Score floors for Trading agent entries

If you would not star a row by hand, paper Trading agent should not enter it either. Score floors are how that sentence becomes configuration. The agent should inherit the same floors you use on the Breakflare board: Breakout Score™, confidence, and no hype-only exceptions. Leaving floors at zero “to see more action” on live is not research curiosity. It is unscoped risk. Core attention research still does not place trades or publish price targets; floors only govern the optional agent.

Key Takeaways

• Copy manual promote floors into the agent: Breakout Score™ and confidence at minimum.
• Reject hype-only entries; require the same Climb and momentum discipline you use on the desk.
• Keep floors on in paper; zero floors teach the wrong distribution of fills.
• Review floors weekly, not hourly, unless a written regime rule changes them.
• Pair floors with size, stops, max hold, and kill switch (guardrails).

How do you copy desk floors into the agent?

Start from your hand checklist after the UTC snapshot:

1. Universe is CoinGecko ranks 51-200.
2. Sort by Breakout Score™.
3. Read hype, 3d/7d momentum, Climb, confidence.
4. Promote only when floors clear.

Whatever numbers or qualitative bars you use in step 4 become agent floors. If your desk rejects high hype / no climb, the agent needs that rejection too. If quiet climbs stay research-only (quiet climb), do not grant the agent a Climb loophole.

Write the floors in the same note as size and stops so they cannot drift independently.

Why is confidence a second lock?

Breakout Score™ can look interesting while coverage is thin. Confidence is the honesty column: missing Tier 2 feeds, unstable history, exploratory rows. A second lock stops the agent from treating incomplete rows like fully observed ones.

Practical pattern:

• Hard skip below confidence floor (low-confidence skip)
• Smaller size near the floor (optional, still written)
• No manual override spam for “I read a thread”

Reddit or sector heat does not repair ticker-level confidence; see Reddit in the hype stack.

Why is band membership a third lock?

Entries outside 51-200 are a different product problem. Top-50 graduates need archive habits (after top-50 upgrade). Microcaps below the band are outside Breakflare’s mid-cap attention design. The agent should not become a backdoor for out-of-universe tickers just because a feed mentioned them.

Stablecoin-like or zero-confidence edge cases also belong in skip logic (stablecoins / zero confidence). Floors are not only “how high is the score”; they are “is this even in scope?”

What does the floor table look like?

Map layers in the scoring stack hub. Floors encode your reading of that stack; they are not a second opaque model.

Review floors weekly, not hourly

Hourly floor edits usually track emotion: raise after pain, lower after boredom. Weekly review asks different questions:

• Are paper fills clustered just above a soft floor that should be higher?
• Did regime change (Bitcoin dominance) warrant a temporary tighter floor?
• Are pack days (Solana, restaking, RWA) producing correlated entries that need stricter relative rules?

Document changes with a date. Silent edits destroy after-action review.

The most expensive floor mistake is asymmetric: tight floors in the note, loose floors in the agent “just for paper.” Paper then teaches you a fill distribution you will not accept live, which makes graduation feel like a personality change instead of a switch flip. Keep them aligned (paper then live).

Floors versus the rest of the control set

Floors decide eligibility. They do not replace:

• Size caps
• Stops and max hold (max hold and stops)
• Kill switch rehearsal (kill switch drill)

A high floor with no stop is still incomplete. A perfect stop with zero floors is still a carnival.

Research cadence stays on the board: daily watchlist, false breakout when hype fades. The agent only automates gates you already trust.

Plus is $399/mo or $3,990/yr with a 14-day trial for live ranks, alerts, whales, and paper Trading agent. Start a Plus trial when you are ready to configure floors on the live product.

How should floors behave when the whole band compresses?

Sometimes Breakout Score™ values cluster and few rows look exceptional. Operators then lower floors to “keep the agent learning.” That trains the agent on marginal names exactly when edge discrimination is hardest. Prefer accepting sparse fills. Learning quality beats fill quantity.

If compression lasts many days, review whether your floor is absolute or percentile-like in spirit. Breakflare scores are already cross-sectional in design; your floor should still represent “I would promote this by hand today,” not “I need the agent to do something.”

Sparse fill weeks are also good weeks to drill kill switch and to re-read research habits on the board without automation pressure. The agent is optional. The 51-200 research workflow is not.

What is a healthy override policy?

If you must override, log: ticker, floor breached, reason, expiry of override, and whether paper or live. One-day overrides that somehow renew daily are just lower floors in costume. Prefer changing the weekly floor with a dated note over accumulating exceptions.

Never override confidence floors because a single social venue was exciting. That recreates the hype-only path floors exist to block. Community depth belongs in research notes, not as a master key around confidence.

How do floors interact with starring limits?

A desk that stars twelve names but encodes floors that would allow thirty agent entries is misaligned. Bring agent eligibility in line with shortlist capacity. If your shortlist size limit is five, floors should rarely clear dozens of rows on ordinary days.

When floors clear too many names, raise the composite floor or add a Climb confirmation rule before you blame the agent for overtrading. The agent is reflecting your eligibility surface. Make that surface match the research desk you actually run on the morning board, including prune cadence from prune rules.

Also sync language: “promote,” “watch,” and “skip” should mean the same thing in journal notes and in agent configuration comments. Shared vocabulary prevents soft floors from hiding inside synonyms.

How do you test a proposed floor change?

Change one floor variable, run a defined paper window (for example five UTC days), and compare fill count, false-entry rate versus your hand stars, and correlation of entries on pack days. If the change only increases fills without improving agreement with hand promotes, revert.

Do not A/B multiple floor changes at once. You will not know what worked. Store the trial in the same journal as research notes so agent configuration history is auditable. Floors are part of research ops, not a separate shadowy settings hobby.

What if hand starring and agent floors disagree for a week?

Stop and reconcile. Either your starring became looser than your written floors, or the agent configuration drifted. Pick the stricter interpretation until they match again. Disagreement is a process bug, not a reason to average the two policies into something neither desk nor agent intended.

Do floors replace reading the columns?

No. Floors are automation thresholds. You still read hype, momentum, Climb, and confidence on the morning board to keep judgment calibrated. If you only watch agent fills, floors will drift from what a careful hand promote looks like. Keep both loops alive.

Frequently asked questions

Can floors be zero?

Not for live, and not as a paper habit if you intend to graduate. Zero floors maximize action, not learning quality.

Do floors apply in paper?

Yes. Paper should mirror the floors you plan to keep.

What if the board is quiet?

Accept fewer fills. Lowering floors to manufacture activity undoes the point of a mid-cap eligibility filter.

Can I override one name?

Prefer changing the shortlist by hand over teaching the agent one-off exceptions. Exceptions become permanent.

Should hype score alone be a floor?

No. Hype is a layer. Use Breakout Score™ and confidence; consult momentum and Climb as in your desk rules.

Related reading

• Guardrails before live: floors inside the five-control set.
• Paper mode first: proving floors with fill logs.
• Crypto signal confidence score: what the second lock measures.
• Breakflare scoring stack explained: layers behind the floors.