Zator turns filings, contracts, markets, loans, narratives and workflows into executable economic state for models, agents, simulations, APIs and decisions.
The financial system is society's resource-allocation machine.
Zator makes its state legible, programmable, and reviewable.
Zator turns source streams into compiled economic objects, workflow systems, market replay, agent work, and cockpit surfaces that share the same state grammar. The visible surface is the command layer; source lineage stays attached underneath.
/products/sec-econ-gaap-api
Filings, macro releases, CRE records, accounting facts and replayable market state become product-shaped responses instead of one-off notebooks.
/library
Raw ideas, PDFs, chats, market traces, code and model outputs become reusable research objects that route into decks, investor notes, agent tasks and future workflow memory.
/platform/agent-substrate
Agents operate through durable state: they watch sources, route work, repair stalled lanes, update memory and leave the deployment with stronger rules and reusable workflow memory.
Observe source, model and workflow state
Normalize typed event → state object / workflow / memory
Route model, agent and control paths by capability
Repair stale assumptions, open exceptions and interrupted control paths
Inspect lineage, verifier, audit trail and replay path on demand
The same source-preserving grammar appears across cockpit panels, reports, APIs and agent workflows: source badges, provenance timelines, connector matrices, and command panes for work a team can replay.
lineage/preview/fact_0007421/source_span/p84/table3
lineage/preview/fact_0007421/source_span/p84/table3
projection: company-state.surface · stale policy: upstream revision
> compile source → object → projection
ok 6 source families normalized into canonical object lanes
path source span → drawer → typed object → lineage → projection
gate contracts lane requires a control check before projection mutation
scope example panels are illustrative; no external systems implied
Click an Atlas node, source lane, topology node, command or replay legend to see how source coverage, topology and command state move together.
projection access and replay path
coverage lane selected without external writes
topology selection remains a public preview transition
lineage/website-route-deck/preview/no-execution
before: route visible, control panes unchanged
after: terminal, review, Atlas and connector panes focused
read-only preview · no environment mutation
before: dependency waits for control posture
after: control surface receives gate and replay context
lineage/website-route-transition/preview/read-only
coverage/source lane stays inside the preview
drawer state and lineage policy
read-only preview · no environment mutation
after: control surface receives gate and replay context
This interactive topology shows source intake, canonical object forge, lineage spine, Atlas projection bus, and the control surface as one visible map behind Zator.
The replay surface keeps price state, source intensity, residuals, legend and lineage visible in one pane stack.
lineage/website-market-replay/preview/sample-feed
This interactive source map shows public-safe coverage lanes, evidence mass, source families and the lineage trail behind the panel.
generated_from_chronicle_source_coverage_registry
portedInvestor filings13F / 13D/G / N-PORT → factor crowding explorer
prototypePrivate markets / VCADV / portfolio pages → private markets spatula
touchedMacroeconomic seriesIMF / FRED / SDMX → FX topology
portedCentral bank communicationsspeeches / calendars → policy regime lanes
prototypeMarket microstructurevenues / books / trades → execution replay engine
portedAgent run tracesagent runs / model studies → research maproom
portedNews / web retrievalGDELT / CommonCrawl / RSS → deal intelligence lane
touchedCredit agreementscovenants / obligations → covenant graph
01load capability universe: search, filings, macro, bank-regulatory, private markets, execution, documents, partner systems
02read coverage evidence=54,781 candidates=27
03bind source mass cache truth_status=cached_registry_sample observed_at=2026-05-01T18:15:12.835588+00:00
04readiness counts ported:4, prototype:6, touched:3
05emit source coverage packet with capability universe and lineage trail
The command palette searches across source coverage, lineage lifecycle, cockpit windows, panel renderers and workflow intake without external side effects.
No preview command matches this query.
Zator turns notes, source packets, run outputs and agent traces into typed evidence, route constraints, reusable workflows and controlled next actions. The system does not ask every model to rediscover the same state from scratch.
PDF in, memo out. Data room in, deck out. Transcript in, bullet points out. Useful, but it skips the layer that makes financial work compound: the maintained workflow state behind the output — entities, claims, sources, periods, units, assumptions, dependencies, controls and monitoring rules.
10-K footnotes, trustee reports, loan tapes, remittance files, county assessor records, call reports, central-bank speeches, market prints, customer documents, vendor feeds.
Source-preserved memory of financial reality: typed claims, canonical entities, reconciled facts, dependency graph, review history, monitoring rules. The asset that makes outputs trustworthy and re-runnable.
Memos, comps, screens, decks, dashboards, alerts, monitors, regulatory packs, agent tasks, API payloads. Once the state exists, every output is a projection — regenerable, updatable, auditable.
A number can be true and unusable. “Revenue was $10B” is not a financial fact until it is bound to an entity, period, unit, source, concept and context. Every output Zator emits ships as a typed claim with its lineage open.
Prompt-centric AI does not survive production finance. Claim-centric systems do. A claim binds the value to its entity, period, unit, concept, source span, dependency graph and downstream uses. Auditability is structural, not optional.
a number can be true and unusable. financial facts are typed objects, not strings.
Most AI systems can answer a question. Zator knows when yesterday’s answer became wrong. Source updates propagate through the dependency graph and invalidate the projections they touched — with the reason attached and the next action queued.
A serious financial AI system needs source-update propagation as a first-class behavior. Restatement → stale fact. Amendment → stale covenant capacity. New filing → stale comp set. New macro print → stale forecast. Each one routes to the projection, the workflow, the reviewer and the next action.
a system owns the lifecycle of the answer.
The compiled state layer is the asset. The surfaces are the work product — built around your sources, your controls, and the workflows that already exist in your firm.
A focused first conversation. We map the workflow end-to-end and identify where AI fits, what is review-ready, and what infrastructure is required first.
Private agents over your own sources. Permission-aware, claim-citing, exception-routing. Built in your environment, against your controls.
Custom compilers for the forms your team works with: 10-K · 10-Q · ADV · 13F · NPORT · Form D · Y-9 · Call Reports · ABS-EE. Typed objects, calc-arc reconciliation, amendment chains.
Stale-dependency monitors over filings, holdings, contracts, language, ownership, macro. Typed events, deduped, with the diff in plain view.
Source-resolved facts, lineage-bearing claims, typed event streams. Used directly by quant, data, ops and downstream agent teams.
Browseable belief surfaces — debt walls, holdings deltas, covenant trips, IPO mechanics — with the evidence open. Built for customer-controlled deployment.
Maps a workflow into source inventory, control points, hidden dependencies and first deployable surface.
Six concrete shapes. Each is a workflow, not a category — with its inputs, its compiled state, and the projections that fall out the other side.
ABS-EE · 10-D · trustee reports · loan tapes · TRACE · ratings docs → loan→pool→tranche→trigger state → delinquency alerts, prepayment state, collateral drift.
Listings · parcels · assessor · recorder · zoning · debt · brokers · tenants · comps → property state → opportunity feed, valuation state, stale-comp alerts.
10-K · 10-Q · ADV · 13F · NPORT · Form D · Y-9 · Call Reports → typed claims → screens, APIs, regulatory packs, monitoring alerts.
Credit agreements · indentures · amendments · financials → clauses → baskets → triggers → covenant monitor, capacity model, amendment diff, distress alerts.
Filings · calls · notes · holdings · macro · news · price → thesis state → thesis tracker, memo regeneration, watchlist alerts, falsifier monitor.
Customer documents · vendor feeds · CRM · warehouse → permissioned retrieval → private agents, exception queues, reports, APIs.
A diagnostic taxonomy from working sessions: the demo passed, production did not, and each failure points back to the missing state layer.
No source span, no lineage, no review trail. The output cannot be checked, defended or improved.
Summary is not synthesis. Real work spans cross-references that no single document carries.
No dependency graph, no staleness signal. The answer rots silently and the desk doesn’t know.
Every agent restarts as an intern. Reviews aren’t captured. Scar tissue does not accumulate.
Clean data without context is dead data. Period, unit, segment, revision and source all matter.
The compiler preserves the original evidence, projects it into canonical objects, applies deterministic checks where structure exists, uses models where ambiguity remains, and maintains the resulting claims as versioned state.
Sources on the left, the workflow in the middle, the inspector on the right. Every step carries its evidence; every output ships with its lineage. This is what the surface looks like in practice.
A focused first conversation. A 2–4 week working prototype on real data. A deployed system you operate. Then we extend.
A focused session on the actual work: inputs, sources, controls, decisions, exceptions. We leave with a typed scope and a written workflow map.
Where the data lives, who can see what, what counts as “right.” Permissions and review points are designed in, not bolted on.
A real system on real data, end-to-end. Verifiable, lineage-bearing, exception-routable. Usable in week three.
Built for customer-controlled data, customer hardware and bring-your-own APIs. Client-licensed sources stay client-licensed.
We monitor the workflow, route exceptions, and harden the parts that pay off. Each pass turns probabilistic work into deterministic rules.
Verified work becomes a reusable module — composable into the next workflow without re-discovering the substrate underneath.
We’re running introductory conversations with financial teams applying AI to source-heavy workflows: filings, research, structured finance, CRE, credit, market monitoring and regulatory reporting.