solhonesty

What each Solana stablecoin product says it pays, next to what it actually paid.

A working prototype of the Solana Honesty Index. Zero runtime dependencies, no API key, no wallet. From a clean clone:

node bin/solhonesty.mjs build

Live output, 2026-08-08, seven products, thirty day window:

OK  kamino-lend-usdc     advertised   3.9611%  realized   4.1065%  gap   -0.1454  issuer_share_price_history
OK  kamino-lend-usdg     advertised   2.4815%  realized   2.9120%  gap   -0.4306  issuer_share_price_history
OK  kamino-lend-pyusd    advertised   4.7135%  realized   4.7082%  gap   +0.0053  issuer_share_price_history
OK  kamino-lend-usdt     advertised   2.6766%  realized   3.6450%  gap   -0.9684  issuer_share_price_history
OK  kamino-lend-usds     advertised   3.9426%  realized   4.3002%  gap   -0.3576  issuer_share_price_history
OK  save-usdc            advertised   2.5300%  realized   2.3619%  gap   +0.1681  thirdparty_rate_series
OK  save-usdt            advertised   1.4200%  realized   1.3901%  gap   +0.0299  thirdparty_rate_series

The finding, which is not the one this project expected

An earlier version of this README led with the widest row of that day: Kamino's PYUSD reserve advertising 6.92 percent against 4.60 percent realized, a 2.32 point gap. Presenting that as a finding about Kamino would have been wrong, and working out why is the most useful thing this project has produced.

Three days later the same code on the same reserve read 3.25 advertised against 4.76 realized. The gap had not narrowed. It had changed sign. On 2026-08-08 a single read caught the USDC reserve advertising 18.19 percent, and a read a few minutes later returned 3.91.

Nothing was broken. A lending reserve's supply APY is an instantaneous rate set by utilisation, and Kamino's rate curve has a steep kink above 95 percent utilisation, where the borrow rate runs from 4.61 to 30.4 percent. A reserve oscillating in the high eighties therefore prints an advertised number whose thirty day range spans a factor of 7 to 18, while the realized figure over the same period moves by hundredths of a point.

So a spot reading minus a thirty day realized figure does not measure the product. It measures when the collector ran. Any advertised-versus-realized board that samples a spot rate once has this problem, including this one, and it is invisible unless you look at the distribution the reading came from.

The board now publishes that distribution:

product	advertised now	min	median	max	range	spot pct'ile	gap vs median
Kamino USDC	3.9611	3.2082	3.8536	24.3491	7.6x	54	-0.2529
Kamino USDG	2.4815	1.8958	2.4853	11.7439	6.2x	48	-0.4267
Kamino PYUSD	4.7135	1.3223	4.0040	24.3431	18.4x	72	-0.7042
Kamino USDT	2.6766	2.2113	3.7693	4.2655	1.9x	7	+0.1243
Kamino USDS	3.9426	3.3428	4.2785	4.7902	1.4x	9	-0.0217
Save USDC	2.5300	not published
Save USDT	1.4200	not published

Measured against the median of its own published history rather than against one screenshot of it, Kamino delivers what it advertises on all five reserves, within 0.71 points, and over-delivers on four of them. That is the honest headline and it is duller than the first draft. Two rows in this snapshot were captured at the 7th and 9th percentile of their own range, so gap_vs_median_pct is the column to quote for those, and the board says so itself rather than leaving the reader to notice.

Save publishes a current rate but no history of it, so those two rows carry the absence rather than a borrowed substitute.

Why this does not already exist

The measurement is easy on Ethereum and hard on Solana, for one reason.

ERC-4626 gives every EVM vault the same convertToAssets call, so you can read a share price from any of them with one line and no protocol specific code. That is why advertised versus realized boards exist for EVM stablecoins.

Solana has no equivalent standard. I checked the obvious escape hatch before building anything: none of USDY, PYUSD, BUIDL, syrupUSDC, sUSDe, USDC or USDG carries the Token-2022 InterestBearingConfig or ScaledUiAmount extension, so there is no protocol agnostic on chain rate to read either. And DeFiLlama, the usual fallback, returns a pricePerShare for exactly one of the top thirty two Solana stablecoin pools by TVL.

So realized yield on Solana has to be assembled per protocol, from raw account data, and that is why the column is empty everywhere.

What is here

src/rpc.mjs                  Solana JSON-RPC, endpoint rotation, no key
src/base58.mjs               base58 both directions, no dependency
src/basis.mjs                the comparability gate
src/advertisedSeries.mjs     how much the ADVERTISED figure itself moves
src/realized.mjs             share price to annualized yield
src/adapters/saveReserve.mjs raw 619 byte reserve decode, share price from chain
src/adapters/kaminoLend.mjs  issuer share price history, advertised supply APY
src/adapters/llamaSeries.mjs third party bootstrap, clearly labelled as weaker
src/engine.mjs               registry in, board out
src/publish.mjs              current.csv, index.json, generated dataset card

52 tests, 47 of which run with no network at all.

npm test                  offline, deterministic
SOLHONESTY_LIVE=1 npm test adds five tests that hit mainnet

The part that makes it checkable

solhonesty verify recomputes a share price from the raw account and holds it against the number the protocol publishes about itself:

$ node bin/solhonesty.mjs verify --key save-usdc
reserve             BgxfHJDzm44T7XG68MYKx7YisTjZu73tVovyZSjJMpmw
slot                437202410
share price (ours)  1.302988992106209
share price (them)  1.3029889921062106
relative difference 1.193e-15
AGREES

That is a 619 byte account decoded by hand, over a public RPC, agreeing with Save's own API to fifteen decimal places. No SDK, no key, no trust.

The comparability gate

This is the part that decides whether a board like this is worth anything.

A protocol quoting a since inception average is not answering the same question as a protocol quoting what a holder earns today. If you rank a board by whichever number is largest on each protocol's own page, you have ranked marketing copy, and every protocol can improve its position by adding a bigger number to its site. I know this failure mode specifically because I shipped it on the EVM version of this board and had to publish the correction.

So: an advertised figure is compared only when its basis is spot-apy or trailing-30d. Everything else is still published, with the reason it was not compared attached. Where a protocol publishes several comparable figures, the most conservative wins. There are unit tests that assert a protocol cannot improve its standing by publishing a larger number, and that a row bundling an external incentive is refused rather than silently compared against a share price that never contained it.

Realized methods, weakest to strongest

Every row states how its realized figure was obtained, because a reader who cannot tell these apart has been misled by the format rather than the numbers.

method	what it means
thirdparty_rate_series	a time weighted mean of somebody else's daily rate observations. Weakest. Used only where no share price series exists yet.
issuer_share_price_history	a share price series the issuer publishes.
onchain_share_price	the share price derived here from raw account data.

The independently collected on chain series starts the day this collector first runs. It does not backfill and it will say insufficient_history rather than annualize a window it does not have.

Prior art, and what I actually checked

Plenty of places publish Solana yields. What I went looking for was something narrower: an issuer's advertised figure, quoted verbatim from the issuer's own surface and dated, sitting next to an independently computed realized figure, under a written rule about when the two may be compared.

What I checked on 2026-08-04, and what I found:

• DeFiLlama publishes rates, not advertised claims, and returns a pricePerShare for one of the top thirty two Solana stablecoin pools by TVL. It cannot answer the question.
• vaults.fyi is paywalled. GET /v2/networks returns 402 Payment required, so whatever it covers is not an open dataset.
• Hugging Face has no dataset matching solana yield. The search returns an empty array.
• Protocol analytics (Kamino's own, Dune boards) report each protocol against itself, which is the one comparison that cannot detect a gap.

I am stating what I searched rather than that nothing exists, because the second is not a claim I can support. If something does this already, the right outcome is that I find it and stop, and I would rather be told than not.

Licence

Code: MIT, in full at LICENSE. Data produced by it: CC BY 4.0, except advertised_verbatim, which is short factual quotation from each issuer's own public surface, attributed and dated. Protocol names are used nominatively to identify the products measured.

Status

Prototype, built 2026-08-04. Seven products. The methodology is a port of a board that has been running on EVM stablecoins since 2026-07-14 and is published as an open dataset at huggingface.co/datasets/kerne-protocol/honesty-index.