honesty-index / README.md
Enerzy17's picture
Refresh: snapshot 2026-08-14T06:37:15.328Z
61b9bcd verified
|
Raw
History Blame Contribute Delete
15.5 kB
metadata
license: cc-by-4.0
language:
  - en
pretty_name: The Kerne Honesty Index
tags:
  - finance
  - stablecoins
  - defi
  - ethereum
  - yield
  - erc-4626
  - onchain
  - transparency
size_categories:
  - 1K<n<10K
source_datasets:
  - original
task_categories:
  - tabular-regression
  - time-series-forecasting
annotations_creators:
  - expert-generated
configs:
  - config_name: current
    default: true
    data_files:
      - split: train
        path: data/current.csv
  - config_name: history
    data_files:
      - split: train
        path: data/history.csv

The Kerne Honesty Index

What each synthetic dollar advertises, next to what it actually paid.

Advertised APY versus realized APY for 21 synthetic dollar vaults, recomputed hourly from ERC-4626 share price growth on chain, and signed.

The realized column is not taken from anybody's dashboard. It is measured directly from the vault contract: convertToAssets(10**decimals) read at two block heights, divided by 10**asset_decimals, annualized over the real elapsed time between those two blocks. Every row is reproducible by anyone with an archive RPC endpoint and the block numbers in this file.

Why this exists

A protocol's advertised rate and the rate its holders actually received are different numbers, and almost nobody publishes both. This dataset publishes both, for a category where the difference is material, on one fixed method applied identically to every row including the publisher's own.

Kerne Protocol builds and operates this index and is also row kerne-skusd in it, measured on exactly the same terms as everyone else. In this snapshot Kerne is the worst row on this board on both axes, delivery ratio and absolute gap. That is not a disclaimer bolted onto the bottom of a marketing page: the flags behind that sentence are computed from the data on every build, so the claim changes by itself when the fact does.

The board in this snapshot

Gap is advertised minus realized, in percentage points, so a positive gap is a shortfall. Delivery is realized as a percentage of advertised, so 100 means a protocol paid exactly what it said.

Protocol Vault Chain Status Comparability Gap (pp) Delivery (%)
Kerne skUSD base live direct 4.98 0.0
Aegis sYUSD ethereum live different-window (not stated) (not stated)
Avant savUSD avalanche live different-window (not stated) (not stated)
Cap stcUSD ethereum live different-window (not stated) (not stated)
Curve scrvUSD ethereum live different-window (not stated) (not stated)
Ethena sUSDe ethereum live direct -0.03 100.7
Falcon Finance sUSDf ethereum live different-window (not stated) (not stated)
Frax sfrxUSD ethereum live direct 0.01 99.7
Maple syrupUSDC ethereum live context-only (not stated) (not stated)
Maple syrupUSDT ethereum live context-only (not stated) (not stated)
Neutrl sNUSD ethereum live direct -0.01 100.2
Noon sUSN ethereum live direct 0.14 98.3
Sky sUSDS ethereum live direct -0.02 100.6
Spark sUSDC ethereum live direct -0.02 100.6
Tori strUSD ethereum live different-window (not stated) (not stated)
Usual sUSD0 ethereum live context-only (not stated) (not stated)
Stables Labs sUSDX ethereum dormant context-only (not stated) (not stated)
Resolv wstUSR ethereum recovery context-only (not stated) (not stated)
YieldFi yUSD ethereum recovery context-only (not stated) (not stated)
Elixir sdeUSD ethereum wind-down context-only (not stated) (not stated)
Level slvlUSD ethereum wind-down context-only (not stated) (not stated)

Of 21 rows, 7 state a gap, 6 publish a figure this dataset declines to compare, and 8 publish no yield figure at all.

The most important rule in this dataset: a blank is not a zero

gap_pp and the delivery ratio are empty, never zero, wherever the method is not entitled to state one. A zero would read as "advertised and realized agree exactly" about a company that simply publishes on a different window, which is the single most damaging thing this file could assert about a named third party.

A gap is stated only when the advertised figure is directly comparable to a 30 day realized window:

advertised_basis comparability Gap stated?
modeled direct yes
trailing-30d direct yes
spot direct yes, a governance set spot rate is stable over the window
trailing-7d different-window no, both figures are shown and no gap is manufactured
inception-to-date context-only no, a since inception average is not what a holder earns today
(nothing published) context-only no

Several protocols publish more than one figure. The picker takes the comparable one, not the largest and not the most flattering. Ranking this board by each protocol's highest advertised number produces a different and wrong ordering, and the comparability column is here so that a consumer cannot reproduce that mistake by accident.

Delivery is gated on the identical rule: where no gap may be stated, no ratio is published either.

If one of these rows is yours

Everything above is free, and stays free whether or not anybody ever buys anything. Nothing on this board is for sale: no row moves for money, no row is added or removed for money, and the method is the same one applied to the publisher's own row. That is the boundary, and this section is the only commercial passage in this file.

There is a longer version of this measurement that a protocol can commission about itself. Every non-Kerne row has a scoped landing page, reached by appending its key from the table above:

https://kerne.fi/disclosure-audit?row=<key>&src=hf

The keys below are picked out of this snapshot, closest to exact delivery first, rather than typed into this file. Today the rows landing nearest their own advertised rate are:

What it is: a Disclosure Integrity Audit, 499 US dollars flat, one report inside 72 hours of scope confirmation, commissioned by the protocol being reviewed and delivered privately to it. It reviews whether your public claims match your chain on three axes: advertised against realized yield (the method in this dataset, run properly over your whole surface rather than one vault), the addresses and figures in your documentation against the live registry, and your oracle and attestation posture. It ends in an EIP-191 signed findings summary you are free to publish or to never mention again.

What it is not: it is disclosure review, not a security audit and not assurance. It certifies nothing about your security, your solvency or your compliance, and it is not a rating. Full scope, terms and the checkout: https://kerne.fi/disclosure-audit

Files

Path Rows What it is
data/current.csv 21 The board at snapshot time, one line per vault, 37 columns.
data/history.csv 5550 The realized series, one line per vault per hourly read, 6 columns.
snapshot/honesty-index-api-response.json - The signed snapshot, so the signature can be checked against these exact bytes.
verify.py - Independent verification: recomputes the attestation hash, recovers the signer, and cross checks the CSV.

data/current.csv columns

key, protocol, symbol, chain, vault_address, status, advertised_pct, advertised_label, advertised_basis, advertised_captured_at, advertised_source_url, advertised_verbatim, comparability, gap_pp, realized_ok, realized_apy_pct, realized_reason, share_price_from, share_price_to, share_price_growth_pct, from_block, to_block, from_ts_utc, to_ts_utc, elapsed_seconds, realized_window_days, window_clamped, total_assets, total_supply, asset_symbol, is_kerne, permalink, badge_svg, snapshot_generated_at, snapshot_window_days, snapshot_signer, snapshot_attestation_hash

Column order is frozen. New columns are appended at the end and existing ones are never moved or renamed, because somebody will index by position no matter how loudly a README says not to.

data/history.csv columns

read_at_utc, read_at_unix, row_key, protocol, symbol, realized_apy_pct

The series in this release runs 2026-08-01T00:37:16.000Z to 2026-08-14T06:37:15.000Z: 5550 entries across 320 hourly reads. One entry is one vault in one read, so entries are not hours and this file is not as deep as its line count suggests. An earlier archive of the same series exists, predates a storage migration, and is not in this release; it will be merged when it becomes readable again. Nothing here should be read as a claim to depth this file does not have.

Provenance of every column

Realized figures (realized_apy_pct, share_price_from, share_price_to, from_block, to_block, total_assets, total_supply): read directly from the vault contract at the two block heights named in the row, over public archive RPC. Reproducible by anyone. This is the column this project exists to produce.

Advertised figures (advertised_pct, advertised_label, advertised_basis, advertised_captured_at, advertised_source_url, advertised_verbatim): read off the public surface named in advertised_source_url on the date in advertised_captured_at. In most rows that surface is the issuer's own site, app or public API; where it is not, it is the public data endpoint that the issuer's own information site is itself built on, and the URL says so plainly rather than being laundered into a claim that the issuer published it directly. advertised_verbatim carries the short factual snippet exactly as that surface rendered it, so a reader can check that the figure was not paraphrased into something the issuer never said. Where an issuer publishes no figure the field is empty and no figure is invented.

Advertised figures are captured periodically, not hourly, so advertised_captured_at will normally be older than snapshot_generated_at. Both dates are in every row, so the lag is visible rather than hidden.

Status (live, recovery, wind-down, dormant): wind-down is used only where the issuer itself announced one. dormant exists precisely so that a vault whose accrual has stopped with nothing announced is not labelled with a decision its issuer never made. Read the delivery ratio beside this column: a protocol in an announced wind down paying nothing is behaving correctly.

Verification

Each snapshot is signed. attestation_hash is "0x" + sha256(utf8(signed_payload_canonical)) and signature is an EIP-191 personal_sign over the raw 32 bytes of that hash.

  • Signing key (public address): 0x84949170e0ad0f9bd8686a4aa4922c10f5fdc4ea
  • This snapshot's attestation hash: 0xb2eeab6d735d4fda41759f9ec9a89fc8f80b4cfc9068140175905748709be187

Hash the published signed_payload_canonical string bytes. Do not re-serialize the JSON: any re-serialization changes key order or spacing and the hash will not match. That is the check working, not a bug. Run verify.py in this repo to check the hash, the recovered signer, and that the CSV describes the same snapshot as the JSON.

The signature covers the snapshot rows. Any delivery and cite blocks served by the live API sit outside the signed payload on purpose, because they are arithmetic on numbers already inside it and must not appear to be attested by the signature.

Update cadence, and which copy wins

The upstream collector runs hourly and writes a fresh signed snapshot. window_days is 30.

This copy is a lagging mirror and is refreshed on a schedule. When it disagrees with https://kerne.fi/api/honesty-index, which is always current, free and needs no key, the live endpoint is correct and this file is stale. Check generated_at in snapshot/honesty-index-api-response.json before using this copy for anything time sensitive, and cite the live endpoint rather than this mirror.

Limits, stated plainly

  • The roster is not the category. It covers vaults that are readable ERC-4626 contracts on chains this collector supports. Protocols that are not ERC-4626, or that live on chains not yet wired up, are absent, and absence is not a judgment. The full coverage limits are published on the rendered board: https://kerne.fi/honesty-index#methodology
  • A 30 day realized window is a choice. Long enough to survive a bad day, short enough to describe the present. It is not the only defensible window, and it is why the trailing-7d publishers get no gap here.
  • This is a measurement, not a rating and not a grade. There are deliberately no letters, no scores and no rankings of companies in this file. A low delivery ratio is not by itself evidence of dishonesty; read it beside status.
  • Not investment advice. Nothing here is a recommendation to buy, hold or redeem anything.

Licence and attribution

The compilation, the schema, and every figure Kerne computed (the realized columns and everything derived from them) are released under Creative Commons Attribution 4.0 International (CC BY 4.0). Attribute to The Kerne Honesty Index, Kerne Protocol, with a link to https://kerne.fi/honesty-index.

Two carve outs, stated so that reuse is safe:

  1. advertised_verbatim is quotation, not Kerne's work. Each value is a short factual snippet reproduced from the third party surface named in advertised_source_url, so that a reader can verify the advertised figure is reported accurately. Kerne asserts no ownership of those snippets and grants no rights in them. They are factual matter, quoted with attribution and a date, and each is a handful of words.
  2. Protocol names, token symbols and marks belong to their owners. They appear here as nominative references identifying the product being measured. Their presence is neither an endorsement by them of this dataset nor a claim by Kerne of any affiliation.

If you are one of the issuers listed and something in your row is wrong, the correction path is public and the row will be fixed: https://kerne.fi/honesty-index

Citation

@misc{kerne_honesty_index,
  title  = {The Kerne Honesty Index: advertised versus realized APY for synthetic dollars},
  author = {{Kerne Protocol}},
  year   = {2026},
  howpublished = {\url{https://huggingface.co/datasets/kerne-protocol/honesty-index}},
  note   = {Snapshot 2026-08-14T06:37:15.328Z, attestation 0xb2eeab6d735d4fda41759f9ec9a89fc8f80b4cfc9068140175905748709be187}
}