Prediction ledger

Predictions sealed before the fact: every fingerprint is public and time-stamped on Bitcoin, so none can be quietly dropped or edited. Texts are revealed and scored later.

doc = FileAttachment("data/ledger.json").json()
P = doc.predictions
revealed = P.filter((p) => p.status === "revealed")
resolved = revealed.filter((p) => p.resolved)
score = resolved.length ? d3.sum(resolved, (p) => ({true: 1, partial: 0.5, false: 0})[p.resolved]) / resolved.length : null
fmt = (d) => d ? d3.utcFormat("%-d %b %Y")(new Date(d)) : "–"

html`<div class="stat-strip">
  <div class="stat"><div class="k">Sealed</div><div class="v">${P.length}</div></div>
  <div class="stat"><div class="k">Revealed</div><div class="v">${revealed.length}</div></div>
  <div class="stat"><div class="k">Resolved</div><div class="v">${resolved.length}</div></div>
  <div class="stat"><div class="k">Hit rate</div><div class="v">${score == null ? "–" : (100 * score).toFixed(0) + "%"}</div><div>${score == null ? "nothing resolved yet" : "right = 1, partly = ½, wrong = 0"}</div></div>
</div>`

P.length === 0 ? html`<p><em>Nothing sealed yet.</em> The first predictions will appear here as fingerprints the day they are written; their text follows when they are revealed.</p>` :
html`<div class="table-responsive"><table class="table table-sm">
<thead><tr><th>ID</th><th>Sealed</th><th>SHA-256 fingerprint</th><th>Proof</th><th>Status</th></tr></thead>
<tbody>${P.map((p) => html`<tr>
  <td>${p.id}${p.title ? html`<div class="small text-muted">${p.title}</div>` : ""}</td>
  <td>${fmt(p.date)}${p.resolves_by ? html`<div class="small text-muted">resolves by ${fmt(p.resolves_by)}</div>` : ""}</td>
  <td><code style="word-break: break-all">${p.sha256}</code></td>
  <td><a href="data/ots/${p.id}.ots" download>${p.id}.ots</a>${p.ots_complete ? "" : html`<div class="small text-muted">awaiting Bitcoin</div>`}</td>
  <td>${p.status === "sealed" ? (p.resolves_by && new Date(p.resolves_by) < new Date() ? html`<strong>overdue</strong>` : "sealed") : p.resolved ? html`<strong>${({true: "right", false: "wrong", partial: "partly right"})[p.resolved]}</strong>` : "revealed"}</td>
</tr>`)}</tbody></table></div>`

revealed.length ? html`<h2>Revealed</h2>${revealed.map((p) => html`<section class="mb-4">
  <h3 class="h5">${p.id}${p.title ? ": " + p.title : ""}</h3>
  <p class="small text-muted">Sealed ${fmt(p.date)} · revealed ${fmt(p.revealed_date)} · <a href="data/${p.id}${p.ext}">original file</a></p>
  <blockquote class="blockquote" style="white-space: pre-wrap; font-size: 1rem">${p.text}</blockquote>
  ${p.resolved ? html`<p><strong>Resolved ${fmt(p.resolved_date)}: ${({true: "right", false: "wrong", partial: "partly right"})[p.resolved]}.</strong> ${p.note}</p>` : ""}
</section>`)}` : html``

How to verify a prediction

Anyone can check that a revealed text is exactly what was sealed on the date shown, without trusting this site:

1. Download the revealed file and its .ots proof from the table.
2. Check the fingerprint: shasum -a 256 P-….md must print the SHA-256 in the ledger.
3. Check the date: drop both files on opentimestamps.org, or run

pip install opentimestamps-client
ots verify -f P-2026-10-01-01.md P-2026-10-01-01.ots

ots verify reports the Bitcoin block that commits to the fingerprint, which proves the text existed no later than that block’s time. (Full local verification needs a Bitcoin node; the website verifier doesn’t.)

Why every fingerprint is listed. Publishing only the predictions that came true would make any forecaster look good. Here every sealed fingerprint appears the day it’s made, so a prediction can be revealed late but never hidden. Predictions still unrevealed after their resolution date are flagged as overdue.

The ledger is predictions/ledger.json in the public repository; the scripts are scripts/predict/seal.py and reveal.py.