Immutable Record
SHA-256 Timestamped
Independently Audited Q4 2025
Updated 2026-05-21 14:30 UTC
Public Audit Log
Every prediction the Pulse Engine has made since March 2023, with timestamps committed before kickoff, results auto-imported from official feeds, and full profit/loss attribution at flat 1-unit stakes. Including all losses. No cherry-picking.
Total Predictions
12,431
since 2023-03-15
Strike Rate
64.2%
7,981W / 4,253L / 197V
Cumulative P&L
+596.4u
flat 1u stake · +4.80% ROI
Best Month
+38.2u
2024-06 · Euro group stage
Worst Month
−18.4u
2024-11 · NCAA upsets
Closing-Line Beat
+1.4%
avg edge vs market close
📈
Cumulative ROI · 38 months
From March 2023 to May 2026 · Visible drawdowns and recovery periods · Audit period start marked
Cumulative P&L (units)
Rolling 30-day win rate
Monthly Performance · Last 6 Months
| Period | Picks | Win Rate | ROI | P&L |
|---|---|---|---|---|
| 2026-05 (MTD) | 42 | 66.7% | +6.2% | +2.60u |
| 2026-04 | 68 | 61.8% | +3.4% | +2.31u |
| 2026-03 | 71 | 67.6% | +9.1% | +6.45u |
| 2026-02 | 59 | 59.3% | −2.8% | −1.65u |
| 2026-01 | 64 | 65.6% | +5.7% | +3.65u |
| 2025-12 | 58 | 62.1% | +4.2% | +2.44u |
Performance by Market · All-Time
| Market | Picks | Hit Rate | ROI |
|---|---|---|---|
| 1X2 (Match Result) | 5,142 | 64.2% | +5.1% |
| Over/Under Goals | 3,840 | 71.4% | +6.8% |
| BTTS | 1,920 | 64.2% | +4.2% |
| Anytime Goalscorer | 980 | 62.8% | +5.7% |
| Asian Handicap | 410 | 56.7% | +2.1% |
| Outright (Long-term) | 139 | 41.2% | −1.4% |
📊
CLV Breakdown by Grade
Closing-Line Value distribution across all settled predictions · Updated via
clv-tracker.py every 15 min ·
How CLV is calculated →
Excellent
1,847
14.9% of picks
CLV > +5%
Good
3,521
28.3% of picks
CLV +2 to +5%
Neutral
4,891
39.3% of picks
CLV -2 to +2%
Poor
1,628
13.1% of picks
CLV -5 to -2%
Avoid
544
4.4% of picks
CLV < -5%
What this means: CLV measures whether market odds moved toward our pick after we published. Positive CLV indicates we had information advantage before sharp money arrived. 43.2% of our picks earn "Good" or "Excellent" CLV grades — the gold-standard signal for sharp betting models. Aggregated from clv-history.json via
audit-clv-integration.js; full math in methodology.
📈
CLV Trend · Cumulative Over Time
Lifetime CLV plotted month-by-month. Rising line = model improving over time. Annotations mark Pulse Engine major releases. Data: monthly aggregates of all settled picks since 2023-03-15 · loading…
Cumulative CLV %
Monthly CLV %
Model release
Period start
2023-03
CLV: 0.0%
Best month
+4.8%
2024-06 · 412 picks
Worst month
-1.2%
2024-08 · 286 picks
Current
+1.42%
Lifetime cumulative
Trajectory
↑ +0.04%/mo
linear regression slope
Why this matters: A single CLV number can be luck. The trend matters more. Positive slope over 38 months means our model has been steadily improving — not just lucky once. Each Pulse Engine major release (v3.0 → v3.2 → v3.4, marked in pink) shows visible inflection. Static fallback shown if live data unavailable. Raw monthly data: .
🔒
How This Log Stays Honest
1. Pre-Kickoff Commit
Every prediction is generated, hashed with SHA-256, and written to the immutable log before kickoff. The 8-char hex prefix shown publicly is verifiable against the full hash stored in our database.
2. Third-Party Timestamp
Snapshots published to archive.org's Wayback Machine within 30 min of publication. Independent anchor proves no after-the-fact modification of timestamp or content.
3. Auto-Imported Results
Match results auto-pulled from API-Football, cross-referenced with ESPN and the competition data feed. Goalscorer markets dual-sourced for accuracy on disputed goals.
4. Append-Only Amendments
If a result is overturned weeks later (disciplinary panel, awarded match), the original entry remains intact. An AMENDED entry is appended. 3 amendments exist in 38 months of history.
⬇
Full Dataset · Free to Audit
Complete 12,431-entry historical log. Row-level
SHA-256 hashes included for independent verification. Licensed CC BY-NC 4.0 — free for academic and journalistic use with attribution.
Audit Log Entries
50 most recent
Showing last 50 settled predictions · Filter, sort, search · Download full log for all 12,431 entries
Filter:
50 of 50 entries · +6.85u
| ID / Hash | Timestamp (UTC) | Match | Pick | Odds | Model % | Edge | Conf | Status | Score | P&L | Ver |
|---|
Showing 1–20 of 50
1x
365
P
Auditing our process, considering a wager? Each row's Odds shows the partner book offering the best price at publication time. You can verify those prices independently — they're a snapshot, not a recommendation.
COMPARE BOOKS →
✓
Verify the Log Yourself
1
Pick a Date
Choose any past prediction date (e.g., 2026-04-15). Search the table or grep the CSV for that date.
2
Check Wayback
Visit web.archive.org/web/2026*/pulsebetty.com/ai-predictions/[slug]/ and confirm the prediction page was archived before kickoff.
3
Verify Result
Cross-check the recorded result against the official competition feed or ESPN. Any discrepancy? Report it — we publish all audit responses.
Audit-Specific FAQ
How are predictions timestamped before kickoff?
Every prediction is generated by Pulse Engine, hashed with SHA-256 (first 8 hex characters shown publicly), and committed to our immutable log within 60 seconds of generation. The full hash includes the prediction payload, model version, generation timestamp in UTC, and a server-side nonce. We publish snapshots to archive.org/wayback-machine within 30 minutes of publication, giving an independent third-party timestamp anchor.
Where do match results come from?
Match results are auto-imported from API-Football and cross-referenced against ESPN and the official competition data feed within 30 minutes of final whistle. Goalscorer markets are verified against two independent sources to handle disputed/overturned goals. Any discrepancies are flagged for manual review by our editorial team.
Can entries ever be modified or deleted?
No. Once an entry is committed, it is immutable. If a result is later overturned (e.g., a match awarded retrospectively, a goal disallowed weeks later by a disciplinary panel), the original entry remains with its original result and a separate amendment entry is appended. We have appended 3 such amendments in the history of the log; all are searchable in the table above by status AMENDED.
How is profit/loss calculated?
Assuming a flat 1-unit stake per pick: a winning bet at decimal odds X returns (X − 1) units profit; a losing bet returns −1 unit. Voids/pushes return 0. We use decimal odds at the time of publication, taken from the partner sportsbook offering the best price for that market. Real-world returns will differ depending on the bookmaker used, line movement after publication, and stake sizing.
Why is the displayed table only 50 entries when you claim 12,400+ predictions?
The table shows the most recent 50 settled predictions for display performance. The full 12,431-entry historical dataset is downloadable as CSV (4.2 MB) or JSON (8.7 MB) from the links above. Both files contain identical data, with row-level SHA-256 hashes for verification. Academic researchers can also request the raw model output (pre-publication) under our research data agreement.
Have any results been independently audited?
Yes. An independent statistical audit by an external consultancy covered all predictions in Q4 2025 and confirmed: (1) reported win rate matched true settlement rate to within 0.3 percentage points, (2) no entries showed timestamp manipulation, (3) all losses were correctly recorded in cumulative P&L, and (4) closing-line value across the audited sample was +1.2%. The auditor's signed report is available on request to institutional partners.