TIDEMARK

A separate research project, reported here. Vintage 2026-08-17.

Educational research tool — not investment advice, and not a recommendation to buy or sell any security. Backtested results are hypothetical, come from one 18-year dataset the model was also tuned on, and are not a promise about the future. The forward track is a model portfolio and a sandbox paper account — no money is invested in either, so no figure here is a return anyone received — and it is short. You can lose money. Do your own research.

The rotation question was asked. It came back unanswerable.

This project set out to test whether tilting new contributions toward markets that are cheap versus their own history beats contributing on a fixed schedule to a fixed allocation. That test cannot be run on this data at any bar worth trusting, and the ruling is binding: see POWER_GATE.md.

The reason is sample size, not sentiment. Markets get cheap together — six separate cheap episodes between 2008 and 2016 are one macroeconomic event seen six times — so three markets pooled hold 3 independent episodes, which is fewer than the charter requires of a single market on its own.

This is the project's headline finding, not a caveat attached to one. “Not answerable on this data” is a complete answer and it was the expected one — written into the charter in advance, before any of it was measured.

NO PHASE-2 QUESTION MAY BE ASKED ON THIS DATA. Not one of the three canonical markets reaches the power the project's own pre-registered arithmetic requires, and neither does the pooled test. The shortfall is not marginal: the best-placed market has 82 of the 155 independent years required, and the worst has 25.

Read the other way round: the edge would have to run at an information ratio of 0.41 to 0.74, depending on the market, before this data could detect it — against the charter's own plausible range of 0.2 to 0.4 for a costed cross-asset value overlay. The problem is not that the answer is uncertain. It is that the edge would have to be larger than such an overlay plausibly is.

3 pooled independent episodes
3.33× co-movement discount on the naive count of 10
24.7 yr all three markets overlap
155 independent years a test would need
82 the best market actually has

Two projects, two denominators

These two projects have different denominators and different bars. Valquo's trial register and TIDEMARK's are separate budgets, so a number from one may not be compared with, averaged against, or read as corroborating a number from the other.

There is no combined verdict, and there will not be one. Valquo measures a cross-sectional stock ranking; TIDEMARK measured whether cross-asset rotation is answerable at all. Neither result is evidence about the other.

Project Trials (N) Critical value √(2·ln N) Basis
TIDEMARK 66 2.89 TIDEMARK's own budget, at its own vintage. Vintage 2026-08-17.
Valquo — equity 307 3.3843 Valquo's own budget, derived at render time.
Valquo — options 324 3.4002 Valquo's own budget, derived at render time.

How to read a reading

A market at the 90th percentile with twelve independent episodes gets a band you can act on. One at the 90th percentile with two gets a band so wide that the honest caption is “we cannot tell you what this implies.” Both are legitimate outputs, so the tier is stated before the number, in words, on every card.

Where it sits: the percentile plus or minus 1.96 standard errors, computed at the effective sample size (raw observations divided by the measured design effect), never at raw n — a raw-n interval would be three to ten times too narrow.

What it implies: 1.96 divided by the square root of the episode count — the tightest interval that count could support, in standard deviations. It needs no return data at all. Both are approximations and are labelled as such.

ACTIONABLE A band you can act on
At least 12 independent episodes AND enough independent history to support a test. The reading means something you could lean on.
0 markets
INDICATIVE Suggestive, not decisive
Clears the charter's 4-episode kill condition and the anchor is admissible, but the record is too short to support a test. Read the direction, not the magnitude.
0 markets
NOT INTERPRETABLE We cannot tell you what this implies
Fewer than 4 independent episodes, or the anchor is inadmissible, or its episode count is not robust to how the count is taken. The percentile is a true statement about the past and supports no inference about the future.
0 markets
REFUSED No number is shown, deliberately
There is not enough of its own history for a percentile to mean anything. The refusal is the output, not a gap waiting to be filled.
0 markets

The two tiers at the top are empty, and that is the finding rather than an oversight. They are shown so the scale is visible: this dashboard is capable of saying “you can act on this”, and on this data it never does.

No readings are shown. The vintage is 96 days old, past the pre-committed limit of 90. Vintage 2026-08-17, limit 90 days.
No strategy claim is made on this page, and none may be made on this data. Every percentile is a true statement about where a market sits in its own measured history. None of them is a statement about what happens next.

Derived statistics only — percentile, episode count, effective n and band half-width. No source series is reproduced on this page. Sources behind the figures, in the TIDEMARK project: POWER_GATE.md, PREREG_ANCHOR_SELECTION.md, PREREG_P1_GATE.md, scripts/p1_8_dashboard.py.

← Back