Continuation
43%
beat no-change
70%
in band · target 68%
135,105 forecasts
Palanor Foresight™ · Track record
Every forecast, scored in the open.
Every forecast Palanor shows on a signal is logged when it is made and scored when its date arrives. Nothing is edited after the fact. The measure is plain: was the forecast closer to what happened than simply assuming nothing would change, and did the outcome land inside the band we drew around it?
Hold-out test
261,824 forecasts across 1,718 signals, rebuilt from history cut at past dates. They use today's revised data and today's lattice, so they read a little kinder than live results will.
Palanor Foresight™
51%
beat no-change
70%
in band · target 68%
126,719 forecasts
Live
6,739 live forecasts scored so far, logged as they were shown. Continuation: 47% beat no-change. Palanor Foresight™: 45% beat no-change.
| Cadence | Checkpoint | Mode | Forecasts | Signals | Beat no-change | In band | Median error | No-change error |
|---|---|---|---|---|---|---|---|---|
| daily | 7 days | Continuation | 3,134 | 1,587 | 48% | 82% | 1.38 | 1.36 |
| daily | 7 days | Palanor Foresight™ | 2,884 | 1,464 | 45% | 84% | 1.38 | 1.39 |
| weekly | 7 days | Continuation | 357 | 179 | 38% | 76% | 0.80 | 0.84 |
| weekly | 7 days | Palanor Foresight™ | 348 | 175 | 40% | 77% | 0.86 | 0.86 |
| monthly | 7 days | Continuation | 8 | 4 | 25% | 75% | 0.59 | 0.59 |
| monthly | 7 days | Palanor Foresight™ | 8 | 4 | 75% | 100% | 0.39 | 0.59 |
Hold-out test, in detail
By cadence and checkpoint. Errors are scaled by each series' typical step, so signals of different sizes can be compared; lower is better.
| Cadence | Checkpoint | Mode | Forecasts | Signals | Beat no-change | In band | Median error | No-change error |
|---|---|---|---|---|---|---|---|---|
| daily | 7 days | Continuation | 37,891 | 1,545 | 41% | 74% | 2.05 | 1.93 |
| daily | 7 days | Palanor Foresight™ | 36,447 | 1,487 | 47% | 72% | 1.96 | 1.96 |
| daily | 30 days | Continuation | 33,255 | 1,541 | 43% | 73% | 3.82 | 3.69 |
| daily | 30 days | Palanor Foresight™ | 32,201 | 1,482 | 48% | 63% | 3.70 | 3.72 |
| weekly | 7 days | Continuation | 11,899 | 465 | 40% | 79% | 0.95 | 0.91 |
| weekly | 7 days | Palanor Foresight™ | 11,732 | 458 | 47% | 74% | 0.92 | 0.92 |
| weekly | 30 days | Continuation | 10,519 | 464 | 39% | 80% | 1.50 | 1.41 |
| weekly | 30 days | Palanor Foresight™ | 10,369 | 457 | 51% | 68% | 1.41 | 1.42 |
| monthly | 7 days | Continuation | 9,089 | 464 | 49% | 64% | 1.09 | 1.17 |
| monthly | 7 days | Palanor Foresight™ | 7,865 | 429 | 57% | 75% | 0.99 | 1.10 |
| monthly | 30 days | Continuation | 7,850 | 462 | 49% | 64% | 1.26 | 1.33 |
| monthly | 30 days | Palanor Foresight™ | 6,759 | 427 | 59% | 74% | 1.12 | 1.26 |
| quarterly | 90 days | Continuation | 3,098 | 167 | 51% | 63% | 1.61 | 1.76 |
| quarterly | 90 days | Palanor Foresight™ | 1,550 | 163 | 61% | 76% | 1.38 | 1.68 |
| quarterly | 1 year | Continuation | 2,564 | 163 | 53% | 58% | 2.67 | 2.78 |
| quarterly | 1 year | Palanor Foresight™ | 1,342 | 159 | 63% | 67% | 2.29 | 2.71 |
| annual | 1 year | Continuation | 15,684 | 1,718 | 46% | 59% | 1.41 | 1.44 |
| annual | 1 year | Palanor Foresight™ | 15,204 | 1,657 | 59% | 73% | 1.33 | 1.44 |
| annual | 3 years | Continuation | 3,256 | 1,170 | 46% | 55% | 2.16 | 1.99 |
| annual | 3 years | Palanor Foresight™ | 3,250 | 1,168 | 60% | 67% | 1.73 | 1.99 |
Engine changes
Every change to how a forecast is made, dated. The weights that blend a signal's trend, pattern and lattice read are re-fitted weekly on the older half of this record and kept only when they do better on the newer half; a change that clears those gates is listed here the day it takes effect. Forecasts made before a change are scored as they were made, never rebuilt.
- Sep 26, 2026Quarterly and annual projections damped to what held up on past data.2026-09-26.2
- Sep 26, 2026Foresight v3: one blend of trend, pattern and lattice, sized per cadence on the hold-out record.2026-09-26.3
- Sep 26, 2026Market prices hold the last value; expiring contracts are never partners.2026-09-26.4
- Oct 6, 2026Monthly and quarterly series read a monthly lead lattice: signals that led them, with measured skill as the weight.2026-10-06.1
- Oct 7, 2026Quarterly series: weights re-sized by the weekly tune, 80% trend, 0% pattern, band ×2 → 50% trend, 30% pattern, band ×1.5.tune
How the ledger works
- Logged when made. Once a week, every public signal with at least 20 observations and a recent print gets its forecasts recorded: every mode the signal page offers, exactly as it draws them, with the band and the engine version.
- Checkpoints follow the series. Daily, weekly and monthly signals are checked at 7 and 30 days. Quarterly signals at 90 days and 1 year. Annual signals at 1 and 3 years.
- Scored against what was published. The outcome is the observation nearest the target date, within half the series' spacing (at least three days). Nothing is interpolated. When no observation falls in that window, the forecast is voided with the reason, never scored against a guess.
- Against no change. Every forecast is compared with the simplest alternative: the last value, carried forward. A forecast that cannot beat it more than half the time is not earning its place, and this page will say so.
- The band is tested too. Each forecast carries a ±1σ band. If the band is honest, about 68% of outcomes land inside it. Much more means the band is too wide; much less means it is too sure.
- Hold-out tests are labelled. Before live results accumulate, forecasts are rebuilt from each series cut at past dates, holding back its usual publication delay. They use today's revised data and today's lattice, which flatters them slightly. They are shown apart from live results and never mixed in.
- Licensed data stays fenced. Market prices are scored with everything else, but only these aggregate shares are published.
- The record is used, not just published. A series on which Foresight has not beaten no change over 26 weeks (at least 20 scored forecasts, under 40% beating it, median error at least 10% worse) is held at its last value for 13 weeks and then tested again; its forecast says so. The weekly re-fit of the blend weights is listed above under Engine changes.
Engine version 2026-10-06.1 · refreshed every 6 hours
Forecasts are statistical projections, not predictions of fact and not financial advice. Past and hold-out results do not guarantee future accuracy. See the Forecast disclosures.