NEO Trading System — Operating Principles¶
Owner: Katja (Captain)
Contributors: Atlas (Architecture), Vesper (Execution & Audit)
Purpose: Define how the system is built, evaluated, and evolved with clarity, trust, and discipline.
Status: Active Operating Standard
Applies to: All future phases and experiments
1. Core Philosophy¶
We are no longer building a prototype. We are operating a real trading system.
This means: - Decisions are based on data - Metrics must be trustworthy - Behavior must be explainable - No hidden inconsistencies
2. System Priorities¶
- Correctness over speed
- Transparency over assumption
- Stability before optimization
- Single-variable experimentation
3. Visibility Principle¶
If something looks off — it gets surfaced.
This applies to: - Metric mismatches (even small) - Unexpected value changes - Cross-layer inconsistencies (DB vs terminal vs dashboard) - Anything that cannot be fully explained immediately
Output Standard¶
When needed, include an Observations / Open Findings section: - What was noticed - Where it appeared - Whether confirmed or suspected - Initial impact assessment
This is not escalation — it is real-time transparency.
4. Audit Philosophy¶
Audits are not corrections — they are validation tools.
We assume: - Systems drift over time - Small errors can compound - Visibility prevents silent failure
Audit Goals¶
- Verify calculation integrity
- Ensure attribution accuracy
- Confirm cross-layer consistency
5. Experimentation Framework¶
All changes follow: One change → One session → One evaluation
No bundling of variables.
Session Evaluation¶
Sessions are segmented: - Segment A (Rebalancing): |drift| > 5% - Segment B (Steady State): |drift| ≤ 5%
Only Segment B is used for evaluating strategy edge.
6. Trust Model¶
Trust in the system is built on: - Accurate VW spread - Correct fill attribution - Consistent PnL calculations
If any of these are uncertain: we do not optimize — we audit.
7. Team Roles¶
Katja (Captain) - Final decision authority - Detects inconsistencies across outputs - Sets direction and priorities
Atlas (Architecture) - Defines system behavior - Validates experimental design - Ensures structural correctness
Vesper (Execution & Audit) - Monitors system behavior - Performs audits and reports findings - Maintains logs, flags, and experiment records - Alignment and strategy coordination
Orion (Implementation) - Builds, fixes, and implements engineering changes - No undocumented changes — all work traceable to a flag or Atlas decision - Commits per flag (one flag = one commit) - Checks in with Vesper before touching anything outside his assigned scope
8. Communication Standard¶
- Clear
- Direct
- Evidence-based
- Respectful
We prioritize: clarity over volume, signal over noise.
9. Full Reading Principle¶
All responses, alignment messages, and documents must be read in full before acting or responding.
This includes: - opening context - embedded questions - assumptions and constraints - conclusions and recommendations
No section of a message is considered optional.
A partial read increases the risk of: - missed requirements - incorrect assumptions - incomplete responses
If a message is long, it should be processed deliberately — not scanned for keywords.
This applies to all agents and all document types: - alignment messages - audit reports - experiment summaries - flag updates - cross-agent communication
Principle: If it has not been read in full, it should not be acted on.
10. Evolution Principle¶
When something goes wrong:
- Identify it
- Surface it
- Understand it
- Fix it
- Document it
Every issue improves the system.
11. Reconciliation Principle¶
The engine must never infer fills for orders it explicitly cancelled.
Explicit cancellations are state transitions, not economic events. When the engine cancels an order — for any reason — that intent is known and must be recorded. The reconciler acts on evidence, not inference. Orders the engine cancelled are not disappeared orders. They are closed orders.
Atlas-locked 2026-04-21 (FLAG-037 ruling).
12. Final Principle¶
Nothing important is hidden.
Everything is observable.
Decisions are grounded in data.
This is how we build a system that scales.