Check us before you trust us.

Why is an AI tool the hard one to approve?

Because your name goes on it, and the vendor’s does not. You shouldn’t have to approve a system on the strength of a slide that says “enterprise-grade” — but most AI questionnaires come back as adjectives, and the honest ones come back as promises about a roadmap.

Then the works council asks whether it scores individuals, or a regulator asks what the model decided and why, and the answer has to be something better than a screenshot. So this page is built to be checked rather than believed: everything below can be verified from outside, most of it without talking to us at all.

Can I check a score instead of trusting it?

Yes — one click from the deal, without asking us. Runtime scores are stored with the input snapshot and rules needed for replay. Open one and the working is recomputed in front of you, so you are checking arithmetic rather than weighing an assurance.

What happens when a check fails matters more than what happens when it passes. The correctness certificate is verified whenever a score is served; if that check fails, the numeric score and its components are withheld. A one-click engine replay reports a mismatch as a failure and never turns it into a verified badge.

What will it refuse to answer?

Whatever the evidence does not support — and the list is on screen in the product, not in a whitepaper. This is the first thing worth pointing your evaluators at, because a revenue tool that scores everything is telling you about its standards rather than about its customer’s pipeline.

• A number with missing or untrustworthy inputs comes back as "we do not know", not as a middle value that reads on screen exactly like a real reading.
• The default publication policy withholds an overall score until three trusted signals are present. A workspace may retain the legacy one-signal policy or disable this breadth gate; the active threshold is recorded with the decision.
• Every refusal names what was missing, so your team knows which gap to close rather than being left with a shrug.
• The list of current refusals ships to every user in the default layout. It is not a premium tier and it cannot be bought out of.

Restraint is only worth anything when it costs something. The moment a vendor has the most to gain from inventing a number is exactly the moment a user is asking for one — which is why the rule lives in the code and the refusal is visible to the person who wanted the answer.

What if I think it got something wrong?

Attack it. Re-run a score, vary the inputs, and see what would have had to be different for the answer to change. A number that survives that is worth arguing from; one that does not should never have reached a forecast call — and finding that out in an evaluation is considerably cheaper than finding it out in an audit.

This is the part we most want a sceptic to use first. The demo is not a tour of dashboards. It is an invitation to try to break something in front of us.

What will an auditor see nine months from now?

A record built to be read by somebody who was not in the room. Decisions are stored with the evidence and the rules that produced them, so the question an auditor actually asks — why did it say this, on this date — has an answer that does not depend on anyone’s memory.

• Customer audit entries cannot be edited or deleted. That is enforced by the database, not by the code that writes to it.
• Recorded retention periods are minimum availability commitments, not permission to erase customer audit history.
• Each customer has their own linked record, so a missing or altered entry shows up rather than being deniable.
• Every published score is stored together with the evidence and the rules that produced it, so the record explains the decision instead of merely dating it.
• What SalesSynq refused to answer is kept too, with the reason — so "why is there no number here" has an answer nine months later.

Could a manager point this at one person?

No, and not by arrangement either. Individual scorecards, rankings and performance metrics are hard stops in application and database controls. Aggregate privacy is a separate, bounded workspace policy: suppression starts below five people by default, can be raised, and can be lowered no further than two.

• No scorecard for a named person

  Ask SalesSynq for a performance score or ranking on a named person and the request is refused outright, in separate application and database controls. There is no tier, contract, setting or support request that unlocks it.

• Group reporting starts at five by default

  Team and department performance rollups suppress groups smaller than five by default. A workspace administrator may raise that threshold or lower it no further than two. Individual rankings and individual-performance metrics remain prohibited regardless of that setting.

• AI off until you switch it on, one workflow at a time

  Each AI workflow stays dark until your team explicitly turns that specific one on. Not one global switch, and not on-by-default with an opt-out buried three screens into settings.

• Emotion inference blocked in storage

  Model-derived emotion, sentiment, tone and personality fields from workplace communications are rejected at the application and database storage boundaries, not merely discouraged by an acceptable-use policy. There is no workflow switch that enables them.

That is a worker-protection position, not a feature we are planning to add later. The capability was removed, rather than the temptation.

How do I actually review this?

In three steps, and you can do the first two before anyone books you a call.

1. 1

   Fetch the disclosure

   The control map is a public URL. No account, no NDA, no sales conversation. Read it, or hand it straight to whoever runs your AI reviews.

2. 2

   Try to break a score

   Bring your hardest question to a live score, re-run it, and vary the inputs. If something does not hold up, you will find out in twenty minutes rather than in month four.

3. 3

   Set the bounded policies

   Choose the documented publication modes your workspace needs. The product records the active values, enforces the group threshold’s hard minimum of two, and keeps named-person scoring blocked.

Done that way, this is an approval you make once. The disclosure answers the questionnaire, the refusal list answers the works council, and when someone asks eighteen months later why a number said what it said, you open it instead of opening an investigation.

What do you publish, and what do you hold back?

Controls and method are published; our infrastructure and other customers’ data are not. That boundary is deliberate, and generated certification-fixture scores are not presented as predictive performance.

• Security

  Session protections, keeping customers separated from each other, privacy request handling, vulnerability and dependency scanning, and central error capture. These are part of what has to be true before we ship, not things we get to later.

• Reliability

  Releases go out behind health checks, with production monitoring, encrypted backups, and restore procedures that have actually been run rather than merely written down. We describe the controls; we do not publish our internal layout.

• How the scoring is judged

  A public result must come from independently labelled evidence with its method and limits attached. The current generated certification fixture does not meet that bar, so SalesSynq publishes no performance result for itself.

What is shown publicly:

• How the scoring works at a level you can evaluate, and what constrains it.
• The controls that bind every part of the product, and what they are anchored to.
• The machine-readable AI disclosure, fetchable by anyone, without an account.

What is never shown publicly:

• Customer names, raw identifiers, and anything specific to another customer.
• Our internal service layout, hostnames, providers and storage paths.
• Secrets, and the level of detail about a control that would help someone get around it.
• Numbers we have not measured — accuracy figures and comparisons against other customers — which appear on no product or marketing page here. The one availability figure we publish is a contractual commitment in the Terms, not a measured result.