# Registered bounded trend-core experiment v1

September 8, 2026, before implementation/performance. Purpose: decide whether a fixed Bitcoin core plus limited trend-dependent borrowing merits a separate prospective test after the volatility candidate weakened. These histories are already explored; no independent market holdout is claimed.

## Fixed strategy

Target exposure = 1.00 + 0.50 times the frozen long/cash breakout state. A negative/absent trend therefore leaves 1x Bitcoin exposure; a positive trend targets 1.5x. This is a target fraction of account equity, not a segregated immutable BTC quantity. Daily rebalancing can buy/sell BTC as equity changes. Borrowing is hypothetical and only the existing paper margin account is used.

Use evaluate_breakout_robustness.all_signals without changes. Central state is breakout_120_60: a close above the preceding 120 observed four-hour highs opens the long state, and the existing 60-observation low exit closes it. These are nominal 20-day/10-day windows, not 120/60 days. Current candle is excluded from the channel. Preserve all existing code details and warmup behavior, including observed-bar rather than exact-calendar lookbacks across gaps.

Nine predefined neighbors are all combinations of entry 90/120/150 observations and exit 45/60/75, exactly the old neighborhood. The central rule is fixed and will not be replaced by a better-looking neighbor. For every candidate, sample the current breakout state only at actual UTC-midnight candle closes, hold the target between those decisions, and use that held target at the first account row. A missing midnight is not backfilled. Before first decision, target 1x.

## Timing and fixed controls

- Inverse central: target 1 + 0.5*(1-central state) at the same daily decisions. This checks whether extra exposure benefits from trend alignment.
- Week-old central: target 1 + 0.5*central state known at or before seven days before the current observation. Use timestamp lookup, not a fixed row offset; choose the latest source observation at/before that cutoff, record actual source age, and default to state zero before history exists. No future state or fabricated price.
- Existing controls, unmodified: BTC bought once and held, and daily fixed exposure 0.75x, 1.25x and 1.50x. Reproduce their prior curves exactly.

## Inputs and account matrix

Use the same reference historical/recent vintages, Binance accepted prices, fourteen windows, native execution grids and full warmup as VOLATILITY_VENUE_PROTOCOL.md. Verify all prior source and trace hashes. No new data, interpolation, rewritten gaps, FX conversion or original-strategy mutation.

Two sources x fourteen windows x fifteen models x two costs x two delays x three APRs = 5,040 accounts: 3,024 candidate cases, 672 timing-control cases, and 1,344 repeated fixed-control verification cases. Costs 6/30 bps per quote unit traded; delays one/three four-hour bars; hypothetical APR 5/10/20%. Primary 6 bps, four-hour delay, 10% APR. Starting capital 10,000 native quote units: USD reference, USDT Binance. The currencies are not converted or interchangeable investor returns.

Reuse margin_book, metrics, quote_metrics and trace_rows unchanged, including daily scheduling, elapsed interest, expired missing fills, adverse low-based liquidation, insolvency retention and zero cash yield. Exposure never exceeds 1.5x. No real financing product or trade is implied.

## Decision criteria fixed before outcomes

This is a screening gate for the central rule, not a guarantee or automatic production promotion. Evaluate the three decision windows since 2019, since 2023 and the separate recent replay. Earlier/annual periods remain reported as risk diagnostics.

A qualifying window must satisfy all of these on BOTH price sources:

1. Central primary return strictly exceeds own-source buy/hold, fixed 1.25x and fixed 1.50x.
2. Central beats all three controls together in at least eight of the twelve cost/delay/APR combinations.
3. At least six of nine neighboring candidates beat all three at primary settings.
4. Central primary return exceeds the inverse timing control and is at least as high as the week-old timing control.

The family passes this screen only if at least TWO of the THREE same decision windows qualify and every central scenario remains solvent. This does not require winning every year or the recent window specifically. Report the central maximum drawdown, downside relative to buy/hold, financing cost and liquidation events regardless of the gate; no numeric personal drawdown tolerance is assumed. A pass still needs prospective evidence and review of severe losses. A fail closes this candidate family as a route to promotion; do not tune its parameters, exposure size or gate after outcomes, or substitute a lucky neighbor. Keep the overall objective honest: a rejected candidate is research evidence, not improved trading performance.

## Validation and reporting

Test exact transformation of existing breakout states; unchanged target cadence; causal timestamp-based lag and missing-input behavior; inverse mapping; future/scale feature and full account-prefix invariance; prior controls and original decisions unchanged. Save all targets/account traces and source feature states/lag ages. Report the gate condition by condition, all candidate/control core returns, annual failures, borrowing/cost sensitivity and worst drawdowns. All scenario counts are correlated descriptions, not probabilities. Run focused/full tests and create digest/provenance/evidence. Update the article honestly. No production/cloud change or new forward registry is authorized by a historical pass alone.