ScoreTrace Ledger: Tamper-Evident Track Records
ScoreTrace Ledger: Tamper-Evident Forecaster Track Records
Know who has actually earned your trust, not who sounds the most confident.
ScoreTrace Ledger keeps a permanent, tamper-evident record of public forecasts and tells you who has earned weight. Add calls before they resolve, add outcomes when they come in, and it reports each forecaster's hit rate, confidence interval and calibration, plus the consensus on every open question: naive next to track-record-weighted.
What you get
- An append-only ledger file you keep. Every entry is timestamped and chained by SHA-256 hash, so an edited, removed or reordered entry is detected
- Hit rate with a 95% interval and a plain verdict per forecaster, plus Brier score for stated probabilities
- A weight per forecaster, built only from calls recorded before they resolved. Short records earn little
- Naive vs weighted consensus on open questions, with the number of forecasters behind each
- Flags: backfilled calls, small samples, overdue calls, corrected outcomes, weight concentration, missing sources
- JSON, CSV and Markdown reports
Example (sample data)
| Forecaster | Record | Weight |
|---|---|---|
| Alice | 30 of 40 hits (75%) | 0.20 |
| Cy | 5 of 5 (short record) | 0.17 |
| Bo | 20 of 40 (50%) | 0 |
| Dan | backfilled calls only | none |
On open question Q101 the naive consensus is 60.0 and the track-record-weighted consensus is 43.6, because the forecasters who have earned weight lean the other way. All names and numbers above are fictional sample data.
Inputs
- Public calls: forecaster, question, dates, claim, source, and a probability or yes/no stance
- Outcomes with their sources
- Your existing ledger file, if you have one
How it stays accurate
A deterministic offline script does every calculation and write, so the same ledger always gives the same report. The agent never invents a call, date, probability or outcome, asks you when a post may not be a real call, and states every exclusion. Calls recorded after their horizon or after the answer was known earn no weight.
Limits
- The ledger proves when an entry was recorded and that it was not changed. It does not prove who made a call or when it was first made
- Save the head hash it prints, to detect entries removed from the end
- Weights describe the recorded sample, not future accuracy
- It does not predict outcomes, judge honesty, or give financial, investment or betting advice
Data sharing
Only the material you supply and public pages you link are used. No passwords, keys or accounts are needed. The ledger script runs locally with no network calls. The text you give the agent is processed by the AI model provider running it: the model you choose in Download mode, or the model configured on this listing when run on Capafy. Keep the ledger file on your own machine, and don't share anything you wouldn't give that provider.
Try it
"Add these three public forecasts to my ledger, then verify the chain and show me the weighted consensus."
