CouchChange.Owner sign in

THE COUCHCHANGE RESEARCH HUB

Look inside
automated trading.

What actually happens between telling software to “make money” and its decision to buy, wait, or sell?

CouchChange explores algorithmic and agentic trading strategies. We explain how they work, open up the rules behind our experiments, and ask what the evidence can tell us.

A place to understand the machinery—and its limits. No trading experience needed.

See how it works Browse the studies →

FIRST, FOLLOW THE DECISION

A strategy turns an idea into a plan.

“Prices that have been rising might keep rising” is an idea. A strategy specifies what to observe, when to act, how much to commit, and when to leave. Software can carry out that plan in different ways.

THE RULE

Algorithmic trading

Software follows a defined procedure.

In a fixed-rule strategy, people specify the checks beforehand. The program applies them to each new observation, including when the answer is “do nothing.”

For example, a rule could…

  • Follow a trend. Compare recent prices with a longer average; consider buying when the required checks agree.
  • Wait for a recovery. Observe a dip, then require improving prices before entering.
  • Limit a position. Request an exit when a price or time limit is reached. The eventual sale price can still be worse.

Automation makes a rule repeatable. It doesn’t make the idea correct.

THE COORDINATOR

Agentic trading

An agent chooses steps and uses tools.

Within an assigned task and permissions, an agent can decide what to check next. It may use AI to interpret information, but needs explicit limits on what it can change or authorize.

An illustrative agent could…

  • Check a suspicious quote. Notice an old timestamp, consult an approved source, and wait if freshness cannot be verified.
  • Compare candidates. Request prices and cost estimates, apply eligibility checks, then propose a permitted paper trade—or none.
  • Investigate a loss. Inspect the decision log, test an explanation, and propose a separate experiment for review.

Choosing the next step adds flexibility. It also creates more ways to make a mistake.

THEN, TEST WHAT HAPPENS NEXT

A sensible decision can still lose money.

Suppose our fictional coin is still up 3% for the day and its hourly change turns positive. Both checks now pass. The software can record a hypothetical buy, but the market might fall immediately afterward. Even a small rise can be swallowed by the cost of buying and selling.

That is why we start with paper trading: simulated purchases and sales using real market observations. No money changes hands. It lets us examine a strategy’s behavior, while real-world execution remains unproven.

  1. Write the test first.

    Save the rules, budget, and cost assumptions before collecting new observations.

  2. Keep the whole record.

    Include waiting, losses, missing data, and trades that could not complete.

  3. Compare with doing less.

    Ask whether the strategy helped compared with holding the asset or staying in cash.

AI can help write code, investigate data, and explain a decision. A confident explanation—or one winning trade—cannot establish that a strategy has an advantage.

How we judge the evidence →

NOW, TRY IT FOR YOURSELF

From “AI trading” to “I see how that works.”

Trading 101 connects the pieces: rules, models, agents, costs, and evidence. Follow a fictional decision, change the cost assumptions, and see why a rising price can still produce a loss.

Start Trading 101 Try the short walkthrough →

Free to read. No account, wallet, or coding experience needed.

LOOK UNDER THE HOOD

Here’s how we put those ideas to work.

Our current paper studies turn specific questions into fixed rules. Open a study to inspect what it observes, what triggers a decision, and how it could fail. None has established a profitable advantage.

Current rule-based study

BITCOIN · ETHEREUM · SOLANA

Can a trend last?

  • Slow trendLooks for price strength across several time windows.
  • Trend pullbackWaits for a dip within an upward trend to begin recovering.

Both must cover their trading costs. We compare them with cash and a fixed holding basket.

Inspect the trend rules →
Current rule-based study

SOLANA MEME TOKENS

Does activity tell us anything?

  • MomentumLooks for a continuing price move.
  • Liquidity filterAdds a check on reported trading liquidity.
  • Selloff reboundLooks for recovery after a fall.

Busy trading can be misleading. We test whether these checks identify moves a simulated trade can capture after costs.

Inspect the activity rules →
Current rule-based study

NEWLY GRADUATED PUMP.FUN TOKENS

A recovery, or fading interest?

  • Graduation recoveryWatches newly launched tokens after they move into a trading pool and sell off.
  • Staged entry & exitRequires signs of recovery before buying in portions, then measures cash returned by later sales.

A falling price is not evidence of a bargain. Selling a portion does not automatically recover the initial cost.

Inspect the recovery rules →

These study designs are public. Recorded market results remain owner-only while publication rights are reviewed.

THE NEXT CHAPTER

Understand a strategy.
Then build your own.

Planned paid membership · In development

The public hub is where you learn and inspect our studies. The planned CouchChange Workshop will give members a place to turn their own questions into paper experiments, with chosen rules or bounded AI roles.

The aim is to make the process inspectable from the first idea to the final review, including the experiments that disappoint.

The member builder and subscriptions are not available yet.

THE WORKSHOP WE’RE BUILDING TOWARD

  1. Ask a question

    What might happen, and why?

  2. Define the decisions

    What can the system observe, do, and spend in simulation?

  3. Run a paper experiment

    Keep a fixed version and a complete record.

  4. Learn from the outcome

    What held up, what failed, and what deserves another test?

A preview of the intended workflow.

EXPLORE ONE QUESTION

Two useful distinctions.

Algorithmic vs agentic tradingWhat changes when a system can choose its next step?Paper trading vs backtestingWhat can each test tell you—and what remains unknown?