ASSODEL Observatory v2 · MVP
Data sources
References used to populate the tables of the PRISMATIC dataset.
Dataset coverage
Methodological note
Dataset note: real PRISMATIC dataset data. Update: semiconductors SIA/WSTS through Q2 2026; memory TrendForce through Q2 2026, Q3 as a forecast; metals Westmetall/LME and FRED, monthly averages through August 2026, spot as of August 7, 2026.
Semiconductors
3 sources
Memory
4 sources
Raw materials and exchange rate
3 sources
Price increases
3 sources
Method notes
How the series are built and why they differ
- Copper, gold, tin
- Monthly LME averages published by Westmetall, from January 2024 to August 2026. Spot price as of August 7, 2026.
- Aluminum, nickel, zinc, lead, EUR/USD
- FRED (Federal Reserve of St. Louis): IMF global price indices for metals (USD per tonne) and the Federal Reserve's EUR/USD exchange rate. Methodology differs from LME cash, but is consistent within each single series.
- Why two sources for metals
- Westmetall's historical pages do not cover all minor metals homogeneously: FRED is used for these. Comparisons between metals should therefore be read as trend comparisons, not absolute-level comparisons between different sources.
- Index number Jan 2024 = 100
- The cycle comparison chart does not show dollars: it rescales each series to its own January 2024 value to make metals with prices of different orders of magnitude comparable.
- Memory contract prices
- TrendForce reports QoQ ranges. Where a single value is needed, the midpoint of the range is used, always indicating the declared bounds. Quarters marked as forecasts are not actual figures.
- Price increases
- The entities reported come from channel communications picked up by trade press. Where the manufacturer has not declared a percentage, the entry remains "not disclosed" and is not estimated.
What's coming next
How the entries on the Outlook page are built
- Two origins, never mixed
- The observatory does not make its own forecasts. Each entry declares its origin: "from data" when it arises from a rule applied to the PRISMATIC series, "from press" when it picks up a forecast statement attributed to a source.
- Entries from data
- Calculated automatically by the rules engine: they turn on and off by themselves as the series update. Each entry reports the rule applied, the numerical evidence, what it anticipates and what would confirm or disprove it at the next data point.
- Entries from the press
- Dated statements picked up from reports and trade press, always attributed to the source and with the time horizon indicated. They are added only after human review and approval: no unapproved entry is published.
Signals and scorecard
Dated snapshots, 8 deterministic rules, forecast-vs-actual verification
- Frozen snapshots
- Signals are captured at a given date and are not changed afterward. Each recalculation produces a new snapshot with the current date: the history of snapshots is the scorecard.
- The 8 data rules
- 24-month high: spot value greater than or equal to the period's maximum
- divergence: spot above the latest monthly average while the average stays below its value from three months earlier
- below peak: the maximum is not the latest point and the current value is lower
- supply tension: inventories declining for three consecutive months with the current price above the level from three months earlier
- deceleration: quarterly change declining for two consecutive periods
- acceleration: quarterly change increasing for two consecutive periods
- rotation: one series falls while a related series rises, at the same pace
- forecast revision: the new forecast exceeds the previous one
- Input series
- Monthly metal prices, monthly inventories, quarterly changes in memory contract prices, semiconductor quarterlies and 2026 forecasts.
- Editorial discipline
- Signals from data always carry a rule and a source and are never written by hand. Signals from the press are added only with "approved" status after human review.
- Verification and hit rate
- In the following quarter, the observed outcome, verification date and verdict (correct, partial, incorrect) are filled in, with optional notes. The hit rate is the share of signals with a "correct" verdict out of all verified signals: it measures, without retroactive corrections, how many signals got it right.