Audit trail

Hypothesis: After a long run of dead ends, a different kind of question — not “which way will gold go,” but “is the volatility the market expects in gold consistently priced higher than the volatility that actually shows up?” That gap is the premium a seller of protection collects. We measured 18 years of it, 2008 to 2026.

For the first time in the whole programme, the answer was yes — and clearly. Expected volatility ran about 2.7 points above what actually materialised, was positive in roughly four windows out of five, and was far too strong and persistent to be luck. But we are publishing this as a lead, not a win — and that distinction is the entire point of how we work. A measured premium is not a tradable profit: collecting it needs options, a venue, and real option costs, and its danger lives in the rare violent spike that can hand back months of premium in a day. Whether it survives that is the real product question. So it goes to a written-in-advance harvest test, costed against the real world, before anyone calls it an edge. It is the most promising thing we have found — and it has proven nothing yet.

The 18-year volatility-premium measurement and its control checks — available for review

Hypothesis: Gold reprices hard around scheduled economic releases — jobs (NFP), inflation (CPI), and Fed decisions — where dealers and funds are forced to re-hedge on a known clock. We tested every route to a tradable move: direction (from price, and from the real-rate reaction), fading the first spike, and a let-winners-run trailing exit — across ~80 events, including ones never studied before.

Every route died once real trading costs were counted. No direction was predictable from price or from rates; fading the spike lost at every speed; and even a disciplined let-winners-run exit — which does catch gold’s occasional big event move (a 32% win rate with wins nearly 2× the size of losses) — still bled out on costs, including on the fresh CPI and Fed events we’d never touched. Our controls confirmed the test was honest (it recovered a planted edge decisively and stayed flat on placebo days). What this does NOT close: the volatility-premium side — how much event “insurance” is overpriced — which we can now measure with free data. That’s the next thread.

Full event-campaign boards (Sets 1–5b) and written verdict — available for review

Hypothesis: The strongest form of the price-only question: take a family of simple, rule-based directional signals — each one individually too weak to trade — and only act when several of them agree, across gold and the four major currency pairs. If conviction from agreement could clear the cost of trading, this was the test built to show it.

It didn’t. Across all five markets, only a single signal in one market (gold) even qualified — and one is not enough to trigger the “several agree” rule, so the engine never placed a trade. This was the most complete, most powerful version of “can you forecast direction from price alone, one decision a day?” we could build — and its failure is a real answer: we’re closing **continuous, daily price-only forecasting** on the major markets. (Our next build is an intraday engine — a different question, not yet answered.) No combination cleared the floor our earlier tests set, roughly 0.0002 per trade after real spread. Our own controls confirmed the test was honest — it recovered a signal we planted on purpose and stayed flat on pure noise.

Full family board (all 16 signal cells) and written verdict — available for review

Hypothesis: The setups nearly every retail trader is taught — double tops and bottoms, head-and-shoulders, triangles, flags — and the “smart money” / order-flow strategies (liquidity sweeps, order blocks, fair-value gaps). Do any of them actually beat a coin toss once you count trading costs? We tested the whole catalogue on both Deriv’s synthetic indices and on real gold and the major currency pairs, from one-minute to one-hour charts.

None of them worked. Not one chart pattern and not one “smart money” concept beat a random walk after real costs — across roughly 775 tests, on synthetics and on real markets alike. That last part is the important one: the usual objection is “sure, nothing works on synthetics, but real markets are different.” They aren’t — the same setups failed on real gold and real FX too. Our controls confirmed the search was honest: it caught a bias we planted on purpose, and it found nothing where there was nothing to find.

All ~775 pattern & smart-money results across four timeframes, and the written verdict — available for review

Hypothesis: Two timing strategies — riding the London open, and fading the New York close — tested across five markets (gold, EUR, GBP, AUD, JPY).

Ten tests in all. Zero survived. Fading the New York close actually lost money in every one of the five markets once real trading costs were counted; riding the London open was a flat dead end. Nothing earned more than it cost to trade. And our own controls confirmed the test was honest — it never invented an edge that wasn’t there.

Full test board (all 10 cells) and written verdict — available for review

Hypothesis: Is there a tradable edge — or an exploitable flaw — hidden in the way Deriv’s synthetic indices are built?

We threw everything at it: 394 separate tests across 38 of Deriv’s synthetic instruments, in 25 different ways — spike timing, randomness, cross-instrument links, multi-timeframe structure, and more. Not one found a tradable edge or an exploitable flaw. Our controls confirmed the search itself worked — it found nothing where there was nothing to find, and it caught a test signal we planted on purpose. Closed for good, unless a genuinely new claim about how these instruments are built comes along.

All 394 test results and the written verdict — available for review

Hypothesis: A gold system built around a drift in specific trading sessions.

In the backtest, this one looked good enough to ship. Then our own process caught a flaw in how we’d set the test up — a subtle way the result had been flattered. We caught our own bad backtest. Run cleanly, it lost money once real trading costs were counted. Killed — and we don’t give a dead strategy a second, easier trial.

Full test board and written verdict — available for review