Transparency

Honest backtesting of the AI engine. Hits and misses published unfiltered.

Honesty

This page is a public reckoning. Every daily narrative and every dip classification is measured ex-post against actual price behavior. We don't hide the misses: if the engine was wrong, it shows up here with the same weight as the hits.

BurMarket is NOT a registered investment advisory firm (no CNMV/EAF registration). The metrics you see below are reference information, not financial advice. Their purpose is to let you judge the engine's quality before deciding how much weight to give its analyses.

Data is recomputed every Monday (weekly cron). If a metric looks suspicious, contact us: I'd rather correct it than hide it.

Opportunity score vs realized return

Do the assets we score high actually outperform the low-scored ones? Real data by bucket × horizon.

Score	Horizon	n	Avg return	Median	% positive
medium (5-6)	3d	30	1.75%	1.64%	73.3%
medium (5-6)	14d	30	1.74%	2.22%	70.0%
high (7-10)	3d	24	1.42%	1.16%	75.0%
high (7-10)	14d	24	0.95%	1.50%	70.8%

n = number of narratives measured. Return computed in EUR against the first available trading day at narrative_date + horizon. Hit rate = fraction with return ≥ 0%.

Dip classification

When the engine classified a drop, did reality agree?

Technical, no damage

38 classified events

Recovered to reference price: 0/38

Hit observation stop-loss: 0/38

Digestible news

5 classified events

Recovered to reference price: 0/5

Hit observation stop-loss: 0/5

Thesis broken

2 classified events

Recovered to reference price: 0/2

Hit observation stop-loss: 0/2

Methodology

Narratives are generated daily at US close (22:00 UTC). The 1-10 opportunity_score is assigned by Sonnet based on technical metrics + sensitivity profile + the day's market pulse.
For each narrative, we record the asset's EUR close on narrative_date as reference. After 3, 14, and 28 days, we compare against the next available close.
Dip classifications fire when an asset drops ≥3% in 1 day or ≥7% in 5 days. Sonnet classifies honestly between tecnica_sin_dano, noticia_digerible, and ruptura_tesis (the last one even when it hurts). Manuel reviews weekly and records disagreements.
Raw data lives in the public tables asset_ai_narratives, narrative_outcomes, dip_events, and dip_outcomes with public-read RLS.