Level to level, with the odds.

Every leg the model wrote, with the statistics it cited. A leg is the probability that its level comes next, given the legs before it; where two paths share a fork, the leg carries one probability on both. The checker refuses a level that is not on the operator's ladder, weights that do not sum to one, the same fork priced two ways, a leg faster than the tape allows, and paths that stop short of the year; a refused answer is asked again with the reasons, and the second answer is the record either way.

The day files

Every day's forecast is a JSON file, signed with the forecast key and carrying the hash of the previous day. series.json · latest.json · public key · a day: /forecast/days/YYYY-MM-DD.json.

The verifier

bgr-forecast verify --out-dir <a copy of the files> recomputes every hash, checks every signature and the chain.

The model

—

The rules it is given

The operator's trading protocol in full, and the operator's rules for how price moves level to level: volume levels over time levels over mathematical ones; every turn is a sweep of an unfinished zone and a reclaim, every continuation a break and retest; value rotates VAL to POC to VAH and a range is left by sweeping its bad extreme; the price, open-interest and CVD matrix for direction; CME decides whether a level is swept; and the previous cycles' structure from this same stage as the template a path is built on.

Drawn how

The model gives each level, the day it expects it, and the shape of the leg: a sweep through the level and reclaim, a break and retest, a trend with pullbacks, a range with both ends swept, or a direct rotation. The line between two levels is drawn by that shape, deterministically — no random noise. A range of six weeks rotates twice, a range of six months a dozen times; a trend pulls back every few weeks. The band is the same path drawn half-way to the model's early and late arrival days, so it is the timing that is uncertain, not an invented price range. The levels, the days and the shapes are the model's; the ink between the turns is drawing.

Not a price target and not a trade signal. Twenty thousand year-long paths are simulated from today's close inside the envelope fifteen years of cycles put around it, re-weighted by how the drawn levels behaved in seven years of tape, and gathered into routes by where they are every thirty days. A route's probability is its share of the paths; the line drawn is a real simulated path, the one nearest the route's middle, named by the drawn level it is closest to at each checkpoint; its colour along the way is how much of the whole ensemble is still within reach of it. The envelope is scored against the plain walk every run, below, and the score is published whether it wins or loses.

The levels price is at or near now, and what the simulated paths did there in the first touches, by the miner's own rules (a hold is no two daily closes beyond the level within twenty days; a bounce reaches the next level on the side it came from while held; giving way reaches the next level on the far side).

Matched on the last 120 days' shape, the last touches of the drawn levels, position in value, RSI, momentum and volatility, and orderflow. "Where the ensemble went" is the share of today's simulated paths inside a widening band around that day's actual path at 30, 90, 180 and 365 days.

The day files

series.json · latest.json · public key · a day: /forecast/engine/days/YYYY-MM-DD.json.

The process

Paths are a lognormal at the trailing year's realised volatility, with hourly steps so a day has a high and a low, drifting toward where the last three cycles sat at this point after their peak and reflected at the edge of where they ranged, with the height above trend shrunk by the measured decay of the cycles' returns. Levels change the odds by a few points, measured against random levels and against the same walk.