Build-up begins
Scaffolding the separate stack — Lambda, DynamoDB, Sheets Control, Swing / 2-Step defaults — without touching the live Horizon crypto bot.
Horizon Horizon – FTMO
Lab · Personal research · Not affiliated with FTMO · Not for sale · Not financial advice
Horizon – FTMO
A private prop trial, built in the open — for me.
Horizon – FTMO is a Lab experiment under Horizon: my own AWS automation, Google Sheets control plane, and FTMO-style rule set on a spot-crypto book. The formal trial opened on 1 August 2026. The engineering started months earlier, in January. It is not a product, not a signal service, and not a shortcut to passing Challenges.
Timeline
One dated start. Everything after is the shape of the work — not a calendar of every deploy.
Formal trial opens
Live learning under prop Objectives begins. Spot crypto first; other categories gated until the book behaves.
Velocity meets reality
High-urgency challenge thinking meets large percentage swings. Recoveries happen — but they teach that speed is not the same as survival.
Preservation model
Risk dials, prop pause / flatten, CEST day baselines, and a lean ~3-minute serverless cycle become the spine — not optional extras.
Steadying growth
Regime gates and basket harvests are in the mix. The curve is more patient. The trial still has months left to prove the model.
Harvest honesty and extraction lock
Basket take-profits become fee-aware (floor + target % of capital, winners-first). Sheets fail closed on bad reads. A payout lock can flatten the book at a target-plus-buffer and refuse new risk until a manual reset — so “at Objective” can mean ready to extract, not still averaging.
What changed along the way
Highlights of the rebuild — undated on purpose. Plenty of smaller fixes sit behind these.
- Sheets as the control plane Risk floors, category toggles, sizing, harvest switches and prop guards live on a Control tab — macro changes without redeploying for every tweak. Transparent dials, not a black box.
- Lean AWS serverless cycle Lambda + DynamoDB + CloudWatch + EventBridge, tuned for a tight ~3-minute main tick (and a slower market worker). Built for lean execution and less local editor lag, not an always-on box.
- Prop-aware guards Growth pause and flatten thresholds aimed at Maximum Daily Loss and Maximum Loss — automation that respects the Objectives that end Challenges.
- Regime filtering Higher-timeframe trend gates (including an H4 “sniper” filter) decide whether cold NEW entries are allowed — fewer falling knives.
- Basket take-profits The open spot-coin book treated as one synthetic basket. Fee-aware sweeps use a floor plus a target % of capital, winners-only by default, and a haircut for round-trip costs — so harvests are realised progress, not optical float.
- Sheets fail-closed Control is trusted when warm; a cold miss or consecutive bad reads pause trading instead of guessing dials from an empty sheet.
- Payout extraction lock Sticky arm at equity target plus buffer: flatten the book, block replenishment, and stay locked until Kill Switch Reset — so “at Objective” means ready to extract, not still averaging.
- Sizing spine Target Bid and lot-cap dials so ideas stay equal-weight and one name cannot quietly dominate margin.
Highs, lows, and the growth now showing
Highs
- End-to-end ownership: AWS, Sheets Control, and the FTMO path on one desk.
- Regime + basket upgrades that finally feel like they match how the book should behave.
- A control surface I can read in one glance — dry run, categories, prop posture.
Lows
- Early percentage drawdowns that made “challenge velocity” look foolish.
- Moments where the stack was busy but the book was still catching knives — until the gates arrived.
Where it sits now: the trail from January’s scaffold through the August trial start is starting to show steadier behaviour — more patience, clearer harvests, and progress measured against Objectives rather than against a stopwatch. Growth is real enough to keep going; it is not a finished pass story.
Lessons so far
Capital preservation beats velocity. Passing Objectives only counts if the book is still intact when you get there.
Visible dials beat clever opacity. If I cannot see the rule on the sheet, I will not trust the tick.
Gates before adds. Higher-timeframe disagreement is a reason to stand down — not to average harder.
Harvest the book, not just the ticket. Aggregate strength can evaporate ticket-by-ticket; a basket sweep respects that.
At target means flat and locked — not still hunting. Extraction is a state machine: flatten, refuse new risk, reset only by hand.
Prop ceilings are the product. Max Daily Loss and Max Loss are not footnotes — they shape every other dial.
Still to try in the remaining trial
A short list — not a roadmap pitch. The trial runs toward mid-2027; these are the next honest experiments.
- Pressure-test regime gates and basket sweeps through a few more market moods before widening categories.
- Keep tuning lot-cap / Target Bid so the book stays diversified without starving good names.
- Only consider a paid Challenge path if dry-run and trial evidence still look calm under daily-loss stress — never assumed.
Under Horizon
Built by Phil Stone
Private Horizon – FTMO work in Northern Ireland. Meet Phil · Get in touch
Sits alongside Horizon’s other crypto automation work: same engineering habits, different rule set. Private development — no public signup and no absolute performance figures here.
Trading and FTMO Evaluation Processes involve risk, including the loss of Challenge fees and failing Trading Objectives — especially Maximum Daily Loss and Maximum Loss. FTMO uses demo accounts with fictitious funds; any Rewards follow FTMO’s terms. This page describes private Horizon – FTMO infrastructure only. It is not financial advice, not a signal service, not a guarantee of passing any Challenge, not an invitation to invest or trade, and not affiliated with or endorsed by FTMO. Always read the current terms on ftmo.com before you start anything with FTMO.