What Arch-AGI is, in plain language

Arch-AGI is the deep-intelligence layer behind every Treeova AI options trading agent. It is the part of the platform that decides whether an idea is good enough to act on — not by guessing, but by scoring agentic AI trading conviction the way a disciplined human trader would. Where a typical model spits out a single buy / sell number, Arch-AGI runs a structured, multi-pass review and asks the same questions an experienced options trader would ask before risking real capital: what is the edge, what would invalidate the thesis, what does the market regime tell us, and how have we been wrong about setups like this in the past?

The Treeova whitepaper on the conviction methodology puts it plainly:

“The output is a conviction score plus a written narrative — not a trade signal in isolation, and never a recommendation.”
— Treeova whitepaper, Arch-AGI methodology

That single sentence captures why Arch-AGI exists. A trading platform that can only say "buy now" is gambling-flavoured automation. A platform that says "here is a conviction score, here is the written reasoning, and here is the historical track record of similar setups in this regime" is closer to having a research desk inside your account. Conviction-gated AI trading is the difference between a black box and a glass box.

How it works — the seven-pass conviction loop

Every Arch-AGI report walks a position through seven structured passes. Each pass is a focused question, and each answer feeds the next. The order matters: weaker passes feed into stronger gates, and the final adversarial pass exists precisely to push back on the earlier ones.

Pass one — edge analysis — what is the thesis, and what specific market behaviour would prove it wrong?
Pass two — scenario projection — walking the position through best-case, base-case, and worst-case outcomes so the AI options trading agent is not just modelling the rosy path.
Pass three — risk and reward framing — breakeven, optimal exit, and stop-loss zones, all expressed against live options-chain liquidity rather than yesterday's snapshot.
Pass four — regime overlay — the current market regime is layered in — trend, chop, high-IV, low-IV, post-earnings drift — and the engine asks whether the setup actually fits the regime it is being deployed into.
Pass five — macro overlay — relevant tailwinds, headwinds, scheduled events, and cross-asset flows are folded in so the thesis is stress-tested against the wider market backdrop, not just the single ticker.
Pass six — calibration — the step that makes Arch-AGI honest over time. The whitepaper describes the loop directly, and it is what turns conviction from a one-time guess into a self-correcting estimate.
Pass seven — adversarial stress test — the same engine deliberately attacks its own thesis: what would a smart short-seller of this idea say? What does the dealer-positioning data imply? If the stress delta is large — that is, if the adversarial pass meaningfully weakens the case — Arch-AGI drops the conviction score before any trade gate is even consulted.

“Every closed position feeds back into a regime-segmented calibration table.”
— Treeova whitepaper, Arch-AGI methodology

In other words, every time an Arch-AGI-scored position closes — winner, loser, or scratch — the outcome flows back into a calibration table that is segmented by regime. If conviction scores of 0.7 in a high-IV regime have historically only resolved profitably 45% of the time, future 0.7 scores in that regime get re-weighted. AI trade conviction scoring is therefore not a one-time prediction; it is a reinforcement-calibrated running estimate of how reliable the agent's own conviction has been, segmented by the market it is operating in.

Where Arch-AGI fits in the wider Treeova stack

Arch-AGI does not work in isolation. It is the conviction layer that sits inside Treeova's Meta-Agent Trading Stack, drawing on every other intelligence source the platform exposes. The Navigator copilot can request an Arch-AGI report mid-conversation when you ask "is this a good setup?" — the agentic options trading flow is built so a discretionary trader, a rules-based seller, and a fully automated trading agent all talk to the same conviction engine.

MetaChart snapshots feed Arch-AGI the same pixel-stable image the trader is looking at, so the reasoning is grounded in the actual chart rather than a hallucinated description. Kronos handles the scheduling — pre-market scans, regular-hours execution, after-hours reviews — so Arch-AGI runs at the moments where its output is actually actionable. Lossless context management keeps the agent's working memory intact across long sessions, so a conviction score generated at the open is still defensible when the close arrives.

Crucially, conviction-gated AI trading is enforced at the executor, not at the prompt. Alert-only agents will never place an order even if a high-conviction signal screams at them; trading agents will refuse to fire below a conviction floor, regardless of how persuasive the language wrapped around the score happens to be. The modality wall is hard-coded so the trader's risk envelope cannot be talked into being bypassed.

What this is not

It is worth being precise about the limits, because conviction-first language can be misread as a profit guarantee. Arch-AGI is not a recommendation engine. Its reports describe a position; they never tell a trader to open, close, or size a position. It is not a backtest either — conviction scores reflect the current context, not a historical replay. And it is not infallible: the adversarial pass exists specifically because the earlier six passes can be wrong, and the regime-segmented calibration loop exists because past calibration does not guarantee future calibration. The platform is intentionally conservative when data for a given regime is sparse, and the conviction score will refuse to climb past what the calibration table actually supports.

This is also why Arch-AGI is not the same as the whitepaper that documents it. The whitepaper is the methodology source-of-truth — the formal walk-through of every pass, every safeguard, every known limitation, and every threshold the platform reserves the right not to publish. This feature page is the readable translation. For the full seven-pass breakdown, calibration loop, and disclosed limitations, the whitepaper at /whitepapers/arch-agi-conviction-methodology is the authoritative reference.

Use cases — who actually benefits, and when

The clearest beneficiary is the discretionary options trader who already has good instincts but cannot watch every ticker. An Alert-only AI options trading agent gated on Arch-AGI conviction will sit quietly through hundreds of weak setups and only ping when the score crosses the trader's chosen floor. The alert is not just a signal — it carries the written narrative, the regime context, and the calibration history for similar setups. That is far closer to the second-opinion conversation a trader would have with a mentor than a single-line buy alert from a chat-room bot.

The rules-based options seller — the trader who sells delta-30 put credit spreads every Monday, for example — uses Arch-AGI as a discipline layer. The agent already knows the structure; what it gains is the regime overlay and the calibration loop. If selling delta-30 spreads in a high-IV regime has historically underperformed for that agent, the conviction score for the same mechanical setup will be lower, and the conviction-gated executor will trim size or skip entirely. Reinforcement learning happens automatically in the background as positions settle.

The systematic builder running a small fleet of specialised agents — one for trend days, one for earnings IV crush, one for mean-reverting range setups — uses Arch-AGI to keep them honest against each other. Each agent gets its own calibration history, so a strong earnings-IV agent does not get to borrow credibility from a weak trend agent. Conviction is per-agent, per-regime, and fully auditable. The platform never let one of those agents quietly drift into trading a regime it was never calibrated for.

Finally, the trader who has been burned by black-box signal services uses Arch-AGI as a transparency contract. Every score comes with the written narrative behind it. Every closed trade updates the calibration table that those future scores will be measured against. That trail of evidence — agentic AI trading conviction with an audit log attached — is the actual product.

What happens when you start using it

On day one, the recommended path is Alert-only modality pointed at a paper account. The trader watches Arch-AGI reports roll in, reads the narratives, and decides whether to act. Within a week or two, the calibration table begins to populate with real closed outcomes — even paper outcomes feed the loop, because the same execution engine and price oracle that fills live orders fills paper ones, with phantom-fill protection keeping the data honest.

In month one, the trader graduates the agent to Trading modality on the paper account. Conviction-gated execution starts placing trades; the risk envelope (stop-loss, profit target, 20% concentration cap, $50 minimum buying power gate) sits in front of every order. The Live Observability Pulse streams a 60-second health beat showing exactly which passes ran, what each one said, and where the conviction score landed. Nothing is hidden behind a chat bubble.

By month two or three — only if the trader is comfortable — the same agent is re-delegated to a live broker account through Treeova's broker integrations (Robinhood, Tastytrade, Webull, TradeStation, Alpaca, and Lightspeed today, with more on the way through Connect Trade). The same Arch-AGI conviction loop, the same calibration table, the same audit trail. The only thing that changes is the destination. The trader keeps the kill switch.

The long-term outcome is not a money-printer — Arch-AGI does not predict the future, and the whitepaper is unambiguous about that. The long-term outcome is process: fewer impulse trades, fewer setups acted on against the regime, more discipline at the moments that actually matter, and a fully auditable record of why every position was opened. Conviction-first agentic AI trading is, in the end, a way of making the trader's own best decisions reproducible — and of catching the moments when those decisions would have been wrong.

Frequently asked questions

What is the difference between Arch-AGI and a trade signal?

A trade signal is a single number — buy, sell, or hold — with no reasoning attached. Arch-AGI is the conviction layer that produces a 0–100 score plus a written narrative covering edge, scenarios, risk, regime, macro, calibration, and an adversarial stress test. The output is evidence, not an instruction; the trader (or the trading agent's executor) decides whether to act on it.

Does Arch-AGI place trades automatically?

Arch-AGI itself never routes orders. It scores ideas. Whether an idea becomes an order is decided downstream — by the trader in Navigator, or by a Trading-modality agent whose conviction-gated executor checks Arch-AGI's score against the trader's own threshold. Alert-only agents can never silently flip into trading.

What is a 7-pass conviction loop?

Each candidate trade walks through seven structured passes in order: edge analysis, scenario projection, risk and reward framing, regime overlay, macro overlay, calibration, and adversarial stress test. Earlier passes feed later ones, and the final adversarial pass exists specifically to push back on the first six. The conviction score is the agreement across all seven.

How does regime-segmented calibration work?

Every closed position feeds back into a calibration table that is segmented by market regime — trend, chop, high-IV, low-IV, post-earnings drift, and so on. If conviction scores of 0.7 in a high-IV regime have historically resolved profitably only 45% of the time, future 0.7 scores in that regime are re-weighted accordingly. Conviction becomes a self-correcting estimate rather than a one-time guess.

What happens when Arch-AGI's adversarial pass disagrees with the earlier passes?

The adversarial pass deliberately attacks the thesis the first six passes built. When that attack meaningfully weakens the case, Arch-AGI drops the final conviction score before any trade gate is consulted. Disagreement is the point — a high published conviction means the thesis survived its own steel-manned counter-argument.