How Mandates Shape SI Behavior
The injection pipeline and precedence hierarchy that governs every interaction
What It Is
When you create a mandate, the output is a structured JSON schema — a set of machine-readable instructions. But a schema sitting in a database does nothing on its own. The power of mandates comes from how Zyntro *injects* them into SI's decision-making at the moment of action.
Every time SI generates an email, processes a reply, conducts a Phona call, responds in live chat, or sends an SMS, it loads all active mandates that match the current context — the contact's status, their audience segment, the pipeline stage they are in, the ware being discussed, and the channel being used. SI reads these mandates alongside the brand definition and Reusable Knowledge Base, then generates communication that respects all of them.
This is not a suggestion system. Mandates are directives. SI does not weigh them against its own judgment and decide which ones to follow. If an active mandate says "max 2 emails per week," SI sends no more than 2 emails per week in that context. Period.
Why It Matters
The injection model means mandates are not just configuration — they are live, enforceable rules operating across every channel simultaneously.
This matters because consistency is hard. A sales team might know that enterprise prospects should be contacted no more than twice per week. But when SI operates across email, SMS, and phone, that rule needs to be enforced automatically — not remembered by a human checking a calendar.
It also matters because *different* mandates can apply to *different* aspects of the same interaction. When SI writes an email to a nurturing-stage realtor about your Pro plan, it might be reading:
- A brand-level Comms Flow mandate (overall tone)
- A ware-level Hard Facts mandate (what the Pro plan includes)
- An audience-level Comms Flow mandate (realtor-specific language)
- A stage-level Cadence mandate (nurturing follow-up rhythm)
- A channel-level Prohibition mandate (email-specific restrictions)
All of these are loaded simultaneously. SI synthesizes them into a single, coherent output.
How It Works
The Injection Pipeline
When SI is about to take an action for a contact, it runs through this sequence:
1. Context assembly — Identify the contact's status, audience segment, pipeline stage, the ware being discussed, and the communication channel.
2. Mandate loading — Query all active mandates that match any of these context dimensions. A mandate matches if its attachment level corresponds to the current context (e.g., a ware-level mandate matches if the ware being discussed is the one it is attached to).
3. Precedence sorting — Sort loaded mandates by their behavior priority: foundational < overlay < execution_specific < safety_net.
4. Conflict resolution — When two mandates give contradictory instructions, the higher-priority mandate wins. Within the same priority level, more specific attachments override more general ones (e.g., a stage-level mandate overrides a phase-level mandate).
5. Schema injection — The resolved set of mandate instructions is injected into SI's prompt context alongside the brand definition and relevant RK items.
6. Generation — SI produces the communication, respecting all injected mandates.
7. Validation — For critical prohibitions and hard facts, SI checks the generated output against the mandate rules before sending. If a violation is detected, SI rewrites the output or escalates.
Where Mandates Get Injected
- Email generation — Mandates are loaded before SI composes any outbound email
- Reply processing — When a contact replies, mandates shape how SI interprets the reply and crafts a response
- Phona voice calls — Mandates are loaded into the call context so the AI follows them during live conversation in real time
- Live chat — Mandate rules are present throughout the entire chat session
- SMS — Mandates govern tone, content, and frequency of text messages
Precedence Hierarchy
| Priority Level | Attachment Types | Behavior | Can Be Overridden By |
|---|---|---|---|
| Safety Net (highest) | Contact Status | Hard limits on what SI can do for contacts in specific statuses (DNC, churned, unverified) | Nothing — this is the final word |
| Execution-Specific | Pipeline Stage, Communication Channel | Concrete, operational rules for specific stages or channels | Contact Status only |
| Overlay | Ware, Audience Segment, Pipeline Phase, Product Category | Specialization rules that layer on top of the brand foundation | Contact Status, Pipeline Stage, Channel |
| Foundational (lowest) | Brand | Global behavioral doctrine — the baseline for all communication | Everything more specific |
Examples
A Contact Status mandate prevents outreach despite a Cadence mandate
A Cadence mandate on the Nurturing stage says "email once per week." But the contact has been set to DNC status. The Contact Status mandate (safety_net priority) says "no outbound communication of any kind for DNC contacts." The Contact Status mandate wins. SI does not send the email, regardless of the cadence rules.
Multiple Comms Flow mandates layer together
A brand-level Comms Flow mandate says "professional and warm." An audience-level mandate for realtors adds "use real estate terminology and reference market trends." A channel-level mandate for SMS adds "keep messages under 160 characters, no formal greetings." SI combines all three: professional-warm + real-estate language + SMS-length constraint. The most specific (channel) overrides the general (brand) where they conflict — so no formal greeting in SMS, even though the brand mandate prefers warm openers.
A Hard Fact overrides an Objection Handling response
An Objection Handling mandate suggests responding to pricing concerns by highlighting a money-back guarantee. But a Hard Fact mandate states: "We offer a 14-day trial, not a money-back guarantee." The Hard Fact wins. SI adjusts the objection response to reference the trial offer instead of a guarantee that does not exist.
