What our baskets and models show today, without signing in. Every tile says how established its figures are:
Live is recorded as it happens, Hypothetical is a backtest, Experimental
is a method still being proven, and Preview is a model still in validation.
NDI Baskets
Hypothetical
As of 2026-10-22
Each basket selects its members from NDinomics' universe of tracked assets by its own criteria, ranks them, and weights them by its weighting scheme, then rebalances on its schedule. The hypothetical series is a reconstituting backtest: at each monthly boundary the price-based criteria are re-evaluated from prices through the previous trading day, and the cost of the resulting turnover is charged. The live series is the daily value of the basket as recorded by the daily tick, rebased to 100 at the first tick; days without a tick are shown as gaps and are never filled in.
The Calf — First NDI Basket
The first basket born from the Arena pipeline. Large-cap momentum, EF Max Sharpe optimized. Named after the Calf — the autonomous agent that builds what the Arena debates.
Weighting
ef max sharpe
Rebalances
monthly
Members
12
Expense ratio
0
Est. annual rebalance cost per $10,000
$9.11
Selects: market cap >= 10
Hypothetical Backtest, not a record
2024-08-20 to 2026-05-22: 100 to 231.4 (starts at 100)
Live Daily record
2026-05-25 to 2026-10-22: rebased to 100, now 81.96
No tick between 2026-06-18 and 2026-07-06; the line is broken there, not filled in.
No tick between 2026-07-06 and 2026-07-27; the line is broken there, not filled in.
No tick between 2026-08-06 and 2026-08-18; the line is broken there, not filled in.
No tick between 2026-08-19 and 2026-08-31; the line is broken there, not filled in.
Data notes (1)
no correlation matrix is published: a uniform correlation is assumed between all members
Limitations (6)
Hypothetical reconstituting backtest, not a track record. Criteria are re-evaluated monthly from price data through the previous trading day. Static criteria (market cap, class, grade, yield, ...) use today's values, so history is flattered by what is known now. Survivorship bias: only tickers that exist today are considered. Trading costs are charged on reconstitution turnover only (bid-ask spread and SEC fee; FINRA TAF omitted); keeping weights at target between reconstitutions is not costed. Prices are split/dividend-adjusted closes.
Criteria that cannot be computed from prices (market capitalisation, class, grade, yield) use today's values for the whole history, so the hypothetical series is flattered by what is known now.
Survivorship: only assets that exist today are considered.
No taxes are modelled. No management fee is charged: the expense ratio shown is 0 because no product is sold.
Hypothetical results are not actual results and do not represent trading in any account.
Past performance, hypothetical or actual, does not indicate future results.
ESI market state
Experimental
As of 2026-10-23
ESI reads the news and filings NDinomics collects, sorts each item onto one of eight event ribbons by keyword rules (some items also by an AI reading), and scores how severe it is. This page shows the last 30 days of that record, three flags for fiscal, monetary and political activity, and which past market episode the last 90 days of events most resembles. For that resemblance it also gives scenarios: what is happening now, how the past episode unfolded (context only), and two or more mutually exclusive turns, each with a probability from ESI's model, a date by which it is judged and a rule for judging it. Each scenario is stored when issued; when its date passes it is judged by the rule, not by hand, and scored. The record below includes the misses.
Market state
Axiom regime signals, 2026-10-22
10y minus 2y Treasury yield (percentage points)
0.33
Yield curve inverted
No
Credit spread (basis points)
333
VIX
19.03
VIX trend
rising
Gold / copper ratio
631.72
Gold / copper regime
neutral
Oil, 90 days
normal
Copper, 90 days
normal
S&P 500 vs its 200-day average
above
S&P 500 vs its 200-day average (%)
5.13
Event ribbons, last 30 days
Events
Highest severity (1 to 5, 0 = none)
Credit
32
4
Institutional
172
3
Geopolitical
87
4
Policy
452
4
Technology
67
3
Pandemic
4
2
Commodity
37
3
Regulatory
102
3
Policy flags
Fiscal: active (22)Monetary: active (401)Political: active (201)
Active = an event of severity 3 or more in the last 30 days.
Closest historical episode (an explanation, not a forecast)
2023-2024 AI Bull Market – stage 5 of 5; 3 stages matched, keyword match score 81.3 out of 100 (not a probability).
Scenario outlooks (possibilities that are scored afterwards, not forecasts)
As of 2026-10-14, events captured in the last 90 days match 3 of the 5 stages of the library pattern 2013-2015 Bull Market (Post-QE Calm) (furthest stage matched: 3). The keyword similarity is 68.0 on a 0 to 100 scale; that is a similarity measure, not a probability. The matched events are on the commodity, policy, regulatory ribbon(s), have severities 2, 3, 3, and are dated 2026-10-12 to 2026-10-14.
3 of 5 stages of the pattern matched; ribbons: Commodity, Policy, Regulatory; 3 events cited
Before (context only; no number below comes from it)
The library holds one earlier episode with this chain of events, 2013-2015 Bull Market (Post-QE Calm). The broad US equity index fell about 0% peak to trough over roughly 36 months. This is one past example for context. It sets none of the probabilities below, and a similar start has not always led to a similar path.
Next
In our model's estimate, over the window to 2027-01-27 there are two mutually exclusive turns. One possible turn is that normal capital market activity and innovation cycle is captured (probability 12%). The other is that nothing of the kind is captured in that window (probability 88%). These are estimates from an unscored model; they are not statements about what happens.
Window ends 2027-01-27.
11.9% One possible turn: a further event of the pattern's next stage is captured by 2027-01-27.Judged by: At least one event captured after 2026-10-14 and by 2027-01-27 (UTC), on the technology ribbon, severity 3 or more, covering at least 25% of the keywords of stage 4 of the 2013-2015 Bull Market (Post-QE Calm) pattern. Events tagged regime_overlap do not count.Confidence label: track record not yet sufficient
88.1% Another possible turn: no event meeting that test is captured by 2027-01-27.Judged by: The advance test above is not met by 2027-01-27.Confidence label: track record not yet sufficient
Basis of the numbers: stage prior (a fixed table set by ESI's developers, not fitted to data) x keyword match score / 100 x structural-conditions factor. They are not fitted to data.
2000 Dot-com Crash issued 2026-10-22
Now
As of 2026-10-22, events captured in the last 90 days match 6 of the 8 stages of the library pattern 2000 Dot-com Crash (stages matched: 1, 2, 3, 4, 5, 6). The keyword similarity is 80.0 on a 0 to 100 scale; that is a similarity measure, not a probability. The matched events are on the credit, institutional, policy ribbon(s), have severities 2, 2, 2, 2, 3, 4, and are dated 2026-10-12 to 2026-10-21.
6 of 8 stages of the pattern matched; ribbons: Credit, Institutional, Policy; 6 events cited
Before (context only; no number below comes from it)
The library holds one earlier episode with this chain of events, 2000 Dot-com Crash. In that episode the stage after this one (stage 7) appears in the record around 2002-01-01. The library records a peak-to-trough fall of about 49% in the broad US equity index over roughly 31 months. This is one past example for context. It sets none of the probabilities below, and a similar start has not always led to a similar path.
Next
In our model's estimate, over the window to 2027-01-10 there are two mutually exclusive turns. One possible turn is that a regulatory event resembling stage 7 of the pattern is captured (probability 56%). The other is that nothing of the kind is captured in that window (probability 44%). These are estimates from an unscored model; they are not statements about what happens.
Window ends 2027-01-10.
56% One possible turn: a further event of the pattern's next stage is captured by 2027-01-10.Judged by: At least one event captured after 2026-10-22 and by 2027-01-10 (UTC), on the regulatory ribbon, severity 3 or more, covering at least 25% of the keywords of stage 7 of the 2000 Dot-com Crash pattern. Events tagged regime_overlap do not count.Confidence label: track record not yet sufficient
44% Another possible turn: no event meeting that test is captured by 2027-01-10.Judged by: The advance test above is not met by 2027-01-10.Confidence label: track record not yet sufficient
Basis of the numbers: stage prior for the number of stages matched (a fixed table set by ESI's developers, not fitted to data) x keyword match score / 100 x structural-conditions factor. They are not fitted to data.
Record of judged outlooks
Issued
4
Open
2
Judged
0
Not judged
0
Voided
2
No judged outlooks yet. Until some have been judged there is no score, and no probability above carries a confidence label.
Voided (issued in error, not scored)
2026-10-14, 2023-2024 AI Bull Market: issued in error: next stage computed from the count of matched stages, not the furthest matched; matched stages not consecutive
2026-10-14, 2008 Global Financial Crisis: issued in error: next stage computed from the count of matched stages, not the furthest matched; matched stages not consecutive
Educational only, not advice. Scenarios are probabilistic estimates, not predictions; they can be wrong.
Limitations (8)
Events are classified by keyword rules and, for some items, an AI reading: items are mislabelled or missed.
The historical episodes before 2020 are a curated library written with hindsight, not a record made at the time.
A sequence match is a similarity reading, not a forecast. The scenario probabilities are a separate estimate from a model whose stage priors were set by hand, not fitted to data; they are shown together with their track record.
The track record is short. Until enough scenarios have been judged, no probability here carries a confidence label, and a small record says little about how good the estimates are.
A scenario is judged only on ESI's own captured events, with a keyword rule, so a turn can be judged 'happened' or 'did not happen' because of how events were captured or worded. A scenario whose window had no event intake is reported as unresolved, not scored.
A flag is active when an event of severity 3 or more that matches it was captured in the last 30 days; the match is by keyword.
The market state comes from Axiom's daily regime signals and may be a few days old; its date is shown.
Counts reflect what NDinomics' sources published and what its collectors captured, not every event in the world.
Axiom
Preview
Figures are on the Axiom page
Axiom estimates, for every asset NDinomics tracks, how it is likely to behave, and how the market as a whole is positioned. The live page shows these per asset; no sign-in is needed to look.