Institutional portfolio operations

MANDATE.

Portfolio implementation and control, made traceable.

Mandate reconciles fragmented portfolio data, evaluates effective-dated controls, and produces an auditable path from rebalance intent to authorized handoff.

Launch Mandate View architecture

Status: Synthetic product concept
Role: Systems design and build, solo
Stack: Next.js · TypeScript · Zod · Vitest
Affiliation: Independent. No firm, no client data

The Mandate account view for the synthetic Alder Family Tax-Managed Equity SMA: a blocked readiness verdict, the scenario rail, three blocking controls with owner and client impact, and a dependency-ordered next-actions panel. — One synthetic account, mid-rebalance. The verdict, the control that produced it, the evidence behind that control, and the action that clears it are all on one screen.Open at full size ↗

01
Readiness
Blocked · 3 controls
02
Blocking control
DATA-FRESH-001 · Fail
03
Evidence
Source 2.7 d old vs an 18 h window
04
Next action
Owned, and it unblocks 4 controls

The problem

The answer is spread across seven systems.

Material sources

Custodian holdings
Internal accounting
Model portfolio
Security master
Mandate document
Tax-lot ledger
Transaction history

MANDATE.

Reconcile · evaluate · order · authorize

One canonical account state

Recorded facts · append-only events · one derived projection

A rebalance decision depends on facts distributed across custodians, policy documents, tax context, model data, and human approvals. Readiness gets asserted because reconstructing it is slower than the trading window. Mandate makes the reconstruction the product.

Mandate's data lineage view: each source snapshot listed with its as-of time, age, freshness window, state, and the controls it feeds, with the custodian holdings file marked stale. — Fresh, complete, and consistent are different properties. One stale source is visible here, along with every control waiting downstream of it.Open at full size ↗

The product insight

An account is not ready or not ready by opinion.

Readiness is derived.

Facts

Recorded, sourced, timestamped

Effective-dated controls

Versioned policy, not opinion

Dependencies

What must be trusted first

Human authority

Who may release, and against what

The current mandate state

Re-derived on every new fact · never edited by hand

Blocked, and why

Blocked

3 unresolved controls

DATA-FRESH-001
Custodian position file is stale against its freshness window
MANDATE-RESTRICT-004
A proposed buy conflicts with a verified client exclusion
TAX-BASIS-003
A selected sell lot has no usable cost basis

Mandate does not simply flag the portfolio. It exposes which control failed, the evidence behind it, the policy version in force, the required resolution, and who must authorize release.

Mandate's portfolio impact view after the restricted buy is substituted, showing model deviation moving from 8.00% to 4.61% alongside turnover, residual cash, largest single name, buy and sell value, and a projected realized result still reported as unknown. — Complying with the restriction cost index-tracking accuracy, and the product says so. Resolution is not a completion checkbox; it is a change with a price, approved against the projection it produces.Open at full size ↗

See it in the product

State transition

Same account. Same policy. New evidence. New state.

01
Blocked
3 unresolved controls. 5 more report they could not evaluate and name the dependency they are waiting on.
Fresh source ingested, authoritative record named
02
Needs authorized review
No blockers remain. Three warnings are carried on the record rather than cleared, including a documented wash-sale review.
Portfolio manager authorization, with a reason
03
Handoff approved
A checksummed synthetic package bound to one exact projection version. It says approved. It never says executed.

Each step appends real domain events and re-runs the same projection. Readiness changes because the facts changed, not because a screen advanced.

Run the sequence

Control trace

One control, and everything behind its verdict.

The blocker that most other controls on this account are waiting on, as the product publishes it.

Control: DATA-FRESH-001Material source freshness
Version: v1.4.0Effective Jan 01, 2026 → open
Verdict: FailRequirement: rebalance blocker
Input: Custodian position fileAs of Mar 13, 2026, 21:00 UTC
Measured: 2.7 days oldAgainst an 18-hour window
Source class: Product assumptionIllustrative, not any firm’s policy
Owner: Data StewardResolution is owned, not unassigned
Next action: Request and ingest a fresh snapshotClearing it unblocks 4 controls
Authority: Release requires a portfolio managerBound to the resulting projection

Verdict and requirement are separate axes. The same finding can be a hard stop under one mandate and a warning under another, which is why the catalog records both.

Mandate's decision trace: every control listed with its version, verdict, requirement class, reason, owner, and a link to the evidence and calculation behind each result. — Every control on the account, with the evidence drawer one click away. Nothing here is summarized into a score.Open at full size ↗

Open the decision trace

Mandate's control registry: twelve effective-dated controls, each with version, effective interval, owner, source class, severity mapping, test count, and the outcome it produced on the flagship account. — The registry is read-only by design. Source class records where each rule came from — regulatory guidance or an explicit product assumption — so a reviewer can tell a definition from an invention.Open at full size ↗

Open the registry

How it works

Reconcile. Control. Authorize.

01
Reconcile
Fragmented evidence becomes one canonical account state, with every fact carrying its source and age.
02
Control
Effective-dated deterministic rules evaluate readiness. Unknown stays unknown; a contradiction stays a contradiction.
03
Authorize
A named human releases material action, bound to the exact projection they reviewed.

Mandate's operator work queue, grouping accounts by what to resolve first, with each action showing the control, how many downstream controls it unblocks, its owner, and its deadline. — Ordering comes from a dependency graph in the domain layer, not a column sort. An action that unblocks four controls outranks an older one that unblocks nothing.Open at full size ↗

Open the queue

Architecture

Facts are recorded. Decisions are derived. People release.

01
Source data
Seven material feeds, each with an as-of time and a checksum
02
Canonical account state
Recorded facts and an append-only event log. Nothing derived is stored
03
Control engine
Versioned, effective-dated, dependency-aware
04
Readiness
Derived, never asserted
05
Resolution plan
Ordered by what each action unblocks
06
Authorized handoff
Checksummed, projection-bound, synthetic

Scope

What is built, and what is stood in for.

Built

Canonical, event-sourced account state
Deterministic control evaluation
Effective-dated, versioned rules
Event replay and deterministic reset
Dependency-aware resolution planning
Reproducible decision trace
Projection-bound approvals and invalidation
Checksummed handoff state

Simulated

Custodian and accounting feeds
OMS and execution venue
Tax system integration
External compliance tooling
Named approver identities

Implementation

Under the hood.

Canonical state: One normalized source of truth for the rebalance case. It stores facts and events, never readiness.
Projection engine: Readiness is re-derived from the current event state and an injected clock, never mutated by hand.
Effective-dated controls: Each rule carries a version, effective interval, scope, owner, source class, and named tests.
Append-only events: Every past decision stays reconstructable, and reset rebuilds an identical baseline.
Decision trace: Every material outcome links to its evidence, rule version, result, and the human who acted.

Boundary

What Mandate does not do.

Optimize portfolios
Route or execute trades
Replace the order management system
Act as a tax engine
Let AI override a control outcome

The boundary is the product decision. Models propose trades, controls evaluate readiness, authorized people release the handoff — and the handoff is a synthetic package, never an order.

Why it exists

A control problem, not an analytics problem.

Observation

Portfolio implementation fails less from a shortage of analytics than from fragmented state, inconsistent controls, and unclear ownership.

Product decision

Build an orchestration and control layer rather than another optimizer, and prove it on one account type deeply enough that the real failure modes appear.

Effective-dated controls: 12Effective-dated controls
States in the flagship path: 6States in the flagship path
Synthetic accounts evaluated: 11Synthetic accounts evaluated
Synthetic data: 100%Synthetic data

See the system run.

Mandate is a working synthetic portfolio-control prototype, built to show how an institutional rebalance workflow can be made explicit, traceable, and governable.

Launch Mandate View GitHub Back to all projects

Independent engineering case study. Every account, security, price, tax lot, policy, feed, and person in it is synthetic. Not affiliated with, endorsed by, or derived from any asset manager, custodian, index provider, broker, tax adviser, or trading venue. Thresholds are illustrative product assumptions, not any firm's policy. Nothing here is investment, legal, or tax advice, and the product never creates, routes, or executes an order.