An RPL-1 program: stateless, run with rand call. The caller proves that a number they hold
privately is at least a threshold fixed when the program was deployed. The chain learns that the
statement is true, and nothing about the number.
public input (deploy) threshold: two words, a u64 little-endian threshold.txt = "1000 0"
private input (call) amount: two words, a u64 little-endian never leaves your machine
outputs (receipt) [1, threshold_lo, threshold_hi, 0, 0, 0, 0, 0]
Below the threshold there is no proof at all: the program refuses by reading a private input index no caller can have committed, so the run has no trace. A receipt for this program can only ever say "yes".
What it proves is knowledge of a number, not ownership of funds: it is the smallest example of a private input and a public one meeting in a proof. A real use would bind the amount to something the chain checks (an RPL-2 cell, a note commitment).
| file | |
|---|---|
src/main.rs |
the guest |
Cargo.toml |
a no_std, no_main crate on guest-sdk, panic = "abort" |
threshold.txt |
the deploy-time public input: lo hi |
build.sh |
build image.bin inside your circuits checkout and reject any panicking path |
run.sh |
run it on the zkVM emulator, off chain, above and below the threshold |
deploy.sh |
deploy with threshold.txt as the public input; saves the id to program.id |
call.sh <amount> |
prove the statement on chain and print the receipt |
export CIRCUITS=~/src/circuits # see the top-level README for the rest
./build.sh
./run.sh 2500
./deploy.sh
./call.sh 2500run.sh, off chain:
threshold 1000 (public), amount 2500 (private):
out[0] = 1
out[1] = 1000
out[2] = 0
…
tier 10: fits (cycles 56 of 1023, Poseidon2 permutations 16 of 128)
amount 1 (below the threshold):
trap: InputIndex(4294967295)
The image is 51 words and every call proves at tier 10, the smallest.
call.sh passes --expect-public threshold.txt, so the wallet refuses before proving if the
program on chain was deployed with a different threshold. The threshold is part of the program id
(rand-program-2 hashes the code and the public input), so changing threshold.txt and
redeploying gives a new program.
Run 2026-10-01 through https://rpc.randprotocol.org with a v0.6.8 (main) rand and these
scripts, unmodified:
program id (deployed with 1000 0) |
71f6c2e0d1772cdcfeba767ceca2939761d7feda0b7bed96dc366096a5aaa541 |
| deploy | fee259de43b482b1396eb8cf8791adef7f10d3fc5dd306ab3e9cabf59e9245b1, fee 0.0063 RAND |
./call.sh 2500 |
1cd6237db84a688e1f1213567b299b58c1e607362eea9a714ef2f2fb8102bcac, outputs [1, 1000, 0, 0, 0, 0, 0, 0], tier 10 |
The program id is the same on every chain: it is a hash of the code and the threshold.
The image build.sh produces (built images are not committed; build.sh reproduces them byte
for byte with the pinned toolchain):
image.bin sha256 |
f07e53a9878da70e1c1e4e46e57b1f48f336993fecd75765814b970cb229abaf |
hc (rand-guest) |
5f74e137f4cb75e8723853f50060c225d2854bb1f6ba8e8e3f6b1902285ffd7e |
| program id, code only | 6b5ccea3633cd11f71a8f59b6a517311861b0dc3f02a6d8c3db22cd8616c0dfd |
program id deployed with threshold.txt = 1000 0 |
71f6c2e0d1772cdcfeba767ceca2939761d7feda0b7bed96dc366096a5aaa541 |
From a fresh faucet wallet, with the scripts above (the pre-rebase feat/rpl2 build):
$ ./deploy.sh
program id: 71f6c2e0d1772cdcfeba767ceca2939761d7feda0b7bed96dc366096a5aaa541 (51 words, …, public input 2 words, …)
authorisation proved in 22.7s: tier 10, 1375787 bytes
proved in 383.4s: tier 14, 1494600 bytes
submitted deploy 7cbb286e16fc8fccec7043555dccfed44d7d93c29bb61465f8a505ad7672c7fe
0 RAND out, 99.9937 RAND change, fee 0.0063 RAND, …
$ ./call.sh 2500
fee 0.0043032 RAND
proved in 26.0s: tier 10, 1372333 bytes, outputs [1, 1000, 0, 0, 0, 0, 0, 0]
authorisation proved in 25.0s: tier 10, 1367819 bytes
proved in 378.4s: tier 14, 1495688 bytes
submitted call 3ef1d975b4cb87967892c18768983bad9e5f7c78bddd1e38bd9e71c83f7160e4
{ "height": 2759, "outputs": [1, 1000, 0, 0, 0, 0, 0, 0], "tier": 10, … }
The call proof (tier 10) took 26 s; most of the wall time is the bundle that pays the fee (tier 14: ~6 minutes on that 4-vCPU cloud machine with the pre-rebase build; the v0.6.8 build proves it in about 20 seconds on an M4 laptop).