Martin ZhangPortfolio · Toronto
Open to opportunities
Back to all recordsAll projectsOpen the project page
LOG-004 / REPLAYPublic record · DOC v3.1.0 ·

Systems & trust

Run Replay

Let finance operators run month-end controls, reconcile cross-system data, prepare journal drafts, test forecast scenarios, and audit decisions without giving AI control over financial values.

Primary audience
Accounting, finance operations, FP&A, payroll, system owners, reviewers, and approvers working under close deadlines and control requirements.
My role
Workflow design · Financial controls · Scenario modeling · AI evaluation · Implementation
Working stack
Next.js API routes · deterministic JavaScript · Claude API · zod · node:test · PostgreSQL design · GitHub Actions
Status
Prototype · The inspiration for Tickmark

01 / Shipped artifact

Summary

Two server-side control engines, four finance source contracts, journal preparation, exception review, a deterministic scenario model, validated AI briefs with fallback, and human-only release gates.

Quantified snapshot
22
Deterministic controls
4
Finance source contracts
3
Balanced journal drafts
98.5%
Payroll-engine exception recall
File & data footprint
  • 22 deterministic controls across 2 engines
  • 63-row, 4-source finance close sample
  • 3 balanced journal drafts and 3 forecast scenarios
  • 74 evaluation cases
  • 98.5% recall, 0 false alerts
  • 10 automated tests

02 / Content and UX decisions

Decisions

DDR-001 · Content decision

Make every stage, reconciliation, journal draft, scenario formula, and exception legible as evidence: what happened, what value was used, and what a reviewer can do next.

DDR-002 · Experience decision

Use an application-style workbench with separate Close, Review, Journal, and Model pages; keep the result area empty before a run and lock release until all high-risk blockers receive a human decision.

03 / Delivery evidence

Delivery evidence

  1. 01

    A five-stage Ingest → Validate → Reconcile → Prepare JEs → Route Review finance-close pipeline with source fingerprints.

  2. 02

    A 15-item exception queue across GL, cards, close tasks, and plan variance.

  3. 03

    Balanced journal drafts, CSV exports, a human release gate, and an interactive actual-to-forecast model.

04 / Role signals

Role fit

  • Read schemas and document system behavior
  • UI copy for errors and review states
  • Automation scripts and checks
  • Journal preparation and financial modeling
  • Human-centered AI boundaries
Validation

Ten automated tests, production build and type checks, deterministic evidence, balanced-entry assertions, decision logs, keyboard operation, reduced motion, responsive QA, and synthetic-data-only demos.

Boundary

No live company or employee data, claimed vendor integration, autonomous accounting calculation, posting, or AI approval.

05 / Live project change feed

Live changes

Sync sourcePublic portfolio integrationPublished portfolio integration

15 changes captured from portfolio pushes

  1. Add Tickmark as its own project; reframe Run Replay as its prototype

    940b99cMartin Zhang

  2. Link Tickmark, the rebuild, from the Run Replay hero

    5953bcbMartin Zhang

  3. Redesign Run Replay as tactile finance workspace

    ee07f6eMartin Zhang

  4. Expand Run Replay into finance operations control room

    ed06cb4Martin Zhang

  5. Run Replay 2.0: AI Operations Exception Review System; cloche hero; Next 16.4.2

    250b263Martin Zhang

  6. Reposition portfolio for AI roles

    576383fMartin Zhang

  7. feat: split Run Replay into app pages

    8a94190Martin Zhang

  8. fix: satisfy production comparison types

    355e3b4Martin Zhang

  9. feat: recreate Run Replay paper workspace

    83cdf8fMartin Zhang

  10. feat: rebuild Run Replay as operations dashboard

    2f65aaeMartin Zhang

  11. feat: turn payroll demo into run replay

    4a7afd2Martin Zhang

  12. Give payroll agent a standalone product identity

    dfac672Martin Zhang

  13. Gate reconciliation results behind run

    7f88fc7Martin Zhang

  14. Expand automation skills and career pathways

    ea2aa88Martin Zhang

  15. Add Payroll Control Room case study

    e619197Martin Zhang

Private source; deployment-safe metadata only.

06 / Documentation record history

Record history

Documentation version, not product release.

Documentation version
v3.1.0
Last updated
  1. v3.1.0

    Reframed Run Replay as a quiet, tactile finance workspace: replaced the dark purple AI-dashboard treatment with a cool mineral palette, clearer application chrome, restrained orange action cues, denser data hierarchy, calmer state motion, and a responsive tabbed workbench informed by Apple interface principles, professional creative tools, and modern close-management products.

  2. v3.0.0

    Expanded Run Replay into a Finance Operations Control Room for the Opendoor Finance Systems & Automation Specialist role: four synthetic vendor-shaped sources, 10 accounting controls, cross-system reconciliation, balanced journal drafts, a 15-item human review queue, an automation opportunity backlog, audit evidence, CSV exports, and a deterministic actual-to-forecast scenario model.

  3. v2.0.0

    Rebuilt as an AI Operations Exception Review System: server-side rule engine, validated Claude briefs with deterministic fallback, a 74-case evaluation harness with recorded failures, observability, CI tests and an honest post-mortem.

  4. v1.6.0

    Turned Run Replay into a paginated application with deep-linked Run, Review, and Project views; replaced the long case-study scroll with one concise project brief and a dedicated exception review workspace.

  5. v1.5.0

    Recreated Run Replay from the selected paper-file concept with source folders, a three-tab comparison sheet, a Review 8 evidence file, a four-frame motion storyboard, and functional run and review interactions.

  6. v1.4.0

    Rebuilt Run Replay as an independent payroll operations dashboard with a persistent tool rail, run controls, an empty first-run state, a Queue / Evidence review workspace, a pinned replay timeline, functional pause control, and responsive mobile behavior.

  7. v1.3.0

    Renamed the product Run Replay and rebuilt the demo around a replayable five-stage timeline, stage receipts, a Review stop, and a four-blocker Release Gate.

  8. v1.2.0

    Connected a live GitHub change feed with source attribution, commit links, and a resilient published-history fallback.

  9. v1.1.0

    Added pipeline and audit metrics, documentation-version metadata, and a public change history.

  10. v1.0.0

    Published the initial public record with purpose, decisions, delivery evidence, validation, and boundaries.