Head-to-head · updated 13 September 2026

DataShield vs Semarchy: who governs the golden record once an agent asks for it?

Semarchy is a real MDM platform. It has fifteen years of work behind it. Gartner made them a Leader in the 2026 MDM Magic Quadrant. Their model-driven builder does the work. It turns a data model into a working steward app. Most MDM vendors bill a system integrator for that instead. They ship inside Snowflake and Microsoft Fabric too. Their homepage claims 190+ partners. Their partners page names three: Microsoft, Snowflake and AWS. Plus a couple dozen system integrators. Either way, their bench is bigger than ours.

We build MDM agent first. An agent can stand the model up with you. It talks with a steward. It builds a spec from that. Then it walks a decision gate you validate. Then it builds it all. Entity types, hierarchies, draft match configs, reference bindings, rules. All in one batch. It promotes nothing right away. The matching under it names its own maths. Fellegi-Sunter linkage. Jaro-Winkler and Double Metaphone comparators. LSH blocking. EM-trained parameters. Five survivorship modes. A golden record reaches an agent only after it clears a score gate. Not before. A consent veto guards that gate too. Masking happens in the view, not a setting. We tokenize datasets at ingest. Agents query tokenized data over MCP. Turning a token back is a privileged, audited step. Here is the honest split, wins for them included.

Get free Demo Center access Get an instant quote

The short version

Pick DataShield when

An agent should help build it. Not a services team. mdm_project talks with a steward. It works a decision gate you validate. Then it builds entity types, hierarchies, draft match configs, and rules. All in one batch. Fifteen commands. No statement of work. See how Ontology models it.
You want to see the match maths. Fellegi-Sunter weights. EM-trained parameters. A drift score on every cleared config. No config ships until a frozen test run goes green. Ask us to explain a merge. You get numbers, not a shrug.
Agents, not just people, will read your golden records. Ours have to earn it. Six quality axes. A composite floor. A consent check that fails closed. Then the read goes out over MCP, under a token with a scope ceiling. The call gets checked before it runs. How Auth does it.
You want a price before a call. A stack on your own hardware. A steward log you can verify yourself. Pricing is published and the verifier is public.

Pick Semarchy when

Business stewards need an app, not a queue. Their model-driven builder builds one. We do not have that. It matters.
You want prebuilt domain models on day one. Their accelerator marketplace ships customer, product and vendor models. We hand you an open vocabulary instead. Plus an agent to help fill it. That's a different kind of head start.
Snowflake or Microsoft Fabric is your estate. A native app inside the warehouse beats any outside tool on plumbing.
You need a large partner bench. Plus a Gartner Leader badge. We are not on that list. We will not pretend otherwise.

Bottom line: Semarchy sells a bigger suite. A better business-user surface. A bench to staff the project. We sell mastering an agent can build with you. You can query it under a scope ceiling. An auditor can check it. Are your golden records about to feed agents? Ask both of us what happens on the call itself.

Feature by feature

Competitor cells describe what Semarchy's public site says. And its press releases, as of the date above. Did we get something wrong? Email support@myorg.ai. We will correct it, credited.

What matters	DataShield	Semarchy	Edge
Standing up the model	An agent does it with you. `mdm_project` talks with a steward. It builds a spec from that. Then it works a gate, in order. Each idea says what it is, why, its impact, and the choices. Override one, and each dependent choice reopens. One batch then makes entity types, hierarchies, match configs, and rules. Each object lands draft, gate-honest. The generator never clears a config. It never skips the test battery. Fifteen commands. No setup services.	A model-driven designer. Plus Copilot AI. It turns prompts into data products. It works with VS Code, Git and CI/CD. Driving the model is your job, or a partner's.	◆
Entity resolution method	Fellegi-Sunter probabilistic linkage. Jaro-Winkler, Levenshtein, Soundex, and Double Metaphone comparators. Plus LSH blocking. EM-trained parameters run against a frozen holdout. A trained flip in polarity is a hard refusal. Five survivorship modes exist. The most trusted one reads overturn rates from steward work. Not a hand-ranked list. Drift monitoring runs on every cleared config. It proposes; it never acts alone. An identifier tier routes conflicts to review, not to a merge. No config goes live without a green test run.	Match and merge with survivorship rules. Plus Snowflake Cortex AI. It runs right there. For matching, enrichment, and validation. The scoring model behind it stays unpublished.	◆
Ingestion and binding	Six on-ramps. Inline, URL, and batch ingest. Watch sources. A file repo. A catalog bind. The catalog bind reads a live PostgreSQL source. No rows leave it. Bind a table to an entity type once. Mastering then stays current on a set schedule. Remaster policies can be auto, agent review, or steward review. Differential loads mark deleted rows stale. They flag updated ones for a re-check. A weekly full pass catches any missed deletes. That is PostgreSQL today, not your whole estate. We are not an ETL product. We will not sell you one.	A real integration lineage with broad connectivity. Now built right into Snowflake and Microsoft Fabric. Plus Git and CI/CD workflows for data engineers.	◇
Multidomain modelling	An open entity-type vocabulary. Three matching modes per type. A People, Places, and Things taxonomy is just a table row. Not a fixed list. Each hierarchy stays inside its parent's scope. A child resolves inside its parent's scope at a base level. Across parents, it needs a stricter one. Relationships carry bitemporal validity. Domains carry an owner and a compliance framework. Rules gate golden publication. Onboarding runs end to end. Interview, gate, one-shot build. Every object draft. What we do not ship is a packaged Customer 360.	Customer, product, vendor, and more. An accelerator marketplace of prebuilt models arrived in February 2026. Plus application data management alongside.	◇
Stewardship experience	Review queues cover batch decide, reopen, assign, escalate, and undo. A dashboard adds more. A flow canvas. A per-gate workbench. A records view. An alerts queue. A dry run tests the writer's own gate code first. A preview can never say yes, then refuse later. It is a workbench, not a generated business app.	Their strongest card. The model-driven builder makes apps. And steward screens, straight from the model. Business users live in it.	◇
Stewardship trust ladder	Stewards dispose patterns, not items. Each idea carries a counterevidence scan. It argues against itself, on purpose. A three-rung trust ladder counts human approvals, not disposed rows. 50 approvals at a 98% rate clears a pattern. Just one dissent demotes it. Zero dissents required. One dissent demotes it, right away. At the top rung, a sampler steps in. It diverts 10% of auto-disposed items to a human audit. You can show why we picked any item. Standing rules start staged. That needs a second release choice. They auto-suspend on an audit overturn. Steward groups carry routing and escalation. Each action can be undone.	Workflow-driven steward work with assignment and approval steps. We found no published trust ladder, audit sampling rate, or suspension rule. Worth asking what earns automation on their side.	◆
Agent-admission gate	A golden record must earn its read, before any agent gets it. Six score axes carry set weights and floors. It needs a combined score of 0.70 or better. Each axis must clear its own floor too. A consent veto fails closed. Consent covers all six GDPR Article 6(1) bases, across nine purposes. No recorded basis means no permission. One rule enforces the gate. It covers both the SQL path and the app path. The two cannot drift apart.	Certification and badges on data products. Semarchy serves context to Cortex agents. We found no score floor. No consent check stops an agent from reading a record either.	◆
Audit evidence	A SHA-256 hash chain with Ed25519-signed checkpoints. Those checkpoints chain too. Verification names the break: tampering, insertion, deletion, or truncation. Each change gets a before and after record, in the same transaction. Try the verifier.	Their docs describes an immutable audit trail. We found no cryptographic tamper evidence. No independent verifier either.	◆
Agent authorization	Each checked tool call passes a scope gate. Plus an access tier, and a fresh revoke check before dispatch. We track usage before the handler runs. The call fails closed.	MCP endpoints give checked data products to Cortex AI agents. Access control is platform-level. We found no per-call check in public docs.	◆
Break-glass	Scoped, time-boxed emergency access for agents, IP-allowlisted and step-up gated. It auto-revokes and cannot be quietly removed from the log.	Not described in their public docs.	◆
GDPR erasure	Crypto-shred of per-subject key material. Plus ISO 27560 consent receipts, and a consent-validity record per subject and purpose. Golden version history is append-only. It survives deletion. A tombstone records the reason and the successor. The audit chain still verifies afterwards.	Oversight and steward workflow. How they erase a mastered record is not public. Worth asking.	◆
Tokenization and masking	Deterministic, join-preserving, vault-reversible tokens at ingest. Plus quasi-identifier generalization. Dates become year, decade, or age band. ZIPs become 3 or 4 digits. Phones, SSNs, and emails get partial masking. Each column gets its own cardinality-reduction score. The golden views mask in the view. And in the reader role. Not a setting. The reader has no grant on the base table. Derived phonetic and n-gram fields inherit their base field's PII verdict. A tokenized matching field makes a run refuse, not merge blindly.	Classification and badges on data products. Tokenization is not vocabulary they use.	◆
MCP and agents	More than 200 MCP tools across Ontology, Auth, Corpus, and Lighthouse. MCP tool tokens carry scope ceilings. Per-call tracking ties back to the agent. Mastering alone is 143 commands: 99 on mdm, 29 on stewardship, 15 on guided projects. That's about a third of the whole agent surface.	MCP endpoints shipped in the Snowflake Connected App in June 2026. They serve cleared golden records with context to Cortex agents. Their docs do not say if that surface exists outside Snowflake.	◆
Golden record consumption	Masked views. A golden read relation tells you what it excluded. And why. Under your own filters. A value-free CloudEvents outbox carries 21 event types, commit-ordered. Golden-to-source lineage runs over typed edges. A DuckDB dimension lets you join against Parquet. We build RAG grounding documents only from agent-ready records. What we do not have is outbound publishing into your target systems. That is on the roadmap. We will not pretend it ships.	Their whole current pitch. Checked data products, given to consumers. MCP endpoints serve cleared golden records to Cortex AI agents, inside Snowflake.	◈
Deployment	Self-hosted, in your own cloud or data center. Or a dedicated single-tenant server we run. Docker images ship for Auth, Ontology, Corpus, and Lighthouse. Guardian verifies a signed deploy manifest. Ed25519 audit-signing keys stay on your instance. In a key directory you control. Key rotation fails closed on an unknown key. KMS or HSM custody is a built seam in Auth. It is not a shipped integration. Tokenization keys sit in your environment today, not in a KMS.	SaaS, Snowflake native app, Microsoft Fabric, and self-hosted since February 2026. They promise no feature gap between them.	◈
Maturity signals	Auth, Guardian, and Lighthouse are live in production. They've shipped several releases since April 2026. We test mastering against about 370,000 real public records. They span three City of Chicago datasets, not just synthetic ones. We publish no record-count ceiling. We will not invent one. SOC 2 is not yet certified. We say so. No analyst quadrant either.	Founded 2011. Gartner MQ Leader for MDM in April 2026. Also 4.8 out of 5 on Peer Insights. Across 39 reviews. Six years as a Customers' Choice. They also claim a trillion-plus consolidated records. That's their number, not an audited one.	◇
Pricing	Published model. A scoped instant quote. No sales wall.	Quote only. No tiers, no free tier, no open-source edition. A Forrester study they paid for reports a 315% three-year ROI. That's their study. It tells you about payback, not price.	◆

◆ DataShield leads◇ Semarchy leads◈ comparable

Semarchy claims come from semarchy.com and Semarchy's own press releases. Last checked 13 September 2026. We link them below rather than work from memory.

Three things you get here that you won't get from an MDM suite

A model an agent helps you build

Most MDM projects start with a workshop. And a statement of work. Ours starts with a conversation. An agent asks a steward what they master, and why. It builds a spec. Then it walks a decision gate, one proposal at a time. Change your mind, and the dependent decisions reopen. Say yes, and one batch makes it all. Entity types, hierarchies, match configs, rules. All draft. Later, a steward asks why two records merged. You get the match weight, per field. The values behind it, too. You get the drift score since we cleared that config. See how Ontology models it.

Golden records that earn their readers

A score is usually a dashboard. Here it is a gate. Six axes carry set weights, plus a floor. A consent check fails closed. It covers all nine purposes. Miss a floor, and the record never reaches an agent. The same rule runs in SQL and in the app path. Nobody can route around it. How the boundary is built.

An erasure you can defend

GDPR says delete the subject. Your auditor says keep the steward log. Crypto-shred settles it. We destroy the subject's key material. The data goes unreadable. The chain still verifies. See the diagram.

Where Semarchy is genuinely stronger

Start with the thing we cannot do. Semarchy's builder takes a data model. It builds a working app for the people who steward the data. Screens, workflow, the lot. That's the difference. A project your users adopt, versus one they quietly route around. It's why their Peer Insights scores stay high, year after year. Add prebuilt models for customer, product, and vendor. Add delivery inside Snowflake and Microsoft Fabric. Then a 2026 Gartner MQ Leader badge. Add a partner bench they put at 190+. Their partners page names Microsoft, Snowflake, and AWS as strategic partners. Plus a couple dozen system integrators. Are you mastering six domains across a big enterprise? That package is hard to beat. We would say so on a call.

Now the push-back. Their pitch to a buyer: a model, plus a bench to build it. Ours is an agent that builds it with you. An interview. A decision gate you validate. Then made entity types, hierarchies, match configs, and rules, all in one batch. We clear nothing until a test run goes green. The second half: what happens when an agent asks for a golden record. Their June 2026 launch says MCP endpoints give cleared records to Cortex agents. With semantic context. Good. But three questions their material does not answer. What happens on the call? Say the agent's access got pulled 90 seconds ago. Can you prove it to an examiner? That the record served was the record logged. What does the log do when the subject exercises Article 17?

We answer with three things. A quick re-check on revoke. A signed hash chain. A crypto-shred that leaves the chain intact. Semantic context makes an agent useful. Those three answers make it defensible.

Questions worth asking both of us

These are the questions we would want answered. If we were the ones buying. Ask them on each call, ours included.

Can you cryptographically prove an audit log entry wasn't deleted?

DataShield: yes. Each record commits to the one before it. We sign and chain each checkpoint. Verification tells deletion apart from truncation and from tampering. Try it on a sample chain at /verify. Semarchy describes an immutable audit trail. We found no published tamper-evidence check. No verifier a third party can run either. Ask them to show one.

What happens to a revoked agent mid-session?

DataShield re-checks access on each checked tool call. The pull lands on the very next call. Say an analyst leaves on a Friday. Their agent is 20 minutes into a 40-minute job. The next call fails closed. Semarchy's MCP endpoints serve checked data products to Cortex agents. We could not find a mid-session pull check in their public docs. Ask how long a stale agent keeps reading your golden records. Even after you pull its access.

How does GDPR erasure interact with the audit trail?

DataShield crypto-shreds the subject's key material. It issues an ISO 27560 consent receipt too. Actor identities in the chain carry HMAC commitments, not names. You can still check the evidence. Even once the subject is gone. We also hold a consent-validity record per subject and purpose. No recorded basis counts as no permission, not a yes. Semarchy's public docs covers steward workflow. They do not spell out how they erase a mastered subject. Neither is the effect on the history. Ask for the method, not the workflow.

Do we need a setup partner to stand this up?

That's the question we would push hardest on both calls. Semarchy's answer is a big bench and an accelerator marketplace. For a six-domain enterprise programme, that's a fair answer. Ours is an agent. It talks with a steward, builds a spec, and works a gate. Each idea says what it is, why, its impact, and the choices. Then it makes the model in one batch, every object draft. Fifteen commands drive that. Mastering has 143 in total. It is real MDM underneath. Probabilistic linkage, blocking, trainable match configs. Hierarchies, reference data, rules, five survivorship modes. What it is not is a low-code app factory for business screens.

Semarchy has MCP endpoints now. What's different about yours?

Theirs came with the Snowflake Connected App in June 2026. It serves cleared golden records with context to Cortex agents. That's a good delivery story inside one warehouse. Ours is the native access path for the whole platform. The difference is what wraps the call. A tool token carries a scope ceiling. An access check runs before dispatch. A quality and consent gate must clear first. Tracking ties to the agent. We seal an audit row into the chain. Context tells an agent what a field means. A scope ceiling decides whether it gets the field at all. More on that in our Snowflake MCP authorization post.

Does DataShield have SOC 2?

Not yet, and we will not imply otherwise. Guardian and Lighthouse have been in production since April 2026. The threat model is public. Anyone can run the verifier. Design-partner terms include source escrow. That way a small vendor is not a single point of failure. We could not find a SOC 2 statement in Semarchy's public docs. No ISO 27001 either. Ask them too. Details on our security page.

Semarchy's current positioning: "Context Changes Everything," with "self-service Intelligent Data Products" for "the agentic enterprise." The site also claims a 315% ROI. Over three years. Plus 1 trillion-plus consolidated master records. Plus 190+ implementation and technology partners. The partners page names Microsoft, Snowflake and AWS as strategic technology partners. Plus a couple of dozen system integrators. - semarchy.com, 13 Sep 2026
Named a Leader in the 2026 Gartner Magic Quadrant for Master Data Management. Also 4.8 out of 5 on Gartner Peer Insights. Across 39 reviews, as of 31 March 2026. - Semarchy press release, 9 Apr 2026
Snowflake Connected App: "Deliver governed data products to Cortex AI agents through MCP endpoints with certified golden records and semantic context." Cortex AI runs right there. For matching, enrichment, and validation. - Semarchy press release, 2 Jun 2026
Self-hosted and customer-managed deployment came alongside SaaS and Snowflake. Plus an accelerator marketplace of prebuilt customer, product and vendor models. - Semarchy press release, 10 Feb 2026
Semarchy Data Platform launch: "the first agentic DataOps experience." Copilot AI turns prompts into data products. It works with VS Code, Git and CI/CD workflows. - Semarchy press release, 9 Sep 2025
CTO demo of "agentic access to data using Semarchy's Data Product MCP capabilities" at Dataversity MDM Demo Day. - Semarchy press release, 20 May 2026

Other head-to-heads

MDM

Get free Demo Center access

You've seen the proof

Ready for a number? Scope your deployment and we'll price it against your own economics.

Get your quote →