The data budget chooses the instrument class

Three of the M0 gates price data: full OHLC as-printed bars (M0.3), a corporate-actions series including delistings (M0.4), and historical constituents for point-in-time universe construction (M0.5). All three exist for one upstream reason — the intention to measure equities over multi-year history. Splits and dividends are an equity phenomenon; changing universe membership is an equity phenomenon; delisting is an equity phenomenon. On an instrument with none of them, the dependency does not become cheaper, it ceases to exist.

$128.09 of the $150 cap is spent and free credit is exhausted. Rather than weaken the evidence standard to fit $21.91, v1 changes the instrument class and keeps the standard intact. The equity-dependent axes are allocated 0 under ADR-0017 and draw from reserve by recorded decision if the data is later acquired.

This does not overturn ADR-0004 or ADR-0025. Both stand exactly as written. They govern an axis that is not running, and they are the reason it is not running.

Considered options

• Free adjusted equity data, with the bias declared. The obvious cheap path, and already rejected twice. ADR-0025: declaring a bias does not remove it — the number produced is still not the number the strategy would have made. ADR-0004 is worse than a bias objection: the adjustment factor is restated retroactively on every dividend, so an archived run stops reproducing with no code change. That is not a quality compromise, it breaks the foundation the Register rests on.
• A short window and a fixed equity list chosen so no corporate action falls inside it. Superficially rigorous and unverifiable: proving no action occurred requires the actions series that has not been bought. The claim would rest on the absence of evidence that was never collected.
• Raise the cap. Available, and it is the author's decision alone (R9 reasoning). Not taken here because it answers how to pay rather than what v1 should measure, and this decision is required either way — the equity axes cannot run before the data exists, whatever the eventual budget. Nothing in this ADR forecloses it.
• Build the machinery and measure nothing — stop at M3. Cheapest and genuinely safe. Rejected because make null proves the pipeline refuses; nothing would ever prove it can produce a verdict. A falsification apparatus never run against real market data is untested in the direction that matters.

Consequences

Two of four axes go to zero. fundamental was already 0 (A2, effective n=2). cross_sectional joins it: comparison across a universe of names is precisely what point-in-time constituents are for. regime and price_daily remain runnable and carry the whole live budget. The invariant sum(axis budgets) + reserve == alpha.total is unaffected.

The success criterion in DESIGN-v4 §10 changes. It reads "on liquid equities". For v1 it is a regime-conditional or daily-price mechanism on an instrument class with no corporate-action series. The equity form of the criterion is deferred, not abandoned.

M4.2, M4.3 and M4.10 — splits, delistings, point-in-time universe — are built as ports with fixture-backed tests and no live source. The semantics are proven; the data is absent. This keeps the equity axis a configuration and licence question later, not a rewrite.

M0.3-M0.5 still run. They cost nothing but time, and they price the option to reopen. M0.6 records the go/no-go with real numbers behind it.

M0.2 becomes the load-bearing gate. On instruments without corporate actions the spread is substantially the entire round-trip cost, and it is what kills short-horizon edges. The detectable floor declared under D5 must clear a cost bound that bites harder here than it would on equities. ADR-0026 also gains weight if the replacement class is FX-denominated.

The replacement instrument set is not decided by this ADR — only that it must carry no purchased-data dependency. That choice follows M0.2 and M0.7.

Amendment note, 2026-09-06 — the price of the deferral, recorded.

This decision stands: the money is not there, and ADR-0017 supplies the mechanism. Two things learned since should be written down, because a deferral whose cost is unrecorded is indistinguishable from a decision nobody has to revisit.

The decision is stronger than it was argued. All three deferred data lines are properties of equities. FX pairs do not delist, index CFDs have no constituent history to reconstruct, commodities pay no dividends. Moving the instrument class did not only avoid three invoices — it dissolved three bias classes (survivorship, delisting returns, restatement look-ahead) rather than declaring them. That is a better justification than cost, and it is the one that would survive a larger budget.

The deferral has a price, and it was not named. The capacity constraint is the only structural advantage a retail participant holds: what is too small for a fund to enter without impact is available, and published anomalies persist in exactly that corner because they cannot be arbitraged at scale. Capturing it requires holding many small positions at once, because breadth in the cross-section is the only substitute for the trade frequency a long holding period gives up (IR ~ IC * sqrt(breadth)). **That is the cross_sectional axis this decision allocated 0** — and it needs precisely the survivorship-free constituent history that was deferred.

So the axis set to 0 is the one where the retail edge structurally lives. That does not overturn anything; the money is still not there. It makes cross_sectional the first axis reconsidered if M0.14 leaves room, rather than one deferred item among three, and it is why M0.7 now records what the chosen instrument class gives up rather than only what it offers.

Amendment note, 2026-09-10 — the replacement instrument set, chosen and confirmed. This ADR left the set to M0.2 and M0.7. M0.7 chose currently-listed, liquid US-listed ETFs and large-cap shares on daily bars, and the author confirmed it the same day (docs/M0-ANSWERS.md §M0.7; TASKS status log).

Two things the choice settles about this ADR's own wording. First, the consequence "an instrument class with no corporate-action series" is superseded by the amendment's narrower test, no purchased-data dependency: the class has dividends and splits, and they cost nothing (Tiingo's bar row carries divCash and splitFactor), while the three deferred data lines stay deferred because a currently-listed deployment instrument needs no survivorship-free universe, no delisting returns and no constituent history. Second, the four-way position is on record: the class gives up the capacity edge entirely, takes the best cost_in_R available, moves the sample risk from count to concurrency rho, and keeps one declared bias — the instrument's own survival — that binds selection hypotheses only. ADR-0025's objection to declared survivorship is therefore not engaged by regime or price_daily on this class, and is engaged the moment a UniverseRule selects from a set.

What did not change. cross_sectional stays at 0, now on sample size as well as money (M0.15). The FX term on this class is 0 or 0.30 % per round trip by the account currency, which is the author's at M0.13 and is recorded, not recommended.