What We Learned From the Rally: Building a Better Tactical Precious Metals Signal

Between July 24 and August 24, gold rose 14.1% and silver 16.9% — and our Tactical Score read Cautious the entire way up. Instead of quietly changing the methodology, we replay the whole episode: what the old engine saw at each checkpoint, what the new one would have seen, what neither catches, and what a seven-year backtest told us that we would rather not have heard.

Published August 28, 2026 · Updated October 24, 2026 · 17 min read

Key Takeaways

• Between July 24 and August 24, 2026, gold rose 14.1% ($4,068 to $4,641) and silver 16.9% ($58.60 to $68.50) — and through the entire move our Tactical Score read Cautious. This article is the post-mortem, not a quiet methodology swap.
• The V1 engine was not wrong about levels: late July was genuinely hostile, with real yields at 2.43% and the bond market pricing another hike. But 75 of its 100 points measured where conditions stood, and the August turn was a story about how fast they were changing.
• Replayed week by week, Engine V2 crosses into Neutral on August 3 and its Macro Direction Indicator flips to Improving on August 5 — days into the move rather than weeks. It also misses the mid-July turn entirely and never once says Bullish, and we think both of those facts are features of an honest model rather than bugs.
• A seven-year backtest returned a finding we publish anyway: over 2019–2026 the score level was inversely related to forward returns — a regime effect that is itself the strongest argument for the fiscal-stress signal V2 adds. The direction indicator survives calibration as context, not alpha, and the backtest moved one threshold in the humbler direction.
• None of this changes what a tactical signal is for. It governs the pace and mix of new purchases. The strategic core answers to the Long-Term Structural Score — which read Bullish all summer — and no tactical reading, old engine or new, was ever an instruction to move in or out of it.

Between July 24 and August 24, 2026, gold futures rose 14.1%, from $4,068 to $4,641 per ounce. Silver rose 16.9%, from $58.60 to $68.50. Through that entire move, the Tactical Score on our Signals dashboard read CAUTIOUS, with what the dashboard then called an allocation bias of Hold, Trim on Strength. It never left the Cautious band once. On August 25 we shipped a rebuilt tactical engine, and the easy thing to do now would be to describe the new formula, say it is better, and move on. We think our readers are owed something more useful: a full replay of the episode, showing what the old engine saw at each point, what the new one would have seen, where each was reasonable, where each was late — and the parts of the rally that no macro framework, ours included, would have caught. All prices in this article are quoted in US dollars (USD) per troy ounce.

One thing this article is not: a demonstration that the new engine would have predicted the rally. It would not have. What it would have done is describe the market's changing condition faster and more honestly — and the difference between those two claims is, in a sense, the entire subject of this post-mortem.

+14.1% Gold's four-week move while the Tactical Score read Cautious. July 24 to August 24, 2026: gold $4,068 → $4,641, silver $58.60 → $68.50 (+16.9%). Across the whole period, the V1 weekly score moved only between 37 and 45 — never leaving the Cautious band.

First, the Ground Rules: What a Tactical Score Is For

Before replaying anything, it is worth restating what this signal governs, because every conclusion below depends on it. Our framework publishes two scores on two horizons. The Long-Term Structural Score asks whether precious metals deserve a strategic place in a portfolio at all; it read 66/100 and Bullish through the whole summer, and nothing in this article touches it. The Tactical Score asks a narrower question — are the next three to six months a favourable environment for adding? — and it governs the pace and mix of new purchases: how quickly planned tranches get deployed, and how far down the risk curve they go.

What Actually Happened: The Rally in One Chart

The shape matters more than any single number. Through most of July, gold drifted: $4,190 after the early-July jobs-report bounce, $4,120 mid-month, $4,068 by July 24. Silver spent mid-July below $58. Conditions were not merely uninspiring — they were measurably deteriorating, and we will show that measurement below. Then, in the first week of August, the market turned, and it turned fast: gold moved from $4,068 on the August 3 snapshot to $4,495 by August 10, and kept climbing to $4,641 by August 24. To understand what our score did and did not see, look at what the underlying macro series actually did across the period:

Read down the middle column and late July looks straightforwardly hostile for metals: real yields near their highs of the year, the dollar up, and a policy gap of +70 basis points — the two-year yield sitting well above the funds rate, meaning the bond market was pricing another hike. That is a genuinely bad tactical environment, and a score that said so was not malfunctioning. Now read across to the right-hand column. By August 24, real yields had barely moved. What changed was everything else: the hike expectation drained out of the two-year, the dollar fell roughly 2.5% in four weeks, breakevens turned up — and the term premium pushed to cycle highs while the 10-year yield stayed elevated. Yields were no longer high because growth was strong. They were high because of the deficit and the supply of Treasuries — and that particular signature has historically supported gold, not punished it.

Why the Old Engine Could Not See the Turn

The V1 Tactical Score was built almost entirely on the level of macro conditions: 75 of its 100 points came from where real yields, inflation expectations, the dollar and financial stress stood. Levels are the right foundation for a three-to-six-month view — but the August turn was a rate-of-change story, and V1 had four structural blind spots that a rate-of-change story sails straight through:

1. Level anchoring. A real yield of 2.38% scores identically whether it got there by rising steadily or by falling 30 basis points in a month. The engine had no way to distinguish a hostile environment from a hostile environment that was rapidly becoming less so.
2. Thirteen-week windows. The momentum V1 did have looked back a full quarter. Its gold-trend signal read −16.4% on July 13 — still digesting the April crash — and only +3.2% on August 24, after a +14% month. The dollar signal's 13-week window likewise buried the fresh four-week decline under stale spring data.
3. Backward-looking Fed measurement. V1 scored the change in the actual federal funds rate over three months, which was roughly zero all summer. The tradable event — hike expectations draining out of the two-year yield — was forward repricing that the signal never touched.
4. No concept of fiscal stress. V1 treated high yields as uniformly bearish for gold. It had no way to recognise August's combination — 10-year up, dollar down, term premium at cycle highs — as the "bad yields" signature that tends to accompany gold strength.

The August turn was not a story about how good conditions were. It was a story about how fast they were getting less bad.

Being fair to the old engine matters, because the temptation in any post-mortem is to flatter the successor. V1's Cautious reading in late July was a defensible description of late July. Real yields at 2.43% and a market pricing another hike are exactly what a cautious tactical posture is for. What V1 could not do was notice the moment those conditions began improving — and in a market that reprices in days, three weeks of lag converts a correct description into a wrong instruction. The review also surfaced two outright defects, which we would rather disclose than have discovered:

What V2 Changes — and What It Refuses to Change

Engine V2 keeps the 0–100 headline, the same five bands, and full per-signal transparency. Under the hood, the composition changes: the four core level signals that carried 75 points now carry 50, a new fiscal-and-Treasury-stress signal carries 10, the confirmation basket of secondary signals carries 10 — and the remaining 30 points belong to a momentum block that measures how conditions have changed over the past two to four weeks. Alongside the score, the dashboard now publishes a Macro Direction Indicator — an arrow from ↓↓ to ↑↑ — derived from the same momentum measurement.

Good Yields, Bad Yields: The New Fiscal Signal

The fiscal-stress signal deserves its own explanation, because it encodes the single most important lesson of the episode. High bond yields are not one phenomenon. Yields that are high because the economy is strong — rising real rates, a firm dollar, tight credit spreads — raise the opportunity cost of holding gold and have historically been a genuine headwind. Yields that are high because of deficits and Treasury supply look different on the tape: the term premium climbs, the dollar weakens even as the 10-year holds up, and investors begin paying for insurance against the fiscal trajectory itself. August 2026 was a textbook print of the second kind, and the new signal — term premium level, term premium trend, and a concurrence check for the "10Y up while the dollar falls" signature — scored it 7 to 9 out of 10 throughout the rally. That single component is most of the answer to a question we heard repeatedly in August: how can gold rally against 2.4% real yields? Because of why the yields were at 2.4%.

The Replay: Six Checkpoints

Now the replay itself. The V1 numbers below are what our dashboard actually published each Monday. The V2 numbers are reconstructed: the level block rebuilt from the recorded per-signal history in our production snapshots, the momentum and fiscal components computed from raw FRED and Yahoo daily data. That makes them an honest reconstruction — but also, unavoidably, an in-sample one, a caveat we will return to before drawing any conclusions.

Six weeks, six checkpoints

• Jul 20 Both engines cautious: V1 39, V2 40, direction Deteriorating. Real yields 2.43%, the bond market pricing another hike. Gold at $4,138 — about to gain 10.2% in four weeks.
• Jul 29 The daily direction exits Deteriorating: The momentum index D reaches −0.20 and crosses back to Stable. Nothing visible yet in the weekly cadence.
• Aug 3 V2 crosses into Neutral: V1 44 (still Cautious), V2 46 — the first band crossing of the episode. Gold still at $4,068.
• Aug 5 The arrow flips to Improving: D = +0.24. Listed drivers: Fed repricing, the dollar, breakevens. Two days later D reaches +0.38.
• Aug 10 The divergence peaks: V1 41 versus V2 53 — twelve points apart on identical inputs. Gold $4,495, silver near $64.
• Aug 24 Strong move, careful label: V1 45, V2 55 — Neutral with an Improving arrow (D = +0.47). Gold $4,641, up 14.1% in four weeks. The score never says Bullish.

July 20: Both Engines Say Cautious — and Both Miss What Comes Next

On July 20, V1 read 39 and the reconstructed V2 read 40 — materially the same message — with the direction indicator pointing down. This was not a malfunction in either engine. Measured macro conditions genuinely were deteriorating into late July: real yields rising, the dollar firming, the policy gap widening to +70 basis points. Gold sat at $4,138 and silver at $57. Over the following four weeks, gold returned +10.2% and silver +16.4%. Neither engine saw it coming, and it is important to be precise about why: the price turn led the macro turn by one to two weeks. The market began moving before the data that any macro-input framework consumes had turned.

August 1–3: The Quiet Crossover

At the start of August there was still nothing dramatic on the tape — gold closed the August 3 snapshot at $4,068, almost exactly where it had been ten days earlier. But underneath, the repricing had begun. The daily momentum index had crossed out of Deteriorating on July 29; by August 3 the reconstructed V2 score reached 46, crossing into the Neutral band. V1 read 44 the same day — its best print in months, but still Cautious, and it would remain Cautious for the rest of the episode. The crossover is easy to shrug at: what practical difference is there between 44 and 46? For a single day, almost none. The structural difference is that one engine had begun registering improvement while the other, three weeks later and 14% higher, still had not changed its message.

August 5–7: The Arrow Flips

August 5 is the day this upgrade was built for. The blended momentum index reached +0.24, and the Macro Direction Indicator — had it existed — would have flipped to ↑ Improving, listing Fed repricing, the dollar and breakevens as its strongest drivers. By August 7 the index reached +0.38. Meanwhile the fiscal-stress signal was scoring 7–8 out of 10, supplying the explanation a reader would actually need: yields were high for fiscal reasons, not growth reasons, and gold was responding accordingly. Gold spent those days mid-leap between $4,068 and $4,495. The value of the arrow here is not prediction — the move was already underway. It is that the dashboard would have described the change within two or three days of it beginning, instead of leaving a reader to reconcile a Cautious label with a running rally for three more weeks.

Aug 5 The day the Macro Direction Indicator would have flipped to Improving. D = +0.24, with Fed repricing, the dollar and breakevens as the listed drivers — two or three days into a move the weekly V1 score never acknowledged at all. V1 did not leave the Cautious band at any point in the episode.

August 10–12: The Divergence Peaks

By the August 10 snapshot, gold had reached $4,495 and silver was near $64 — the week Rick Rule's silver rotation was in the news. The two engines now told opposite stories from identical inputs: V1 read 41, actually falling three points on the week as its 13-week windows churned through stale spring data, while the reconstructed V2 read 53. Twelve points is the widest gap of the episode, and it is worth being precise about what it represents. It is not that V2 was more bullish. It is that V2 was measuring the four weeks that had just happened, while V1 was still averaging them against a quarter that no longer described the market.

August 17–24: Strong Move, Careful Label

Through the back half of August the reconstructed V2 held in the low 50s with an Improving arrow, and on August 24 — gold at $4,641, the top of our replay window — it read 55, Neutral, with the momentum index at +0.47. Here is the part some readers will find unsatisfying: the score never said Bullish. At no point in a +14% month did the new engine flash a green light. We think that is correct, and the reasoning matters. Real yields still stood at 2.38%. A 55 with a firmly improving arrow says precisely that: the environment is middling, and it is getting better quickly. In V2, the direction indicator — not the level — carries the tactical message, and "Neutral and improving rapidly" is a different, more honest instruction than "Bullish". If future review concludes that reading is too timid, the lever we would reach for is the fiscal signal's weight, not a bigger momentum block.

Read the July rows and the forward-return columns together, and the table refuses to flatter anyone: the weeks with the most bearish readings — from both engines — preceded the strongest four-week returns of the year. The table also shows what the upgrade actually buys. It is not clairvoyance. It is the difference between the August 10 row reading 41 or reading 53 while gold trades at $4,495.

What We Deliberately Did Not Claim

A replay like this comes with a built-in trap: the V2 parameters were designed by people who had already watched July and August happen. Tuned on the very episode it is being graded on, of course the new engine looks better. That is why the honest summary of the replay has three parts, not one:

• Where V2 genuinely helps: once conditions turned in early August, it registered the change within days — band crossover on August 3, direction flip on August 5 — while V1 never changed its message at all. Its June readings were also better: lower scores and a deteriorating arrow ahead of a month in which gold fell roughly 10% from its early-June levels.
• Where V2 fails identically: the mid-July checkpoint. A price turn that leads the macro turn is invisible to any macro-input framework, and we prefer a framework that admits that boundary to one that quietly chases price.
• Where the claim itself is weak: one month of in-sample replay proves almost nothing. Which is why we did not ship the engine on the strength of this episode — we made a seven-year backtest the gate.

Then We Ran It Against Seven Years

Before V2 went live, we scored every week from July 2019 through August 2026 — 374 weekly observations — with the full V2 formula: level, fiscal, confirmation basket and momentum, thirteen of the fourteen signals (speculative-positioning data does not extend far enough back to include). Forward returns came from actual gold and silver futures history. We were looking for two things: whether the momentum thresholds calibrated sensibly out of sample, and whether the relationship between score and subsequent returns actually holds. The second question produced the most uncomfortable table we have ever published.

Over this sample, the score level is inversely related to forward returns: the correlation between the total score and 13-week-forward gold returns is −0.25, and for the level block alone it is −0.38. Cautious weeks out-returned Bullish ones. Before concluding that the whole enterprise is upside down, look at where the high-score weeks live: 22 of the 24 Bullish-band readings are from 2020–2021, when deeply negative real yields produced high level scores immediately before gold's long 2021–2022 consolidation. And the strongest forward returns came from low-score weeks in 2024–2026 — the fiscal-era regime in which gold repeatedly rallied against high real yields. The finding is a regime story, not noise, and it is the strongest empirical argument for exactly what V2 did: cut the classic level signals from 75 points to 50, add a fiscal-stress signal that scores the "yields high for fiscal reasons" regime as supportive, and read direction alongside level. It is also why the backtest table on the Signals page now carries this caveat in plain sight rather than being quietly filtered.

What the Direction Indicator Is — and Is Not

The direction indicator fared better, within strict limits. Grouping all 374 weeks by direction state:

Improving states beat Stable at the four-week horizon, and the silver split is the cleanest in the table. But the overall correlation between the momentum index and four-week-forward gold returns is +0.07 — mild, correctly signed, and gone entirely by thirteen weeks — and the deteriorating side is muddied by mean reversion, since weeks of deteriorating conditions often immediately precede rebounds. Our own conclusion, now printed on the dashboard itself: the arrow is context, not alpha. It tells you where conditions are heading. It does not tell you where price is going, and we will not pretend otherwise.

The calibration run changed exactly one parameter. We had drafted the "Rapidly" thresholds at ±0.40; over seven years that fired on 34% of all weeks, and the boundary between ↑↑ and ↑ separated nothing. The shipped engine uses ±0.50, making a Rapidly state a genuine one-in-ten-weeks event. That change has a cost we accept: under the calibrated threshold, August 24's momentum reading of +0.47 earns ↑ Improving rather than ↑↑ Improving Rapidly. The backtest took away our best headline, and we kept the calibrated label. Just as deliberately, two things did not change: we did not invert the real-yield brackets despite the negative correlation — one macro cycle is far too small a sample to flip a causal prior into its opposite, and the fiscal signal already conditions on the regime that drove the inversion — and we did not raise the momentum block above 30 points, because a +0.07 correlation does not justify more.

A backtest that only ever confirms the model you already shipped is not a test. This one cost us our best headline, and we published it anyway.

How to Use the New Signal (Without Overusing It)

The practical question is what a reader should actually do differently now that the dashboard shows a level score and a direction arrow instead of a single weekly number. The answer is: the same things as before, at a better-informed pace.

Run the worked example. A reader following the two-horizon framework in early August, seeing "Neutral, direction Improving, drivers: Fed repricing, dollar, breakevens", does nothing dramatic — and that is the point. They resume the tranche schedule they had slowed during July, keep new purchases tilted toward gold and quality producers, and leave the core untouched, exactly as our August 18 piece on chasing the price argued from the behavioural side. What the new engine removes is the specific failure mode of August 2026 under V1: a reader watching gold run 14% against an unchanging Cautious label, losing confidence in the framework, and then buying the top out of frustration precisely because the model had stopped describing reality. Our DCA calculator totals the ounces, average cost and paper gain of a regular buying schedule from the figures you enter.

What Changed on the Signals Page

• The Tactical Score is calculated every weekday at about 20:00 UTC — the weekly Monday cadence is retired. Each input shows the date of its own reading, which is older than the run: FRED's daily Treasury series are two business days old when they are scored.
• The Macro Direction Indicator is published beside the headline score, with its two or three strongest momentum drivers listed beneath it, and a two-publish confirmation rule so a single day's data cannot flip the arrow.
• Every signal carries a staleness flag: when an input is a cached or fallback value rather than a fresh fetch, the dashboard says so with a visible badge — the silent-fallback failure of July 27 cannot recur unannounced.
• The score-history chart marks the V1→V2 boundary explicitly, so no one mistakes the two formulas for one continuous series.
• The dead money-market signal is removed, and the backward-looking Fed-policy signal is replaced by the 2Y policy gap — the forward-looking version of the same idea.
• The Backtested Signal Performance section now shows the V2 formula's own reproducible output, the direction-state table, and the regime caveat from this article — including the uncomfortable parts.

The Bottom Line

The honest summary of the July–August rally is that our original tactical model gave a reasonable description of macro levels and a poor description of macro change — and in a fast market, that difference compounds weekly. Replayed through the same period, the rebuilt engine crosses into Neutral on August 3 and flips its direction arrow on August 5, days into the move; it also reads deteriorating in mid-July immediately before the turn, and it never once calls the rally Bullish. We have shown you all three of those facts at the same resolution, because a signals page that only replays its wins is an advertisement, not a tool.

Market relationships change — the 2019–2026 backtest shows the classic real-yield relationship inverting inside a single sample, which is precisely why frameworks have to keep being tested against the market rather than defended from it. We will run the same backtest annually, publish what it says, and change what it justifies changing — and nothing more. In the meantime, the current readings, the direction arrow, the per-signal detail and the full backtest table are on the Signals dashboard. If the episode left you with one habit, make it this: let the tactical readings set your pace, and let only the long-term thesis — not a strong month in either direction — decide whether the core position exists.