Reproduce a signal from scratch

This is the guide the whole system exists to make possible. Every step is something you can do independently.

1. Fetch the signal and its bundle

curl -s http://localhost:3101/api/signals/SIGNAL_ID | jq

The response contains the signal, the complete evidence bundle, and evidence_summary.gaps — the sources that produced nothing, each with its reason.

2. Read the gaps before the reasoning

curl -s http://localhost:3101/api/signals/SIGNAL_ID | jq '.evidence_summary'

A verdict resting on two present sources and four gaps is a different claim from the same verdict resting on six. Read the denominator first.

3. Check the citations yourself

Every key in evidence_used must exist in the bundle:

curl -s http://localhost:3101/api/signals/SIGNAL_ID > s.json
jq -r '.signal.evidence_used[]' s.json | while read -r key; do
  jq -e --arg k "$key" '[.. | objects | select(has($k))] | length > 0' s.json >/dev/null \
    && echo "OK      $key" || echo "MISSING $key"
done

Any MISSING line would mean the model cited something it was not shown. The publication pipeline discards such signals before they reach the feed, so this should always pass — the point is that you can check rather than being told.

4. Recompute the evidence hash

curl -s http://localhost:3101/api/signals/verify/SIGNAL_ID | jq

To do it genuinely independently, recompute from the bundle yourself:

import json, hashlib
# NOTE: the canonical format uses keccak256, which the standard library
# does NOT provide. SHA-3 is a different function and will not match.
from Crypto.Hash import keccak

bundle = json.load(open("bundle.json"))
canonical = json.dumps(
    bundle,
    sort_keys=True,
    separators=(",", ":"),
    ensure_ascii=False,     # ensure_ascii=True silently breaks the hash
).encode("utf-8")

h = keccak.new(digest_bits=256)
h.update(canonical)
print("0x" + h.hexdigest())

import { keccak256, toUtf8Bytes } from 'ethers'

const bundle = JSON.parse(await readFile('bundle.json', 'utf8'))
const canonical = JSON.stringify(sortKeysDeep(bundle))
console.log(keccak256(toUtf8Bytes(canonical)))

Two traps that produce plausible-looking wrong hashes:

ensure_ascii=True (Python’s default) escapes non-ASCII characters and silently changes the bytes. The canonical form requires UTF-8 passthrough.
Python’s standard library has no keccak256. hashlib.sha3_256 is a different function — it produces a valid-looking hash that never matches.

5. Check the anchor honestly

curl -s http://localhost:3101/api/signals/SIGNAL_ID | jq '.signal | {anchor_status, anchor_tx}'

{ "anchor_status": "queued", "anchor_tx": null }

queued is not anchored. The Alpha Ledger contract is not deployed, so this is the expected output for every signal today.A signal is anchored when anchor_tx holds a transaction hash you can look up on a block explorer. The status field is written by the process that queued it; the tx hash by the process that got a receipt. Trust the second.

6. Know where verification stops

Proven

The bundle was not edited after hashing. The reasoning cites only what it was shown. Whether an anchor transaction exists.

Not proven

That the evidence was true when collected. Nothing here can establish that a position reading was correct at the moment it was taken.

Verification establishes tamper-evidence and honest citation, not correctness. A badge implying otherwise would be worse than no badge.

What a full reproduction looks like

Independent hash match

Your recomputation equals the published evidence_hash.

Citations resolve

Every evidence_used key exists in the bundle.

Gaps accounted for

Every non-present source has a stated reason, and the verdict is consistent with what remained.

Anchor status matches reality

anchor_tx is null and anchor_status is queued — consistent with an undeployed contract.

If all four hold, you have verified the signal without trusting anything the vendor said about it.

Scienter publishes impersonal market signals of general and regular circulation. Nothing here is investment, legal, or tax advice, or a recommendation to buy, sell, or hold any asset. Scienter is not a registered investment adviser or broker-dealer. Trading digital assets can result in the total loss of your funds — see Disclaimers for the full text.

Get Started

Concepts

Guides

Reproduce a signal from scratch

1. Fetch the signal and its bundle

2. Read the gaps before the reasoning

3. Check the citations yourself

4. Recompute the evidence hash

5. Check the anchor honestly

6. Know where verification stops

Proven

Not proven

What a full reproduction looks like

​1. Fetch the signal and its bundle

​2. Read the gaps before the reasoning

​3. Check the citations yourself

​4. Recompute the evidence hash

​5. Check the anchor honestly

​6. Know where verification stops

Proven

Not proven

​What a full reproduction looks like

1. Fetch the signal and its bundle

2. Read the gaps before the reasoning

3. Check the citations yourself

4. Recompute the evidence hash

5. Check the anchor honestly

6. Know where verification stops

What a full reproduction looks like