Follow-up audit: Read the newer claim, number and history review. It supersedes the cumulative-number and coverage assumptions in this earlier pass.

Accrue data integrity and coverage audit

Review date: 10 September 2026. Scope: token buybacks and burns funded by protocol activity.

Findings and confidence

The audit found real omissions, conflicting product mappings, unsupported allocation estimates, and inactive programs included in the public universe. The corrected snapshot contains 159 priced eligible projects, compared with 148 before the audit. Fourteen projects were added or restored; three were excluded. The expanded view contains 181 eligible rows, including unpriced coverage. These counts describe this dataset, not the entire crypto market.

Every row in the starting snapshot was checked structurally and its derived metrics recomputed. The review also examined duplicate token mappings across the complete registry, screened excluded priced projects, queried source methodology endpoints, and checked primary documentation for the material changes described below. Arithmetic consistency is a narrower finding than input validity: a perfectly calculated ratio can still use the wrong token, denominator, program status, or reporting period.

This is a source and implementation audit, not an independent financial statement audit or a transaction-by-transaction reconciliation. No live burn-balance reading was obtained in the local verification run. The environment lacked the explorer key needed for its EVM enrichment. Source citations, curated figures, and source-reported execution updates must therefore retain their provenance. A recently fetched feed does not make an old policy current.

Coverage and method

The starting snapshot contained 2,301 rows, of which 702 had a positive token price and market cap. The registry contained 546 entries and 198 positive buyback/burn claims. An automated scan found 41 CoinGecko identifiers shared by multiple registry entries, 19 positive registry claims without a priced eligible counterpart, 115 percentage claims requiring denominator review, and 44 descriptions containing possible status concerns. These are screening counts; neither duplicates nor keyword matches automatically prove an error.

The source sweep queried 757 candidates: every priced row plus eligible unpriced rows. It obtained 461 successful responses, including 288 with nonempty methodology. The remaining 296 returned HTTP 400 at the queried holders-revenue summary route. Some identifiers do not support that route, represent curated businesses, or differ from summary API slugs. A missing response does not prove that a program does not exist. Parent summaries frequently provide no methodology even when child products provide it.

The downloadable inventory records the starting universe alongside corrected visibility, queried URLs, source outcomes, review flags, and mechanism evidence. The separate registry inventory preserves the original percentage claims and the denominator-review status. Those claims remain research inputs; they are not automatically accepted as valuation inputs.

The final local snapshot contains 2,252 rows. Removing same-token synthetic duplicates reduces the raw row count without deleting the underlying curated research records. Internal feed coverage remains broader than the public buyback/burn universe.

Confirmed additions and restored mappings

Bluefin / BLUE. A parent protocol had inherited the classification of its Spot child while the positive BLUE record became an unpriced duplicate. The canonical record now joins the Bluefin parent. Bluefin's official announcement reports completed Q1 2026 purchases, establishing that this is more than an announced intention. The review does not assume that the earlier “up to 25%” claim is a fixed share of all consolidated revenue. The annual allocation estimate is withheld.[1]

Optimism / OP. The parent previously inherited a Superchain revenue-share record without buybacks. It now uses the reviewed Optimism policy. The official communication thread records completed purchases funded by January, February, and March 2026 revenue, with an August update reporting a cumulative 9,451,924 OP. Later execution is not confirmed by that thread. Purchases go to treasury; they are not tagged as burns or direct holder distributions. OP Mainnet revenue alone is not an adequate denominator for a Superchain-wide policy.[2]

DoubleZero / 2Z. The parent inherited a staked-SOL product's classification, obscuring Network and Edge security burns. The corrected record includes the documented 2Z mechanism. DefiLlama distinguishes the Network and Edge burn streams from staked-SOL revenue without 2Z sharing. Applying a child product's 10% fee rule to aggregate revenue would be misleading, so no aggregate percentage is modeled. This finding relies on source methodology rather than an independently reconciled execution ledger.[3]

Gearbox / GEAR. Conflicting registry entries separated the priced protocol from its buyback description. The canonical Gearbox record now carries the policy described in the protocol's January 2026 update. GIP-219 directs 25% of realized monthly revenue toward GEAR-WETH liquidity. That LP budget is not a measured amount of GEAR-only purchases, a holder distribution, or a burn. The classification is treasury/owned-liquidity accumulation and the annual purchase estimate is withheld.[4]

THORSwap / THOR. Official documentation describes a 20% revenue buyback-and-burn allocation, alongside a separate 55% holder distribution and 25% treasury share. The source revenue definition matches protocol swap revenue. The dashboard models the 20% allocation rather than treating all holder revenue as burned.[5]

Streamflow / STREAM. The official application describes protocol-revenue-backed STREAM rewards and continuous buybacks; the feed methodology identifies purchases for stakers. The project is included as buyback-and-distribute. The audit does not infer a fixed percentage of total revenue or a burn.[6]

Reserve / RSR. Official documentation describes Index DTF platform fees purchasing and burning RSR. This differs from Yield DTF distributions also included in aggregate holder revenue. The project is included, but no assumption is made that every dollar of consolidated holder revenue buys or burns RSR.[7]

xExchange / MEX. Official documentation allocates 0.05% of swap volume to MEX buyback-and-burn and another 0.05% to Energy holders. The reviewed swap revenue adapter includes both components. The modeled burn allocation is therefore 50% of that measured revenue, rather than 0.05% of revenue or all holder distributions.[8]

Somnia / SOMI, Kaia / KAIA, and Astar / ASTR. Their official documentation confirms native transaction-fee burns. The reviewed feed revenue lines already represent burned fees. The modeled allocation is 100% of those revenue lines; applying the fee percentage again would undercount. Fee burning does not establish net deflation because token issuance remains a separate quantity.[9][10][11]

DeepBook / DEEP. The developer repository describes periodic burning of surplus DEEP trading fees after rebates. The project was missing despite this mechanism. The dashboard does not equate collected revenue with same-day burns: settlement occurs in batches and the parent can contain additional products.[12]

Alkimi / ALKIMI. Official documentation describes advertising protocol fees funding market purchases distributed as staking rewards. The project is included with the current token identifier from the market feed. No burn or fixed percentage of the broader advertising business is assumed.[13]

Bancor / BNT. Later governance documentation specifies buying and burning BNT through the Carbon Vortex connection, whereas older technical documentation describes vBNT. The reviewed record makes that version distinction explicit. A single percentage is not applied across the parent's different fee mechanisms.[14]

Exclusions and tagging corrections

Balancer / BAL is excluded from the active public universe. BIP-919 describes an exercise window twelve months after the snapshot, followed by a twelve-week claim period. An earmarked treasury budget is not evidence that purchases are currently executing. The proposal and historical research remain in the registry with an explicit proposed status.[15]

Gains Network / GNS is excluded from the ongoing revenue-funded program view. The June restructuring redirects the buyback/burn budget toward growth, and the reviewed governance discussion reports passage. Separate collateral-balancing burns should not be confused with continuation of the fee-funded buyback program.[16]

Kaito / KAITO is withheld pending verification. Conflicting records mixed launchpad distributions of other projects' tokens with a low-confidence buyback claim. The audit did not establish current execution of a KAITO repurchase program from sufficient primary evidence. This is an unresolved inclusion decision, not a definitive claim that KAITO has never been bought back.

Flaunch / FLAY remains included. Its current documentation explicitly describes discretionary purchases funded by selected liquidity-position revenue. The disabled governance fee switch is a separate mechanism and does not establish a 10% buyback budget. Purchases of tokens launched through Flaunch must not be attributed to FLAY.[17]

Lido / LDO is classified as treasury accumulation rather than holder distribution, consistent with the stored accumulation-program description. Existing ether.fi treatment is preserved: purchases for redistribution are not burns, and the withdrawal-fee allocation is not generalized to every revenue stream.[18]

Parent identity corrections also prevent Hyperliquid, Chainlink, Lista, Sky, BONK.fun, and Inverse Finance from borrowing the policy of an arbitrary representative child. These are mapping corrections; they do not independently certify every historical execution or percentage in those protocols' research notes.

Financial calculation safeguards

Denominator matching is now explicit. A curated percentage can enter the public revenue allocation calculation only after its basis is marked as the protocol revenue measured by that row. Otherwise the public percentage and derived annual allocation are unavailable. The original claim stays in the registry inventory for review. This deliberately reduces the number of populated estimates: only six priced rows in the corrected snapshot currently pass that gate. An unavailable estimate is not a finding of zero spending.

The retained estimates are Ethereum, THORSwap, xExchange, Somnia, Kaia, and Astar. Even these are annualized allocations or burn values under the stated source definitions, not forecasts of future purchases. The change prevents fees, profits, selected product income, treasury budgets, and liquidity budgets from silently sharing one denominator.

Measured USD needs a reporting window. A live dollar reading with no period is no longer accepted as annual spending. For a valid measured window the formula is amount × 365 / periodDays. An all-time total cannot become an annual rate merely because it was fetched today. Existing cumulative fields remain cumulative.

Explicit burn tags take precedence. A record with hasBurn: false no longer generates a burn-linked valuation estimate because an older mechanism label happens to contain “burn.” Proposed, paused, historical, or unverified programs cannot pass the active mechanism gate through their registry booleans.

Ambiguous token fallback is rejected. Shared CoinGecko IDs are removed from automatic token-only registry matching. Parent policies require an appropriate parent or identity match; a representative child's slug cannot silently determine the entire parent's policy. Same-token unused registry aliases no longer create additional synthetic public projects.

The starting snapshot had no discrepancies when stored derived metrics were recomputed with the then-current formulas. That result did not validate the old assumptions. Regression tests now cover the corrected assumptions, including real production registry mappings, status exclusion, denominator withholding, period annualization, duplicate prevention, and conflicting burn tags.

Source conflicts and remaining uncertainty

Algorand illustrates why an upstream “burn” label is insufficient. The queried adapter describes transaction fees as burned; the current Algorand specification and staking FAQ describe fee-sink-funded proposer rewards. ALGO remains excluded. Similarly, a generic chain fee-burn description does not establish that the burned gas asset is the governance token shown in a market row. Asset identity must be checked before inclusion.[19]

The broader sweep surfaced additional review candidates such as Tree News, Zinc, Neverland, Peptides, Kintara, Swop, Haystack, Polycule, Jade, Surf Liquid, and several chains. Keyword screening also matches negative statements such as “no buyback,” redemption operations, and burns of a different token. These candidates are recorded for review rather than automatically promoted to confirmed coverage.

The registry's remaining program descriptions and cumulative figures are sourced claims with varying confidence and dates. This audit does not establish that every legacy claim remains active, every supplied address is exhaustive, every buyback is revenue-funded, or every external supply figure is correct. Provider accounting changes, contract upgrades, migrations, OTC purchases, and irregular execution can all affect interpretation.

Market capitalization, supply, fee revenue, and burns can have different timestamps and scopes. Revenue multiples are not profit multiples; token burns are not necessarily net supply reduction; holder revenue can include distributions without repurchases. Users should compare similar mechanisms and inspect the linked methodology and verification dates.

Validation and maintenance

The verification suite passed 1,058 tests after the integrity changes, with 24 existing tests skipped. TypeScript checks passed. The final local snapshot was built from live revenue and market feeds. All 181 eligible rows received cached, validated image assets. The deployed browser check confirmed all 181 logos decode, project popups open, and comparisons and the restricted project picker work. Three additional history aliases map Sky, BONK.fun, and Optimism to live source routes whose 30-day revenue totals match the snapshot. The Optimism source does not provide a historical holders-revenue series; that gap stays explicit.

Future additions should identify the exact token, distinguish documented operation from proposals, record a primary source and verification date, establish the funding denominator before exposing a percentage, and specify the reporting period for measured spending. Periodic review should revisit paused and conditional programs and investigate new source candidates. The machine-readable inventories make those outstanding checks visible rather than hiding them behind a completeness claim.