Meridian
Operator manual

How to use the desk

Meridian does two jobs: collect published claims, then try to break them on one futures series at a time (NQ, ES, COMEX gold, or CME Bitcoin). Harvesting is not research. Research is not validation. Only a Yahoo confirmation on that market’s tape may say “edge,” and the library stays empty until that second pass holds.

  1. 01

    Read first. Do not grind.

    The seeded catalog is already the useful set: Jegadeesh–Titman, time-series momentum, first half-hour, overnight tug-of-war, Avellaneda–Lee, and the rest. Open one paper. Read the transfer-risk box. If the paper is monthly equities, a 5-minute NQ fade is a new claim — not a replication.

  2. 02

    Harvest hunts NQ, ES, gold, and Bitcoin papers. It still does not validate.

    Sign in, then Papers → Harvest journals · X · Substack. Every click is a new wave: newest first, rotated queries and pages so you can keep discovering instead of looping the same six citations. Waves mix Nasdaq, S&P, COMEX gold, and CME Bitcoin on OpenAlex, arXiv, Semantic Scholar, and Crossref, plus a scanner of curated quant X timelines and Substack RSS.

    • OpenAlex journals / SSRN / NBER / RePEc — NQ/MNQ, ES/MES, GC/MGC, BTC/MBT.
    • arXiv q-fin.TR / ST / PM / CP, submitted-date descending.
    • Semantic Scholar and Crossref (exotic / practitioner venues).
    • Concretum authors (Zarattini / Barbon / Aziz) as a dedicated pass.
    • X timelines of a fixed quant roster (Robot Wealth, Moontower, Hoffstein, …) — rule-shaped posts only.
    • Substack / RSS (Quantitativo, Concretum, Quantpedia, Alpha Architect, …).

    A work is kept only if it is NQ, ES, COMEX gold, or CME Bitcoin related (or a Concretum day-trading note). Social posts also need a strategy rule, not a meme. Generic equity or spot-crypto papers are dropped. Duplicates update cites and abstracts. Harvest does not walk-forward, does not emit Pine, and does not validate anything.

  3. 03

    Test one thesis at a time.

    Best path:

    1. Open the paper whose claim you actually understand.
    2. Test this thesis — pick NQ, ES, gold, or Bitcoin, literature mode, that family only, parameters biased toward the paper, 12 candidates, walk-forwarded on 5m / 15m / 30m / 1h / 4h / 6h / 1D / 2D.
    3. Or Research → Literature, pick families, run a broader cycle off the top-cited papers in those families.

    Systematic / evolutionary modes ignore papers. Use those when you want a blind search, not a citation.

  4. 04

    Test on more than one chart.

    Default timeframes are 5m, 15m, 30m, 1h, 4h, 6h, daily, and 2D. Higher charts resample the 5-minute series (clock-aligned ET; 1D/2D use calendar day buckets). Each timeframe is its own walk-forward — a 5m survivor is not a 4h edge. Pine is locked to the chart it passed on. Yahoo confirmation: 4h/6h from 1h (~2y), 1D/2D from daily (~10y). Leave them on unless you are retuning one candidate.

  5. 05

    Families are hypotheses, including Concretum.

    The desk searches 38 families, all of them session- or inventory-shaped: ORB / compressed ORB / day-type, VWAP fade vs rail, EMA and HTF-aligned pullback, Donchian, squeeze and NR7, gap fade vs gap-and-go, IB extension, failed ORB, ETH→RTH handoff, overnight hold, Asia range, lunch coil, power hour, PDH break and sweep, Turtle Soup, RVOL, climax, Concretum fast-alpha, plus exotic: open-drive stall, wide-bar absorption, overnight high/low, failed IB, prior-close location, burst fade, balance retest, unfilled-gap magnet. Oscillators, candle patterns, ICT killzones, DeMark counts, floor pivots, and Heikin-Ashi were removed — they do not survive honest WFO on index futures. Core is the default set (now includes absorption, ONH/ONL, failed IB). Unusual (stalls, sweeps, magnets, afternoon revert) is opt-in.

    Concretum papers (Zarattini, Aziz, Barbon) are practitioner notes on SPY/QQQ/US stocks. They are catalogued as hypotheses with a transfer-risk box. They are not NQ edges. Harvest will also pick up Zarattini works from OpenAlex when they exist there.

  6. 06

    Most cycles should fail. Do not run until one ‘works.’

    A paper is not validated when you have tested it. A candidate is only a Real tape badge when every gate holds on Yahoo NQ=F / ES=F / GC=F / BTC=F out-of-sample folds — after it already passed discovery on the synthetic series. If it fails, that is the result. Retune once if the near-miss is close. Do not farm random seeds hoping the library fills up.

    • OOS trades ≥ 18 (scaled down on the short 5m/15m/30m Yahoo window)
    • OOS profit factor ≥ 1.12
    • OOS expectancy positive
    • OOS max drawdown ≤ 28 %
    • IS→OOS degradation ≤ 60 %
    • Fold pass rate ≥ 50 % (skipped if the tape is too short for two folds)
    • Trade rate between 0.12 and 6 / day
  7. 07

    What the badges mean

    • Rejected — failed discovery. Keep the Pine only to see the rules. Do not book it.
    • Near miss — some OOS life, at least one gate dead. Inspect, maybe retune once, never size.
    • Synth only — passed the synthetic sieve, failed Yahoo on that market’s tape. This is the bucket that was losing on your TradingView chart. Not an edge.
    • Real tape — cleared Yahoo NQ=F, ES=F, GC=F, or BTC=F (CME / COMEX continuous, same family as NQ1! / ES1! / GC1! / BTC1!). Promoted to the library with Pine v6. Still not TradingView’s engine. Re-run on the matching 1! symbol before any capital.
  8. 08

    Meridian confirms. TradingView is still the live exam.

    We cannot log into TradingView. There is no public Strategy Tester API. After discovery, Meridian replays the same OHLC fill model on Yahoo NQ=F, ES=F, GC=F, or BTC=F — the public tape closest to that cycle’s 1! continuous. That is the automated stand-in so you do not have to paste every copy by hand. Stay on TradingView for the live script. You do not need Sierra Chart for this desk.

    Nasdaq, S&P, gold, and Bitcoin are separate desks that share families. An NQ walk-forward does not transfer to ES or BTC. Gold uses COMEX 08:20–13:30 ET, not the NYSE open — power hour and lunch coil are skipped there. Bitcoin is CME futures (BTC1! / MBT1!), not a Binance perp; US 09:30 is liquidity overlap, not an auction. Mini vs micro is sizing and commission, not a different tape.

    1. Copy Pine only from a Real tape badge. Delete any older Meridian copy on the chart first — older copies used absolute stop/limit from the signal close and will not match.
    2. NQ1! / MNQ1!, ES1! / MES1!, GC1! / MGC1!, or BTC1! / MBT1! — the exact timeframe on the script header (5 / 15 / 30 / 60 / 240 / 360 / D / 2D).
    3. Leave Properties alone. Order size stays contracts. Leverage stays default — the script already sets 10% margin so 1 mini fits in $100k. Bar Magnifier stays off.
    4. Yahoo 1h is ~2 years. Yahoo 5m / 15m / 30m are the last 60 days — that is Yahoo’s cap, not ours. A Real tape badge on those charts is a short confirmation, not a decade of edge.