# trending-1005 pre-registration

Lane m14, 5 Oct 2026. Written and committed **before any new cell was run on market data**. The only thing run before this commit was a mechanics smoke test of the 12 new rule files on synthetic random-walk bars (no market data), to check that each loads, passes `validateMeta`, and changes position.

Sources and attention evidence for the five techniques: `sources.md`. The exact cells, with the sha256 of every strategy file: `cells.json`. Runner: `scripts/research/trending-1005.mjs`.

## The five techniques and their cells

Each technique has one **best-faith** cell (decides the verdict) and 2 or 3 common variants (reported, counted in the multiple-testing family). "Earlier run" means an existing ledger row, reused as is (not re-run).

Hourly crypto (`CRYPTO_1H`: BTC, BNB, XRP, ADA, DOGE, LTC, LINK, TRX USDT) is the only intraday data the harness has. The futures and stock versions of ORB and ICT were already tested in earlier intraday studies (`docs/research/orb-1002`, `orb-1003`, `research/liquidity-1001`); the articles quote those results by file and do not re-run them.

### 1. Opening range breakout (`orb`)

| cell | role | rule |
|---|---|---|
| `tr5-orb-utc-1h-2r-cutoff` | best | Range = the 00:00 UTC hourly bar. Enter on the first close outside it at 01:00 or 02:00 UTC only (time cutoff), long above / short below. Exit on a close at entry +/- 2R (R = range width) or a close back beyond the other edge; flat from 23:00 UTC. One trade a day. Hourly crypto. |
| `tr5-orb-utc-1h-eod` | variant | Same range, entries until 22:00 UTC, stop at the other edge (close), no target, flat from 23:00 UTC. |
| `s29-h-utc-orb` | variant, earlier run | 4-hour UTC opening range, first close outside, no stop, out at day end. |

### 2. ICT session liquidity sweep (`ict`)

| cell | role | rule |
|---|---|---|
| `tr5-ict-asia-sweep-be` | best | Asia range = 00:00 to 06:00 UTC hourly bars. In London (07:00 to 11:00 UTC) a bar that trades below the Asia low and closes back above it goes long at that close; stop = that bar's low, R = close minus stop. Mirror for shorts at the Asia high. A bar sweeping both sides is skipped. Stop moves to entry after a close at +1R; exit on a close at +2R or at the stop; flat from 20:00 UTC. One trade a day. Hourly crypto. |
| `tr5-ict-asia-sweep-2r` | variant | Same, no break-even step. |
| `tr5-ict-asia-sweep-hold` | variant | Same entry, no stop or target, flat from 20:00 UTC. |
| `s29-etf-turtle-soup` | variant, earlier run | Daily cousin on 25 ETFs: a failed 20-day breakdown recovered the same day, held 5 days. |

### 3. VWAP plus the 9 EMA (`vwap`)

| cell | role | rule |
|---|---|---|
| `tr5-vwap-ema9-trend` | best | VWAP anchored at 00:00 UTC each day (typical price x volume, Binance volume). Long when the close is above both VWAP and the 9-hour EMA, short when below both, flat when they disagree and in the 00:00 bar. Hourly crypto. |
| `tr5-vwap-reclaim-ema9-exit` | variant | Long only: enter when the close crosses from below to above the session VWAP; exit on a close below the 9 EMA; flat from 23:00 UTC. |
| `social-ta-vwap-bounce` | variant, earlier run | Bounce off a rolling 20-hour VWAP band. |

### 4. 9/21 EMA crossover (`ema`)

| cell | role | rule |
|---|---|---|
| `tr5-ema921-1h-both` | best | Hourly crypto, always in: long while EMA 9 > EMA 21, short while below. |
| `h3-hourly-ema-9-21-cross` | variant, earlier run | Hourly crypto, long only while EMA 9 > EMA 21 and the close is above the 200 EMA. |
| `tr5-ema921-daily-stocks` | variant | Daily, 40 US large caps (`US_STOCKS_DAILY`), long while EMA 9 > EMA 21. |

### 5. Bitcoin's 200-day line, golden cross, Uptober (`btc200`)

All on daily crypto (`CRYPTO_DAILY`, 18 USDT pairs). Bitcoin alone cannot pass the registry gate (it needs at least 2 assets), and ETH is in the holdout, so the rule is run on the 18-coin basket with BTC in it.

| cell | role | rule |
|---|---|---|
| `tr5-crypto-200dma-30d` | best | Long after 30 consecutive daily closes above the 200-day simple average; out on the first close below it. |
| `tr5-crypto-golden-cross` | variant | Long while the 50-day simple average is above the 200-day. |
| `tr5-crypto-uptober` | variant | Long for every day of October (UTC), cash the rest of the year. |

## Success criterion, per technique

A technique **held up** only if its best-faith cell passes the registry's PROMISING gate (below). A variant passing alone is reported as "one variant passed; the best-faith version did not", and flagged as likely luck given the number of tests. If no cell passes, the article says "did not pass these tests", never "does not work".

## Common to every cell (fixed before any run)

- **Engine.** `research/hunt2/harness.mjs` + `data.mjs` + `ledger.mjs`, unchanged (the harness behind the published registry: 624 strategies, 707 FDR tests at the base of this lane, 0 passed, lowest q 0.141). Runner: `scripts/research/trending-1005.mjs run`. It refuses to run a strategy file whose sha256 differs from the one recorded in `cells.json`, and refuses to re-run an id already in the ledger.
- **Data.** Cached discovery data only: bars that close before 2025-03-15 (`DISCOVERY_END`), discovery-bucket assets only (the loader throws on holdout assets: ETHUSDT, SOLUSDT, SPY, MSFT, IWM, EFA and the rest of `src/lib/quant/holdout.ts`). Network is blocked in the runner (`fetch` throws), so a missing cache file fails the run instead of downloading. No Databento call of any kind. No holdout bucket file is opened.
- **Costs.** The harness cost model: crypto 0.10% a side plus the measured per-symbol slippage, US stocks and ETFs 0.05% a side, charged on every change in position. Shorts pay no borrow or funding (stated as a limit). Gross numbers are the same run at zero cost (`evaluate(file, { costPerSide: 0 })`, never ledgered).
- **Benchmark.** Equal-weight long of the same assets, rebalanced daily, no costs (the published comparison).
- **Random control.** The harness's 200 seeds: the rule's own trades (same lengths, same directions) moved to random non-overlapping dates, same costs. The runner exposes those 200 series (a runtime copy of the harness with the same seeds, so its percentile equals the ledgered one) to draw the distribution chart.
- **Multiple testing.** Every new cell is appended to `research/hunt2/ledger.jsonl` (one test each, no re-runs) and so joins the published Benjamini-Hochberg family. Reported per cell: p, q inside this lane's family (all new and reused cells here), and q inside the whole published family after this batch.
- **Pass bar (the registry's PROMISING gate, unchanged).** q < 0.10 on the excess-vs-benchmark p over the whole published family, beats at least 95% of its random controls, positive excess in both halves of the window, at least 30 trades, net CAGR above 0, at least 2 assets. A technique "held up" only if its best-faith cell passes. Anything else is reported as "did not pass these tests", never as "does not work".
- **Reuse.** Where an earlier ledgered run already is the best-faith version or a common variant, it is reused as is (its existing ledger row, no re-run), named in `cells.json` with `reuse: true`, and counted in this lane's family.
- **Wording.** No result here is an edge claim. A risk improvement (lower drawdown, lower return) is reported as that, not as a win.

## Also reported (descriptive, no pass/fail)

Gross (zero-cost) excess, max drawdown vs the benchmark, Sharpe, exposure, trades, split-half excess, and the 200-seed random-control distribution (charted). For ORB and ICT, the earlier futures and stock studies are quoted from their own result files.

## Known limits, stated up front

- Hourly bars, not 5-minute or 15-minute bars: an hourly range is a coarser version of the 15-minute ORB, and exits are on hourly closes (no intrabar stop or target fills). Crypto trades around the clock, so the "session" is the UTC day, not the 9:30 ET cash open.
- Shorts pay no borrow or funding.
- The discovery window ends 2025-03-15; nothing after it is used.
- The universes are today's survivors (survivorship bias favours long rules).
