TideAlphaFive themes. One rule set. Every decision on the record.
The Oracle is an AI agent that holds five investment themes at 20% each. It buys a theme that reads Ignition, then Rising. It reviews the book after the last close of every week and leaves a theme on the first Peaking, Falling or Bust reading. It ranks the queue by money flow and takes at most two sleeves from one theme group. A second book holds the same sleeves at the exposure two dials set.
The technical page needs an account.
The gold standard
The gold standard is report_gold_lifecycle.html: stage labels set on the complete cycle of every theme, 150 themes and 56,850 theme-months from 1995 to 2026. It carries a ladder of long-only theme books, where selection uses the return of the month it holds. The top rung holds the best theme of each month and makes 3,531.0% a year; the five-theme rung makes 1,159.3%. The rung the Oracle reads holds every theme marked Ignition or Rising at equal weight and makes 67.5% a year at Sharpe 2.49 with no losing year in 32.
| Stage | Return that month | Hit rate | Theme-months | In the book |
|---|---|---|---|---|
| Ignition | +10.3% | 82% | 3,801 | Yes, first |
| Rising | +3.5% | 64% | 14,166 | Yes, second |
| Peaking | +0.4% | 45% | 3,034 | No |
| Falling | -2.0% | 39% | 5,122 | No |
| Bust | -4.9% | 34% | 4,501 | No |
| Dormant | +1.4% | 57% | 26,226 | No |
- Equal-weight theme-index returns of the month a stage labels. Adding Peaking to the book gives 53.3% at Sharpe 2.13. Peaking alone gives 1.6% at a -66.2% worst loss, which is why the Oracle leaves on the first Peaking reading.
- The live rule set reads the same six stages from the price at the close that decides. It makes 34.8% at Sharpe 1.18. That is 32.7 CAGR points and 1.31 Sharpe below the stage book, and 1,124.5 points below the five-theme rung of the ladder. Both gaps are the distance the live reading has to close.
Both need an account.
What makes it different
Theme lifecycle research, not a screen
150 themes were labelled by hand across 32 years to measure what each stage pays. On the gold standard, Ignition returns 10.3% in the month it labels and hits 82% of the time; Peaking returns 0.4% and Bust -4.9%. The Oracle sells busts. It does not call tops and it does not forecast returns.
A point-in-time rubric
Six ordered tests on the theme index only: the 200-day average, the distance from the 252-day high, the 126-day return, and the ratio of 21-day to 126-day volatility. Every input is known at the close that decides. Nothing in the reading uses a later price.
An AI that shows its work
The agent writes a thesis for every held theme, names the trigger for every exit, and appends one ledger row per book per week. A daily note judges each thesis against the day. The next week scores the last week from closes.
The live rule set
One basis for every row: theme indices of members holdable after 126 sessions and $1M of daily dollar volume, decisions at the close, fills at the next close, 10 bps a side, idle cash at the 3-month bill. Window 2003-12 to 2026-07.
| Book | CAGR | Sharpe | Worst loss | Universe |
|---|---|---|---|---|
| SPY buy and hold | 11.0% | 0.79 | -51% | The market |
| Equal-weight theme basket | 17.3% | 0.88 | -51% | 150 themes |
| Every favorable theme, equal weight | 22.8% | 1.21 | -29% | 94 live themes |
| The Oracle | 34.8% | 1.18 | -30% | 94 live themes |
| The gold standard stage book | 67.5% | 2.49 | -17% | 150 themes |
| The gold standard, top 5 themes each month | 1,159.3% | 4.09 | -3% | 150 themes |
- CAGR and worst loss on month-end values, so the worst loss is a drawdown and not one month. Sharpe on monthly returns at 0% risk-free. On the 150-theme basis the Oracle makes 34.3% at Sharpe 1.12 and a worst loss of -40%.
- A theme group of ORACLE.md may hold two sleeves at most. The cap refuses an entry at every review and it re-ranks the incumbents under the same cap at a queue review. Without it the book held three or more sleeves of one group in 37.5% of the 1157 weekly decisions, and up to five of five. The cap removes every one of those weeks and moves the book from 33.5% / 1.10 / -42% to 34.8% / 1.18 / -30%.
- The filter gives the Sharpe, the rank gives the return. Holding every favorable theme at equal weight already reaches Sharpe 1.21 at two thirds of the Oracle return.
- The two gold standard rows come from the complete cycle of every theme, 1995-2026 at 10 bps a side. The stage book row reads the same six stages the live rule set reads; the five-theme row is the ladder rung with the same slot count as the Oracle. Both are the standard the live reading is built to reach.
- Every row is in sample. The rules were chosen with the full history in view and the theme list is the 2026 list, so themes that died without a surviving member are missing. The live record from 2026-08-31 is the only holdout.
- Recent window 2025-01 to 2026-07: the Oracle made 46.0% at Sharpe 1.12 against 27.4% at 1.44 for the equal-weight basket. Without the group cap the same rules made 20.1% at 0.65 over the same window, so the cap carries most of its gain in the recent years.
What you get
The Oracle report
One Saturday report carries the book, performance, theses, week and due month-end review, market context, changes, candidates, exposure and watch levels.
The live book
The current sleeves, weights and members, the day's move, and a chart for every theme. The terminal carries the stage of any theme you look at.
One decision
The long-only and dialed views are derived from the same validated engine proposal and the same AI review; they are not separate Oracle jobs.
The audit trail
Every run records its exact prompt, tool calls, decision, note, token use and output files so the published result can be traced back to its inputs.
The risk
A theme book loses 30% to 50% in a bad year. A worst loss of -30% to -45% on the long book is inside expectation, not a failure of the rules. 13 of the 35 deep falls in the research were theme falls with the market nearly flat, so the deep falls are a sizing problem. The two-sleeve group cap is the first answer to that risk: it cut the worst loss from -42% to -30% and added return. The dialed book is the second: it runs the same sleeves smaller when volatility expands or the sleeves converge, and holds the difference in cash at the bill rate.
Both books are long only. The Oracle holds no leverage and no derivative, and it holds only themes that read Ignition or Rising.
Judge the Oracle after 24 months on Sharpe against the equal-weight basket and on worst loss against the expectation above. One year at 40% to 55% volatility cannot separate skill from noise. This is research, not investment advice.