaces.

Each branch carries its own unit, and the unit fixes what a number can mean. A Canada Council row is one published approved award record above the publication threshold, so counts, nominal amounts, and the 0.619204 dispersion figure describe awards rather than applicants, people, or resolved recipient organizations. Polkassembly rows are indexer listing records or indexer-transformed vote rows at one snapshot, which makes the 1,564-record subset a conditional status census rather than the proposal population. Gitcoin rows are imported aggregates whose query and execution metadata are missing. Describing the captured outputs and their provenance gaps is the only use that record supports.

Layer separation bounds every code-based statement. The Polkadot SDK and runtimes clones establish source-level capability at the revisions cited in section 5.4. They do not identify the release or WebAssembly runtime active for every referendum in the archived population, or any executed extrinsic, because the packet lacks an exact block anchor, specVersion, runtime code hash, storage proof, and independent RPC captures.

Outside those bounds lie applicant success, rejection, appeal incidence, recipient-level concentration without documented entity resolution, and causal effects of Canada Council program reform. For Polkadot, a final on-chain delegation share, a stable approval population, and executed-runtime identity lie outside as well. For Gitcoin, so do contribution-level inequality, participant retention, causal effects, and clean independent replay. The withdrawn TI/C/S and adaptive-capacity scores stay outside every branch until the instrument is rebuilt and validated on the terms set out in section 10.1. The withdrawn pilot recommendation stays outside because the record contains no legal opinion, privacy impact assessment, algorithmic impact assessment, threat model, independent security audit, procurement analysis, administrative cost model, accessibility assessment, or consultation with applicants and affected communities.

Legal and temporal statements carry a further limit. The Canada Council for the Arts Act and the Privacy Act are read as current to 26 May 2026 (Department of Justice Canada 2026a, 2026b), and the Council’s decision page as retrieved on 15 July 2026. The Council marks its more detailed assessment page under review and out of date, so the cancellation, reduction, Major Warning, and Concerned Status details require current-rule confirmation before operative reliance. Historical Grants rounds likewise do not stand as a proxy for Gitcoin’s later protocol architecture.

The affirmative claim is methodological: a public data surface becomes evidence only when its semantic units, transformations, population state, and, where code is invoked, release, deployment, and execution identity can be recovered. Five observations would defeat that claim or one of the subsidiary findings the audit retains. Four name the object a reader would have to inspect to settle them, and the first names a replication others would have to run. One can be tested against material the packet already holds. The record does not presently contain what the other four require, and each entry says so.

Four of the five conditions name a retrievable object: rows in an archived endpoint response, chain state at an anchored block, a query with its execution record, or the applicant-level data named in section 10.2. The first requires an independent reconstruction by other researchers rather than an object in this record. None requires a forecast about future practice, which keeps the refutation test separate from the reconstruction programme set out above.

Submission of an empirical comparison or policy recommendation remains blocked until four gates are closed: the Polkadot branch is independently reconstructed on-chain, the Gitcoin branch is replayable from identified raw records, the accountability instrument is validated, and Canadian legal and institutional feasibility is reviewed by qualified specialists with affected-community input.

The earlier manuscript treated public visibility as if it were a common metric, and this audit shows why that move was premature. The Canada Council export is reproducible but normatively incomplete. Polkassembly is unusually inspectable, yet a plausible-looking endpoint statistic failed because its aggregate fields were misunderstood. The Gitcoin outputs are locally present but not reconstructible from their causal and computational source chain. Code repositories describe capabilities but do not identify deployed or executed systems.

The audit ends in a demand for sharper nouns rather than a verdict for bureaucracy or against decentralized governance. An award is not an applicant, an address is not a person, a referendum total is not voter inequality, a source file is not a deployment, and a visible event is not an accountable institution. Publicity lowers the cost of asking questions. Only disciplined reconstruction can tell us which answers it contains.