How Cambent forecasts the TKX

Direction of the TKX

A higher TKX means AI capability costs less. Each reading compares with the Aug 4, 2026 baseline of 100.

Experimental or citeable

Every forecast run is in exactly one state: Experimental or Citeable. The state is computed automatically from the run's evidence.

Experimental (default at launch, and for any run that misses the citeable bar)

• Appears on the Forecast tab, labeled Experimental, with uncertainty bands shown.
• Presented as an experimental estimate, separate from official Cambent numbers.
• Kept out of the citeable TKX readings on the live index, the Index API, press, and any "Source: Cambent TKX" citation.
• Language: projected, model-implied, experimental estimate — always with the range.

Citeable (earned on each run)

• Reached when every citeable-bar criterion below passes for that run.
• Can appear on the Forecast tab without the Experimental label, and in the citeable TKX readings, the Index API and citations as an official Cambent forecast.
• Still range-first: still a projection with an interval.

What earns citeable

A forecast is citeable when all of the following are true for that run. If any fails, the forecast stays Experimental.

1. Sound index — the forecast trains on the locked, verified index methodology.
2. Enough comparable history — at least 90 daily index values under the current method.
3. Backtest beats naive — the model beats simple last-value and drift baselines out of sample on the target horizon.
4. Calibrated intervals — the bands cover outcomes about as often as they claim (starting default: within about ±10 percentage points of nominal).
5. Fresh close — the run used a finalized official daily close no older than the freshness default (starting default: 2 days).
6. Confirmed series — the forecast trains on the confirmed TKX series; provisional market-impact estimates stay out of training.

Citeable is earned by the data on each run. There is no button to mark a forecast citeable by hand.

Intervals and bands

• A projection is always shown with its uncertainty band.
• Allowed words: projected, model-implied, estimate, range, band, interval.
• Even in Citeable state, the band stays. The midpoint is the center of a projected range.

Boundary handling

What you see when something goes wrong.

Situation	What the product does
Forecast job fails or errors	Keep the last successful forecast, shown with its date (e.g. Last updated [date]). A partial or broken run stays unpublished.
Official daily close missing or awaiting finalization	Skip the new run. Keep the prior forecast, shown with its date (e.g. Last updated [date]).
Too little comparable history	The Forecast tab says insufficient comparable history. The forecast stays Experimental until there is enough comparable history.
Method change in the comparable history	When the method changes, a forecast can use history gathered since the change, plus any earlier stretch that has been rebuilt on the new method and labeled as rebuilt. A run that mixes in older history still on the old method is left unpublished, and the last forecast stays up with its date. If the usable history is too short, the forecast stays or drops to Experimental. The change is marked when the forecast is shown.
Unusually large jump vs prior run	Send the new run to review. The prior forecast stays current until review clears the new run. The Forecast tab marks a run under review, and any replacement is shown with its date.
Interval calibration falls below the citeable bar	Automatically drop Citeable → Experimental, and off the citeable TKX readings, until the bar is met again. The state is recomputed on each run.

Ruled out:

• A partial or errored forecast presented as official
• An Experimental forecast among the citeable TKX readings or in "Source: Cambent TKX"
• A point forecast without its band
• A forecast marked Citeable by hand

Limitations

• The forecast models capability-adjusted index levels, separate from vendor price sheets and your own invoices.
• Bands describe uncertainty; future TKX readings can fall outside them.
• Experimental forecasts are for exploration on the Forecast tab.
• Citeable forecasts are official and still uncertain: each is a projection with its interval.

The citeable readings (today's confirmed TKX, TKX Open and TKX Frontier, each with its as-of date) stay on the live index. A forecast can join them once it is Citeable; an Experimental forecast stays on the Forecast tab.

Methodology