Most strategy ideas die in one of two ways: untested, because setting up a proper test felt like a weekend project, or half-tested, because the quick backtest looked good and nobody asked the follow-up questions. This walkthrough is about the third path. We take a very ordinary TradingView-style script, the kind half of us have in a folder somewhere, and put it through a full validation pass in TideView: import, AI review, costed backtest, walk-forward, report. Wall-clock time, about fifteen minutes.
The point is not that fifteen minutes produces a tradeable system. It does not. The point is that fifteen minutes is enough to find out whether an idea deserves any more of your life, and most ideas do not survive the first ten.
The candidate: a plain moving-average breakout
Our test subject is deliberately unglamorous: go long when the 20-period simple moving average crosses above the 50, exit when it crosses back under. In Pine-style code:
//@version=5
strategy("SMA 20/50 Breakout", overlay=true)
fast = ta.sma(close, 20)
slow = ta.sma(close, 50)
if ta.crossover(fast, slow)
strategy.entry("Long", strategy.long)
if ta.crossunder(fast, slow)
strategy.close("Long")
No stop, no filters, no position sizing. That is intentional. If the validation pipeline cannot tell us something useful about the simplest possible strategy, it will not help with a complex one either.
Minutes 0-3: import
Paste the script into TideView's strategy editor, or point the importer at the file. The engine parses the Pine-style syntax into its internal strategy representation: entries, exits, and the indicator graph. Pick an instrument and timeframe with enough history to be meaningful; for this run I used a liquid index future proxy on daily bars, eight years of data. The import step also asks a question TradingView never forces you to answer up front: what execution assumption do you want? Choose "signal on close, execute next open." Remember that choice; it becomes relevant in about ninety seconds.
Minutes 3-6: the AI review
Before any backtest runs, the AI review pass reads the strategy the way a skeptical colleague would. On this script it comes back with a short list, and the list is the same one an experienced quant would write by hand:
- Same-bar execution risk. The original Pine defaults would fill entries on the bar where the crossover is computed. Since the crossover uses that bar's close, this is a mild but real look-ahead. Our "next open" import setting already fixes it; the review confirms the fix is active.
- No cost model attached. The raw script says nothing about commission or slippage. The review flags that the strategy trades often enough (roughly 15-20 round trips a year on daily bars) for costs to matter.
- No risk exit. There is no stop of any kind; the only exit is the opposite crossover. Not necessarily wrong, but worth knowing before you stare at a drawdown chart wondering why it is so deep.
- Parameter provenance. 20 and 50 are folklore numbers. The review notes they should be treated as a starting neighborhood, not a discovered truth.
None of this is magic. All of it is the stuff people skip when they are excited. Having it produced automatically, in seconds, is the difference between a checklist that exists and a checklist that gets used.
Minutes 6-10: backtest with real-world friction
First run: zero costs, just to establish the fantasy baseline. On my illustrative run the naive backtest showed roughly +180% cumulative over eight years with a maximum drawdown around 22%. This is the number a TradingView screenshot would show you, and it is the least real number in this article.
Second run: attach the cost model. I set commission at 0.03% per side and slippage at one tick per fill, both on the conservative-but-plausible end for a liquid instrument. Same strategy, same data. Cumulative return drops to roughly +115%, and the drawdown deepens slightly because losing streaks now carry friction. About a third of the paper profit was never going to reach an account.
That is actually a pass. A strategy with around 17 trades a year has room to absorb costs. The same experiment on a five-minute version of this script is a bloodbath; friction eats the entire edge, which is exactly the kind of thing you want to learn in minute eight rather than month three. If you want the full taxonomy of ways an uncosted backtest deceives, that is its own article.
Minutes 10-14: walk-forward
A single costed backtest still grades the strategy on history it could have been tuned to, so the last validation step is a walk-forward pass: 24-month training windows, 6-month test windows, optimizing only the two MA lengths within a sane range. TideView runs the windows and lays out the result per window. Abridged from my run:
Window Train span Test span Params OOS net OOS maxDD
1 2018-01..2019-12 2020-H1 18/52 +4.1% -6.8%
2 2018-07..2020-06 2020-H2 20/50 +7.9% -4.2%
3 2019-01..2020-12 2021-H1 22/48 +2.3% -5.1%
4 2019-07..2021-06 2021-H2 20/55 -1.8% -7.5%
...
11 2023-01..2024-12 2025-H1 21/50 +3.4% -4.9%
Reading it the way the previous post recommends: eight of eleven windows positive, parameters drifting gently around 20/50 rather than lurching, and out-of-sample performance retaining roughly half of in-sample. No single window carries the total. For a strategy this dumb, that is a genuinely respectable profile: not a money machine, but a real, boring, regime-dependent edge that behaves consistently when re-tuned on a schedule.
Minutes 14-15: the report
The last click exports a validation report: strategy source, every assumption made (execution timing, cost model, data span, optimization ranges), the costed backtest, and the full walk-forward table. The assumptions section is the part I care most about. Six months from now, "the strategy backtested well" is a useless memory; "next-open execution, 0.03% per side, one tick slippage, 24/6 walk-forward, eight of eleven windows positive" is a fact you can re-check, challenge, or hand to a friend who will try to tear it apart.
What fifteen minutes buys you
To be clear about the boundaries: this pass says nothing about position sizing, portfolio fit, or whether you can psychologically sit through an 18-month flat stretch, and an idea that passes deserves paper trading at small size next, not capital. What the fifteen minutes bought is a verdict on the question that matters first: is there anything here at all? For this little breakout script, the honest answer is "a modest something, in trending regimes, if costs stay low." For most scripts in most folders, the answer is "no," and getting that answer before lunch, rather than after a losing quarter, is the entire reason validation tooling exists.
For research and education only. Not investment advice.