Technical Journal — Egis (Admin, Positions, BI, Time Invariants)

What I have done

• Continued shaping Egis Admin BI and operations with a strict backend-first, evidence-first approach.
• Extended Admin Dashboard metrics:
  • Updated Total Fee to correctly include funding fees.
  • Added Total Volume as a pure execution-derived metric.
  • Defined navigation from Total Fee → a new Admin > Fees page.
• Designed the Admin > Fees page with:
  • A clear summary section (total, trading fees, funding fees).
  • A paginated, backend-driven table listing all fee records.
• Refactored Admin > Positions for clarity and correctness:
  • Merged opened_at and closed_at into a single Duration column.
  • Duration is human-readable, backend-generated, and excludes seconds.
  • Open positions display - instead of live timers.
  • Added hover tooltips to reveal exact opened / closed datetimes when needed.
• Investigated and fixed multiple timestamp consistency issues:
  • Detected impossible cases where closed_at < opened_at.
  • Fixed the root cause instead of masking it at the UI layer.
  • Discovered a follow-up issue where some CLOSED positions lost closed_at.
  • Enforced the invariant: status = CLOSED requires a valid closed_at or an explicit reason.
• Identified a new anomaly:
  • Positions with IDs 41–55 showing opened_at == closed_at.
  • Initiated a structured investigation to trace this back to execution data, repair logic, or timestamp normalization.
• Defined multiple hard invariants and investigative playbooks for Cursor:
  • Positions lifecycle must be derived from executions, not assumptions.
  • No silent timestamp “fixing”.
  • All BI and display values must be explainable under stress.

---

What I have learned

• Time is one of the most fragile domains in trading systems:
  • A single guard clause meant to “fix” bad data can quietly corrupt historical truth.
  • Every timestamp must have a clear origin: execution time, not import time or UI time.
• UI simplification increases backend responsibility:
  • Replacing raw timestamps with Duration made data inconsistencies more visible, not less.
  • This is good: it forces correctness instead of hiding problems.
• Status-based assumptions are dangerous:
  • status = CLOSED does not automatically mean all lifecycle fields are valid.
  • Status transitions must be atomic with their timestamps.
• Backend-first formatting is not just aesthetics:
  • Centralizing datetime, currency, and duration formatting eliminates entire classes of bugs.
  • It also makes cross-page consistency enforceable, not aspirational.
• BI features should reinforce invariants, not create overlap:
  • Separating execution quality, fees, volume, and PnL keeps each metric auditable.
  • Removing seconds from duration is a UX choice that also reduces false precision.
• Every fix must create a recovery path:
  • Nulling bad data without recomputation logic only defers the bug.
  • Repair commands and evidence logs are essential for long-lived systems.