Bottleneck Monitor

How to read this

• Regime — Price uptrend means the price is in a sustained uptrend (12-month change and level vs its 2-year average both firmly positive); Price downtrend is the mirror image. Be clear about what this is: a continuation gauge. In our copper back-test (1992–2026) entering an uptrend was followed by further gains on average (+15% over 6 months, 77% of episodes positive) — the regime label confirms a move already underway; it cannot warn you that the move is about to end. The turn-watch layers below exist for that question.
• Futures curve — the spot (cash) price versus the 3-month contract, daily. Backwardation (spot premium) means buyers pay extra for metal today — physical shortage now. Counterintuitively, that is a late-cycle marker, not a buy sign: in 2008–2026 months with meaningful backwardation were followed by below-zero average 12-month returns (27% positive), while deep contango — the capitulation state — preceded +18% average (68% positive). The market pays for tightness before the cycle turns, not after.
• Exchange stocks — metal sitting in exchange warehouses, tonnes, daily. Same late-cycle logic: stocks down sharply year-over-year means the tightness is already visible and largely priced; stocks up sharply has historically marked glut-capitulation phases that preceded recoveries. The combinations worth watching are flagged on the card: a high price with stocks turning up (top-watch) and a depressed price with stocks draining (bottom-watch).
• Producers’ capex clock — trailing-12-month capital spending of the listed producers, year-over-year, timed by filing dates. Mining supply follows investment by years, so the clock runs early: in 2006–2026 every capex upturn (2011, 2018, 2022) marked late cycle and every downturn (2015, 2020, 2025) came near a cycle bottom — six events, six correct directions, which is also why we insist it stays labeled «context»: six events is a thin base.
• Years of reserves — official world reserves divided by current annual mine production. A low or falling number means the cushion is thin; note that reserves grow with exploration and higher prices, so treat trends, not levels, as the signal.
• Positioning — the z-score of speculative net length in the futures market versus its own 3-year history. Extremes (|z| ≥ 1.5) describe a crowded trade. We tested the popular contrarian reading on copper (1992–2026) and it failed: crowded longs were followed by above-average returns, because speculative positioning tracks the trend. So we show the extreme as information about who is on board — not as a reversal predictor.
• Memory cycle phase (Semis & Memory) — memory chips have no public exchange price, so the cycle is measured through the producers themselves: their combined operating margin versus its own 10-year history (a z-score, like positioning). Boom = margins far above the norm and still rising; Rolling over = elevated but falling — in past cycles (2018, 2022) this phase coincided with the start of the downturn; Downturn / Recovery mirror it at the bottom. Extreme readings (|z| ≥ 1.5) are flagged, with the historical context shown on the card.
• Linked producers — companies from our screener universe classified in this commodity's industry. Producers carry operating leverage: a commodity price move is amplified in their margins, in both directions. Each link opens the full fundamental analysis.

Method & sources

Prices: Federal Reserve Economic Data (monthly, global benchmark series). Positioning: CFTC Commitments of Traders (weekly). Futures curve and warehouse stocks: LME daily reference data (cash vs 3-month settlement and reported stock, 2008 to date; we store and display only derived readings — spread percentage, tonnage and their changes). Producers’ capex: as-reported quarterly cash-flow statements of the listed producers named on the card. Official-sector gold: IMF International Reserves and Foreign Currency Liquidity dataset (monthly reserve gold volumes per country; we track a fixed basket of the 20 largest holders). London vault holdings: LBMA monthly vault data (gold and silver, since July 2016; we store and display derived readings — tonnage levels and their changes). Supply: USGS Mineral Commodity Summaries (annual world aggregates; refreshed automatically when each year's edition is published — if the automatic update ever fails, a notice appears on this page). Memory cycle: the as-reported quarterly financial statements of listed US memory and storage producers, aggregated revenue-weighted; only quarters where every producer has reported are used, so the gauge lags reality by roughly one reporting quarter and never relies on estimates. All series are stored and updated server-side daily; nothing on this page is edited by hand.

Turn-signal methodology (Metals — copper pilot)

The metals cards carry three layers that we back-tested point-in-time before showing them (monthly forward returns; every rule timed by the date the data was publicly available). Each rule is scored against two different assets, because they are different things: the metal price itself, and the basket of linked producers (FCX+SCCO, equal-weight, price returns) — the thing an equity investor actually holds. Producers carry operating leverage, so the same condition tends to move their stocks harder than the metal, in both directions. The record behind each layer:

• Price regime — a continuation gauge, tested 1992–2026: entering an uptrend was followed by +14.6% average over 6 months (77% of 13 episodes positive). It confirms; it does not warn.
• Positioning extremes — the contrarian reading failed the test (crowded longs preceded above-average returns, +17% avg over 12 months), so crowding is displayed as description, never as reversal risk.
• Futures curve & exchange stocks (daily, 2008–2026 test) — both work as late-cycle markers, in the opposite direction of the naive reading: backwardation preceded −4.8% average 12-month copper returns (27% positive) and below-baseline producer-stock returns, while deep contango — the capitulation state — preceded +18.4% on the metal (68%); stocks building >25% y/y preceded +17% on the metal and +27% on the producer basket. The flagged combinations: price high + stocks turning up → 35% of episodes positive over 12 months on both assets, with the producer basket roughly flat (+1.5% avg) against a +16% baseline (top-watch); price low + stocks draining → copper +13% average (77% positive), producer basket +39% average (92% positive) over 12 months (bottom-watch).
• Producers’ capex clock (2006–2026, filing-date timing) — the strongest layer, and it is stronger on the stocks than on the metal. Capex upturns (2011, 2018, 2022) preceded −10% average 12-month copper returns and −22% average producer-basket returns — 0 of 3 positive on both. Downturns (2015, 2020, 2025) preceded +26% average copper and +59% average producer-basket returns (2020: +76%, 2025: +95%). Six events, six correct directions on both assets — and six is a small number, which is why the card says «context, not a signal».

The same caveats as the memory checklist apply verbatim: few full cycles, rules examined on the same history they are scored on, no timing claim. The copper pilot is the template — other metals get these layers only after passing the same test.

Cross-metal validation (why each metal shows different layers)

We ran the same battery on gold, silver and aluminum before extending anything. The system did not transfer wholesale, and what failed is as informative as what passed:

• Price regime passed everywhere (entering an uptrend: gold +17.8%/83%, silver +28.0%/65%, aluminum +11.1%/79% forward 12 months vs their baselines) — it stays on every card.
• Aluminum keeps the tightness layer with its own numbers: backwardation −1.3% avg forward 12m (39% positive) vs +4.2% baseline; heavy stock build (+25% y/y) +27.7% avg (95%); bottom-watch +10.4% avg (86%). Its checklist is deliberately shorter: the copper-style top-watch showed no discrimination on aluminum, deep contango is this market’s normal state (two-thirds of all months), and the capex clock failed — the US-listed basket does not represent the global smelter cycle.
• Gold gets a demand gauge, not a checklist: mine supply is roughly 1.5% of the above-ground stock per year, so no supply-side clock can work — and indeed the capex clock failed the gold test outright. What binds is monetary demand: 12-month gold returns co-moved with 12-month changes in the US 10-year real yield at r ≈ −0.5 (2004–2021) and r ≈ −0.76 (2022–2026). But the direction of real rates had no forward predictive power in either era, so the card presents the real yield as concurrent context. The current divergence — gold rallying while real yields rise — is itself the message: since 2022 the official sector, not the rate-sensitive investor, has been the marginal buyer.
• Silver gets a relative-value checklist: its own COT failed, and its capex clock failed (silver mine supply is mostly a by-product of copper, zinc and gold mining). What passed is the gold/silver ratio versus its own 10-year history (1992–2026): at cheap extremes (z ≥ 1.5) silver returned +28.0% avg forward 12m (74% positive) and beat gold by +13.1% avg; at rich extremes (z ≤ −1.5) silver lagged gold in 14 of 14 episodes (−18.7% avg relative).
• Uranium has none of these inputs (no CFTC positioning, no exchange warehouse data, and the listed producers lack sufficient reported history), so its card honestly carries the price regime only.

Turn-checklist methodology (Semis & Memory)

Every row of the memory turn checklist is a discrete «message» that we back-tested point-in-time against the basket of listed US memory producers, timed by the date the public could first read each report (filing / publication dates, never quarter labels). The rules and the historical record behind each row:

• Monthly revenue trend — Taiwan-exchange monthly filings of the listed memory makers (mandatory, ~10-day lag); trend = 3-month vs 6-month average of y/y growth. In 2013–2026, deceleration onsets reached the tape 0–5 months from the producer-stock peak (quarterly gauges took 4–8 months); acceleration onsets mattered when growth was still depressed (2016, 2023, 2025 bottoms).
• Inventories — producers’ combined InventoryNet (10-Q/10-K) over trailing-12-month revenue. A build-up while margins are far above the norm arrived within 1–2 months of the stock peak in 3 of 4 cycles (2015, 2018, 2022); the 2013 exception was restocking at a cycle bottom.
• Buyers’ capex — aggregate capital spending of the four largest US cloud/AI platforms, y/y with a quarter-over-quarter trend. Past sell-offs resolved down only when this decelerated (2022); it kept growing through the 2019 scare.
• Margin momentum — producers’ revenue-weighted operating margin vs its own 10-year history (z) plus 2-quarter change. High z with falling momentum («rolling over») was followed by −22…−64% over 12 months in 5 of 7 episodes; an upturn after a fall of ≥8pp marked real cycle starts (6 of 7).
• Short positioning — FINRA short interest per producer, z vs its own ~2-year history, always shown per ticker and never aggregated (the two flags are opposite signals and fired together only once, 2026-06). Crowded shorts (z ≥ +1.5): basket +50% avg / 75% hit over 6 months (69 dates). Shorts gone (z ≤ −1.5) while the stock sits within 10% of its 18-month high: −12% avg / 13% hit (18 dates).

Caveats we insist on: each message has a small sample (4–14 events per rule — memory has had only a handful of full cycles), the rules were selected after examining the same history they are scored on, and none of them is a timing device: our separate tests show the market prices reported fundamentals in the same weeks they are published. The checklist answers one narrower question honestly: how many of the conditions that historically accompanied cycle turns are present right now? A sell-off with zero messages fired is the market front-running the cycle — a configuration that has resolved in both directions historically. The full research log (every event date and outcome) is kept internally and the computation runs server-side on official sources only.

This monitor is a qualitative research layer: it describes conditions, it does not generate buy or sell signals, and commodity regimes can persist or reverse without warning. Educational content only — see the site disclaimer.