Trading Stocks

The Trading Stocks capability is where a portfolio decision becomes a concrete order. It is also where the Finance family’s authority boundary is drawn most sharply, because the gap between "the agent thinks this is a good trade" and "the agent has placed this trade" is the gap that matters most.

Analysis, autonomously#

• Instrument analysis — fundamentals, valuation context and recent price behaviour are assembled into a dated view with sources attributed.
• Order construction — a proposed order is specified in full: side, quantity, order type, limit or stop levels, time-in-force and the assumptions behind each.
• Cost and impact estimation — expected spread, commission and market impact are modelled so the true cost of a trade is visible before it is authorised.
• Scenario framing — the outcomes if the thesis is right, wrong, or slow are stated up front rather than discovered afterward.

Execution, only across the boundary#

Placing an order is a distinct, explicitly authorised action. The capability constructs the order and presents it; a human with execution authority approves it; only then is it routed. No confidence level, market condition or phrasing of the objective collapses these into one step. This is enforced by the permission system described under Permission System, not by the model’s judgement.

proposal    BUY  120 shares  limit 0.5% below last
            time-in-force: day    as-of 2026-09-06 14:02
rationale   completes the fixed-income-funded equity trim from
            the approved rebalancing proposal of 2026-09-05
checks      within single-name concentration limit        pass
            within daily notional budget                   pass
            idempotency key attached                       pass
authority   HELD  —  requires human execution approval
            agent will not route until approval is recorded

After execution#

An executed order is reconciled back against the portfolio and the originating proposal, so the record shows not just that a trade happened but which decision it fulfilled and whether it filled as modelled. Slippage between the modelled and realised cost is captured and fed back, because a capability that never compares its estimates to reality never improves them.