Condense receipt / Solana mainnet

Receipt

Qwen2.5-1.5B-Instruct answered a prompt and the hashes of both were committed on Solana mainnet at 2026-10-07 21:02:31 UTC. Re-run it below and compare the hash yourself.

Answer

Prompt

KORTX emitter prompt v1 (emitter-prompt-v1). Emitter: Vapor Drift Disposition: bias none; horizon 1 hour; conviction threshold 50 of 100. Disposition hash: 7c7ce4ab5828b263a635b94c406ac7433183da05b6c504f49333d41ec68613e7 Snapshot time: 2026-10-07T21:02:16Z Source: Coinbase Exchange public 1-minute candles. ETH (ETH-USD), last 8 one-minute candles, oldest first (UTC time, open, high, low, close): 20:54, 2572.22, 2572.45, 2572.22, 2572.34 20:55, 2572.34, 2572.8, 2572.31, 2572.38 20:56, 2572.33, 2573.04, 2572.11, 2572.93 20:57, 2573.04, 2573.04, 2572.22, 2572.66 20:58, 2572.7, 2573.2, 2572.18, 2572.36 20:59, 2572.49, 2572.84, 2572.46, 2572.76 21:00, 2572.74, 2573.91, 2572.74, 2573.32 21:01, 2573.35, 2573.56, 2573.15, 2573.19 Task: say whether one of these assets will be higher (UP) or lower (DOWN) 1 hour after the snapshot time than its last close above, or say NONE. Reply with exactly two lines and nothing else: CALL: <ASSET> UP or CALL: <ASSET> DOWN or CALL: NONE CONVICTION: <whole number from 0 to 100>

CALL: ETH UP CONVICTION: 85

Recomputed in this browser

The text on this page, hashed here with KCF-1 strict and compared with the committed hashes in the record. The input hash covers more than the prompt: the model, the runtime, the system prompt and the decoding settings are all inside it.

Answer against the output hash

Hashing

Prompt against the input hash

Hashing

Rebuilt from the model declaration stored when this trace was committed.

Record

Trace: Account6NdhWfTK66TXQs6kvSogwZb2Eifac29srbsGwFpbn2A
Open the trace record
On-chain statusread from chain by the service: committedStored answer matches the chain. Stored prompt matches the chain.
Accountthe trace account on chain: Account open -- rent returns to the provider after the challenge window closes (2026-10-08 21:02:31 UTC).
Input hashKCF-1 over model, prompt, decoding
Output hashKCF-1 over the answer
Model: Qwen2.5-1.5B-Instruct
Weights fingerprint
Plate
Transaction: Solana mainnet4xFE2kK3jAapG4EdKPv2nr54BZnXHZBLqM5fRKHcgp6hWWQumsKzwhEUBW7y2y3wmdRpdpvxsBqK5EGY6yAHYwRE
Committed: 2026-10-07 21:02:31 UTC
Server time: 14,607 ms, run to confirmed commit

Re-run

Send the committed trace back through the same weights with the same decoding settings. If the model is deterministic, the new answer hashes to exactly the committed output hash.

Earlier re-runs

Kept by the Condense server. Same-server re-runs are not written to chain.

Nobody has re-run this trace yet.

If this record does not hold up, a bonded challenge puts the trace in front of independent verifiers. It is an on-chain transaction signed by your wallet.

Open for 22 more hours, until 2026-10-08 21:02 UTC.

Challenge this record

Tier: Sampled

Traces are committed under the sampled tier: eligible for re-execution by independent bonded verifiers under a pinned environment. Until an independent node reproduces a trace, its on-chain status stays committed, not verified. Traces that are never drawn are not covered.

How the tiers differ

What this receipt does not prove

One server ran the model and committed the hashes. A rerun on that same server checks that the model is deterministic; it does not check that the server is honest.
The prompt and the answer are stored by the Condense server, not on chain. The chain holds their hashes, so the text shown here can be checked against the chain but not recovered from it.

The link carries the answer and the hashes. Whoever opens it can re-run the trace themselves.

Post on X