Hesper Atlas

The historical replay, in numbers.

Every buy-to-sell signal in the current-rule historical replay, in aggregate: by year, by theme, the return distribution, the best and the worst. Closed round trips and the positions still open are both counted. Regenerated after every US close; retrospective, not live.

Historical replay Derived / calculated Not live performance
578
closed replay signals since 2022
64.9%
ended in profit
+39.7%
average return per closed signal
+65.1% / −7.2%
avg win vs avg loss: the asymmetry is the edge
67
median days held
10.0%
of signals ran longer than a year
111
signals still open, marked at the last close (median +61.8%), counted in the open book below
Data status: what these statistics are, and are not

Data status: historical replay. Generated 2026-09-11 16:39, today’s Hesper Atlas rules produce 578 closed BUY-to-SELL signals across 155 tracked names since 2022-06-21. 64.9% ended in profit; the average closed signal returned +39.7% over 135 days. The average winner made +65.1% while the average loser cost −7.2%, and 8.8% of all closed signals at least doubled.

The replay leaves 111 more signals open, 16.1% of all replay signals, and they are included rather than left out of the count: marked at the last close, 80.2% of them are in profit (median +61.8%, worst −20%). Counting the closed and the open book together, all 689 signals average +65.9% with 67.3% in profit. Open marks are unrealized and can still turn.

Source and status: every figure on this page is computed from the machine-readable JSON replay (field stats_ext), regenerated with current rules after every US close. Retrospective, losing signals included, not live performance. Educational tool, not financial advice.

The open replay book

Closed-trade statistics have a well-known failure mode: bank the winners, let the losers run, and the closed book still looks excellent. The only honest answer is to publish the positions the replay still leaves open. All 111 of them are aggregated here, 16.1% of all replay signals, so the closed numbers above can be checked against what was left out. The full open replay list is public on the track page. The latest end-of-day action and trade-plan levels remain the paid layer.

111
signals the replay leaves open at the latest close
16.1%
of all replay signals remain open
80.2%
of open replay signals are in profit at the latest close
+61.8%
median open mark
+3469% / −20%
best and worst open mark
463
median days held so far
Every signal, cut the same wayClosed (578)Open (111)All (689)
In profit64.9%80.2%67.3%
Up 20% or more28.5%61.3%33.8%
Up 50% or more15.2%55.0%21.6%
At least doubled8.8%39.6%13.8%
Down 10% or more8.5%4.5%7.8%
Down 20% or more2.1%0.9%1.9%
Median return+4.2%+61.8%+5.8%
Average return+39.7%+202%+65.9%

Closed returns are realized at the exit price. Open returns are marked at the last close, are unrealized, and can still turn: a position up 40% today can close flat. The point of the column is not to claim those gains, it is to show that the signals left out of the closed statistics are not the bad ones. Nor is the open column a like-for-like read: a position is still open precisely because its trend has not broken, so the open book skews to longer holds and larger marks by construction. What matters is the direction of that skew, and it runs against the closed figures rather than flattering them.

Results by year

Every closed signal, grouped by the year it exited. Trend engines earn their keep in trending years and get whipsawed in choppy ones; both show up here.

YearSignalsEnded in profitAvg returnMedian
20224233.3%−2%−2.7%
202313060.8%+10.4%+2.2%
202413068.5%+36.2%+11.3%
202515772.6%+53%+9.1%
202611966.4%+72.8%+3.9%

The current year contains only the signals that already closed; open positions join it when they exit.

What a closed signal looks like

The distribution behind the averages. A trend engine's math is asymmetry: it takes many small losses to stay in the game for the few very large wins.

8.8%
at least doubled (+100% or more)
15.2%
returned +50% or more
28.5%
returned +20% or more
64.9%
closed positive at all
8.5%
lost 10% or more
2.1%
lost 20% or more

Results by theme

Closed signals grouped by the theme of the underlying name, largest first.

ThemeSignalsEnded in profitAvg returnMedian
Genomics & AI-Health9957.6%+11.5%+2.6%
AI Software6460.9%+33.3%+4%
AI Power & Grid5862.1%+47.8%+4%
AI Chips5259.6%+31%+2.9%
Consumer & Internet5271.2%+14%+3%
Fintech & Crypto3961.5%+42.5%+7.4%
HBM & Packaging2871.4%+41.4%+9.3%
Quantum2458.3%+87.9%+16.4%
AI Servers & Systems1776.5%+89%+15.7%
Energy1668.8%+6.9%+3.8%
Industrials1662.5%+28.5%+2.6%
AI Cloud1566.7%+102.9%+9.4%

Ten best closed signals

SymbolEntryExitReturnDays
AXTI2025-08-282026-06-10+2820.9%286
PLTR2023-05-112026-01-28+1492.6%993
IREN2025-05-012025-11-05+1099.5%188
SMCI2022-09-062024-07-24+1055.5%687
RGTI2024-09-132025-02-26+987.9%166
POWL2022-10-252025-02-18+809.8%847
LITE2025-06-252026-07-06+696.8%376
VRT2023-04-262025-01-27+656.3%642
NVDA2022-12-092025-03-14+616.3%826
RGTI2025-04-212025-10-17+471.9%179

Ten worst closed signals

SymbolEntryExitReturnDays
RGTI2024-07-152024-08-05−33.1%21
MDB2022-08-112022-09-08−32.9%28
TER2025-02-262025-04-25−32.8%58
TEM2025-06-182026-03-20−32%275
IREN2024-07-262024-08-07−29.1%12
IONQ2023-09-112023-09-27−27.6%16
NU2024-09-192025-02-28−27.4%162
AXTI2023-01-272023-02-24−22.9%28
ON2024-02-222024-04-22−22.6%60
MSTR2025-05-062025-10-21−21.7%168

Walk-forward validation, separate from this replay

The replay above inherits hindsight. The stricter test re-selects each name's rules every January on earlier data only and trades the next year, 2016 to 2026, after costs: across 144 names the class engine returned a median +18.5% a year versus +20.7% holding, with a median worst drawdown of -50% versus -57%, shallower on 128 of 144. The self-audit has the benchmarks and the per-name detail; the walk-forward rows are public JSON and the method is on the methodology page. Backtested, not live.

The honest fine print, before quoting the numbers
  • These are the signals of TODAY'S engine replayed over history: the rules were researched and improved over time, so early-period signals are partly retrospective.
  • The universe is today's catalog (survivorship-biased: it contains names that earned their place; delisted or faded names aren't represented).
  • Returns are raw per-signal price returns (dividend-adjusted), end-of-day, without position sizing, costs or slippage; the in-app backtests apply 0.1% round-trip costs and vol-targeted sizing.
  • A trend engine exits AFTER tops and enters AFTER bottoms by design; win rate matters less than the size of wins vs losses.
  • Closed-signal statistics cover completed round trips only. Every still-running position is published alongside them (the 'open' list here, aggregated in stats_ext.open_book), marked at the last close and therefore unrealized: read the two together, because a record that banks its winners and leaves its losers open would look identical on the closed book alone.
  • Past performance (measured or backtested) does not predict future results. Educational tool, not investment advice.
Generated 2026-09-11 16:39 across 155 tracked names from end-of-day data · Source data: the machine-readable JSON ledger (field stats_ext) · Educational tool, not investment advice · Past performance (measured or backtested) does not predict future results.
Reading this with software? These statistics, their provenance and caveats are also a tool call away on the agent-verifiable evidence page, free and without an account.
Open Hesper Atlas →