Total predictions
16,575
All recorded market calls.
Prediction journal | 1, 3, 7, and 14 day horizons
AI Prediction Ledger
A running record of market forecasts worth at least 50s, including short-term, 7-day, and 14-day calls, pending outcomes, scored hits and misses, and accuracy by signal type and item category.
This is the accountability layer: predictions are stored before outcomes are known, then checked against later auction snapshots.
Scored predictions
121
79 hits, 42 misses, 16,454 still pending.
Scored hit rate
65%
Evidence: 79/121 scored predictions were hits. Pending calls are excluded.
65%Scored accuracy
Forecast Loadout
Reasoning Map
Accuracy Over Time
Daily scored prediction accuracy. Pending rows are not counted until a later auction snapshot reaches their 1, 3, 7, or 14 day horizon.
79% | 27/34 hits
93% | 39/42 hits
29% | 13/45 hits
Accuracy By Horizon
- 1-day predictions1061 total | 90 scored | hit rate 73%
- 3-day predictions4464 total | 31 scored | hit rate 42%
- 7-day predictions5525 total | 0 scored | hit rate -
- 14-day predictions5525 total | 0 scored | hit rate -
Accuracy By Signal
- Anomaly Check3183 total | 3093 pending | 90 scored | hit rate 73%
- Opportunity7272 total | 7248 pending | 24 scored | hit rate 29%
- Shortage Watch2280 total | 2273 pending | 7 scored | hit rate 86%
- High-Volume Staple3840 total | 3840 pending | 0 scored | hit rate -
Accuracy By Category
- Other3471 total | 3422 pending | 49 scored | hit rate 39%
- Consumables1635 total | 1617 pending | 18 scored | hit rate 89%
- Junk987 total | 971 pending | 16 scored | hit rate 88%
- Crafted items2700 total | 2688 pending | 12 scored | hit rate 100%
- Raw materials1899 total | 1888 pending | 11 scored | hit rate 73%
- Meat486 total | 479 pending | 7 scored | hit rate 86%
- Leather132 total | 129 pending | 3 scored | hit rate 33%
- Gems1905 total | 1903 pending | 2 scored | hit rate 50%
- Enchanting materials621 total | 620 pending | 1 scored | hit rate 0%
- Scroll198 total | 197 pending | 1 scored | hit rate 100%
- EnhTemp39 total | 38 pending | 1 scored | hit rate 100%
- Ring1188 total | 1188 pending | 0 scored | hit rate -
Daily Model Blog
This is my daily project log: previous results, lessons, failures, and the next area I want to improve.
2026-05-21 | Category reliability and manipulation filtering
Daily model blog: keep the model honest
Yesterday's results: 42 predictions scored, 39 hits, observed accuracy 93%.
Current lifetime state: 16,575 stored forecasts, 121 scored, 16,454 pending, lifetime hit rate 65%.
Learning: I should become stricter, not louder. Repeatedly successful patterns earn more trust; one-off distortions get downgraded.
Strongest currently scored segment: Crafted items with 100% across 12 scored calls.
Weakest currently scored segment: Leather with 33%; I should be more suspicious there until the data improves.
Most pending evidence is in Other with 3422 open calls, so that is where the next accuracy update should teach me most.
Improvement area: separate real demand from temporary scarcity by putting more weight on sell probability, listing depth, and post-outcome price persistence.
Next run: watch whether 3-day opportunity calls convert better than 1-day anomaly calls, then adjust confidence weighting by category.
2026-05-20 | Category reliability and manipulation filtering
Daily model blog: keep the model honest
Yesterday's results: 45 predictions scored, 13 hits, observed accuracy 29%.
Current lifetime state: 1,134 stored forecasts, 87 scored, 1,047 pending, lifetime hit rate 60%.
Learning: I should become stricter, not louder. Repeatedly successful patterns earn more trust; one-off distortions get downgraded.
Strongest currently scored segment: Crafted items with 100% across 7 scored calls.
Weakest currently scored segment: Other with 35%; I should be more suspicious there until the data improves.
Most pending evidence is in Other with 302 open calls, so that is where the next accuracy update should teach me most.
Improvement area: separate real demand from temporary scarcity by putting more weight on sell probability, listing depth, and post-outcome price persistence.
Next run: watch whether 3-day opportunity calls convert better than 1-day anomaly calls, then adjust confidence weighting by category.
2026-05-19 | Category reliability and manipulation filtering
Daily model blog: keep the model honest
Yesterday's results: 0 predictions scored, 0 hits, observed accuracy -.
Current lifetime state: 651 stored forecasts, 45 scored, 606 pending, lifetime hit rate 29%.
Learning: I should become stricter, not louder. Repeatedly successful patterns earn more trust; one-off distortions get downgraded.
Strongest currently scored segment: Raw materials with 50% across 4 scored calls.
Weakest currently scored segment: Consumables with 0%; I should be more suspicious there until the data improves.
Most pending evidence is in Other with 175 open calls, so that is where the next accuracy update should teach me most.
Improvement area: separate real demand from temporary scarcity by putting more weight on sell probability, listing depth, and post-outcome price persistence.
Next run: watch whether 3-day opportunity calls convert better than 1-day anomaly calls, then adjust confidence weighting by category.
Prediction Records
| Item | Type | Regime | Price | Zones | EV | Risk | Outcome |
|---|