SME Access

Enter your password to view the onboarding materials.

SME Onboarding

Welcome, Subject Matter Expert

You've been assigned as a domain expert for the Synapse Design Partner program. As an SME, you'll work across two platforms — Synapse Prompt Playground and Testiment — to drive product improvements. This page covers your role, tools, and how you'll work with Design Partners.

Watch First
Onboarding Videos
Getting Started
Your Onboarding Checklist

Complete these steps to be fully set up as an SME.

Access to Synapse Playground
You'll receive an invite to the Synapse Prompt Playground. This gives you access to view and improve all prompts used on the platform — on a per-venture level or globally.
Access to Ventures from Design Partners
Each Design Partner will invite you to their venture. Once added, you'll see their data, session recordings, and feature usage in your dashboard.
Access to Testiment
You'll receive an invitation to testiment.io where you'll review session recordings, analyze user behavior, and generate structured fix documents.
Your Role
Your Role as an SME

You're the bridge between raw user feedback and actionable product improvements. You never attend DP sessions directly — you review the raw data and write the improvement documents. You'll work across two platforms:

1. Synapse Prompt Playground

View all prompts used on the platform and improve them where needed — on a per-venture level or globally. This gives you complete flexibility to improve the quality of results from research, intelligence, and how the agents currently run.

2. Testiment

Review session recordings, analyze user behavior with AI, and generate structured fix documents. This is where all DP session data lives and where you create actionable improvements.

  • Review raw session recordings, transcripts, and logs from Design Partner test sessions
  • Generate structured fix documents that translate friction into actionable improvements
  • Improve prompts via the Synapse Prompt Playground — per venture or globally — to enhance research, intelligence, and agent quality
  • Own specific features and be accountable for their improvement arcs across sprints
  • Review AI-drafted fixes, refine with your domain expertise, and submit for core team approval
  • Coordinate retests after fixes ship to confirm the improvement landed
  • Never attend DP sessions — this ensures unbiased, independent analysis
Time Investment
Time Commitment

Your time is split between reviewing sessions, generating fixes, and coordinating with the core team.

2–4
Hours per week
1–2
Fixes per week
~30
Min per session review
Rhythm
Weekly Cadence with Design Partners

Your work begins after DP sessions wrap up. Here's the weekly flow.

Mon – Thu

DP Sessions Run

Design Partners complete their scheduled test sessions. You're not involved during this phase — sessions are run independently.

Thursday

Session Data Delivered

Recordings, transcripts, bug reports, and NPS results are compiled in Testiment and handed off to you for review.

Fri – Mon

SME Review Window

Watch recordings, read transcripts, analyze logs. Identify patterns, friction points, and opportunities. Draft your fixes.

Tuesday

Fix Submission

Submit completed fix documents with evidence, root cause analysis, and proposed changes. Core team reviews on Wednesday.

Tooling
Synapse Prompt Playground

The Prompt Playground lets you view all prompts used across the Synapse platform and refine them to improve the quality of results. You can edit prompts on a per-venture level or globally, giving you complete flexibility over research output, intelligence quality, and agent behaviour.

🔎 View All Prompts

Browse the complete library of prompts used across the platform. See exactly what instructions drive each agent, research task, and intelligence output so you know where improvements are needed.

Per-Venture Editing

Customise prompts for individual ventures. If a specific Design Partner's venture needs tailored research instructions or adjusted agent behaviour, you can override the defaults at the venture level.

🌐 Global Editing

Apply prompt improvements across the entire platform. When you identify a systemic quality issue — like research depth or output formatting — a global edit fixes it everywhere at once.

Impact on Quality

Your prompt changes directly improve the quality of research, intelligence outputs, and how agents run. This is the most direct lever SMEs have to improve the results Design Partners experience.

Tooling
Testiment — Your Review Platform

Testiment (testiment.io) is the platform where all DP session data lives. It's your primary workspace for reviewing sessions, analyzing feedback with AI, and generating fixes.

🔑 Get Access

You'll receive an invitation to Testiment via email. Accept the invite to create your account. If you haven't received it, contact your program lead. Once in, you have full access to all test data for your assigned features.

📋 View Tests & Sessions

Browse all tests and their recorded sessions. Each session includes the full video recording, screen capture, audio transcript, developer logs (console + network), NPS scores, and debrief answers. AI-generated summaries give you the key findings at a glance before you dive into raw data.

  • Session recordings with timestamped events
  • AI summaries of each session's key findings
  • Bug reports with full context captures
  • NPS scores and trend data per DP
  • Think Aloud transcripts with word-level timestamps
  • Developer logs: console errors, network failures, action steps

🤖 Chat with AI on Each Session

Every test and session has an AI assistant you can chat with. Ask it to highlight friction points, compare sessions across DPs, identify patterns, or draft sections of your fix document. AI acts as your copilot — you bring the domain expertise, it handles the data crunching.

  • "What were the top 3 friction points in this session?"
  • "Compare NPS trends for this feature across all DPs"
  • "Draft an issue description based on the transcript at 2:34"
  • "What console errors occurred during task 3?"

🛠 Generate Fixes

After reviewing a session, use Testiment to generate a fix proposal. The AI analyzes the session data, your notes, and the codebase context to draft a concrete fix. Generated fixes are submitted to the Synapse core team for approval — once approved, they go into the improvement pipeline.

  • AI drafts the fix based on session evidence
  • You review and refine before submitting
  • Synapse core team approves or requests changes
  • Approved fixes enter the sprint and get built
Visit Testiment →
Fix Deep Dive
How to Generate Fixes

Testiment uses a structured template system to capture every finding. When you review a session and identify an issue, AI drafts a fix document from the session evidence. You refine it, and the core team approves it. Every fix follows one of these templates.

The Template System

There are two categories of fix documents. Use ICD for bugs and broken functionality. Use A1–A16 for feedback-area findings (quality, UX, trust, etc.). AI selects the right template automatically based on the session data.

Bug / Issue Template

ICD
Issue Context Document
Something is broken, throws an error, or a workflow fails. Includes reproduction steps, console errors, network failures, and a proposed code fix.

Feedback Area Templates (A1 – A16)

A1
User Experience — Navigation, cognitive load, hierarchy
A2
Output Quality — Accuracy, relevance, completeness
A3
Data Presentation — Charts, tables, info density
A4
Agent Interaction — Directives, HITL friction, trust
A5
Feature Value — ROI, JTBD fit, willingness to pay
A6
Time Saved — Speed vs manual, efficiency delta
A7
Methodology — Prompt structure, step sequencing
A8
Reliability & Trust — Consistency, confidence in outputs
A9
Onboarding — Time-to-value, discoverability
A10
Integration Fit — Workflow compatibility, data I/O
A11
Stage Appropriateness — Right feature at right stage
A12
Missing Capability — Gaps vs JTBD not served
A13
Performance — Load times, latency, responsiveness
A14
Accessibility — WCAG, assistive tech, inclusive design
A15
Content & Copy — Labels, errors, microcopy, terminology
A16
Security & Privacy — Data handling concerns, trust barriers

Document ID Format

Every fix document gets a unique ID auto-generated by Testiment. The format encodes the template type, year, and sequence number.

{TYPE}-{YEAR}-{SEQ}
ICD-2026-0342
Bug report #342
A1-2026-0015
UX finding #15
A2-2026-0087
Output quality #87
A8-2026-0023
Reliability finding #23

Step-by-Step Process

  1. Review the Session Data

    Open the session in Testiment. Watch the recording, read the AI-generated summary, check console logs and network activity. Note specific friction points with timestamps and exact quotes.

  2. Chat with AI to Dig Deeper

    Use the AI copilot to analyze patterns: “What friction points appeared across all sessions for this feature?” or “Compare NPS trends for F2 across all DPs.” Let AI do the data crunching while you bring domain expertise.

  3. Identify the Core Issue

    Isolate the specific problem. Be precise: “Canvas fails to load when venture has no competitors” beats “Canvas doesn’t work sometimes.” Classify whether it’s a bug (ICD) or a feedback area finding (A1–A16).

  4. Generate the Fix

    Click “Generate Fix” in Testiment. AI analyzes the session evidence, selects the right template, and drafts the full document — including observation, evidence, root cause, and proposed fix. You get a complete first draft in seconds.

  5. Review & Refine

    Read the AI-drafted fix carefully. Add your domain expertise: refine the root cause analysis, adjust the proposed fix strategy, correct any inaccuracies. The AI is your starting point, not the final word.

  6. Set Priority & Submit

    Assign a priority (P0–P3) based on the framework below. Submit the fix document. The Synapse core team reviews it, approves or requests changes, and approved fixes enter the sprint pipeline.

Template: ICD — Bug / Issue Report

Use when something is broken, throws an error, or a workflow fails. This template captures the full technical picture.

Meta

Document ID
Auto-generated: ICD-YYYY-NNNN
Title
One-line description of the bug
e.g., “Canvas fails to load when venture has no competitors”
Severity
P0 / P1 / P2 / P3
Status
Pending Review → Approved → In Progress → Resolved
Source
Bug Report / Usability Test / Agent Scan
Session ID
Links to the session where the bug was captured

01: What Happened

Narrative
Clear description of the bug in 2–3 sentences
AI writes this from session data. Includes what the user did, what happened, and what should have happened.
Reproduction
Numbered steps with timestamps and screenshots
Step-by-step path to reproduce: action → timestamp → screenshot reference
Expected vs Actual
Side-by-side comparison of expected and actual behaviour

02: Technical Evidence

Console Errors
Captured error messages with source file and line number
e.g., TypeError: Cannot read properties of undefined — canvas-summary.tsx:142
Network Requests
Failed or problematic API calls with method, URL, status, and payload
Browser / Env
Browser version, OS, viewport size, session recording link

03: User Sentiment

NPS Score
Rating and reason from the DP
Debrief Answers
Responses to post-session debrief questions
Think-Aloud
Transcript excerpt of the DP's real-time commentary
e.g., “It just broke. I assumed the product wasn’t ready.”

04: Root Cause Analysis

Root Cause
Plain-language explanation of what’s causing the bug
AI analyzes the codebase after reviewing session evidence
Primary Location
Repository, file path, and line numbers of the problematic code
Related Files
Other files that need changes, with explanation of why

05: Impact & 06: Proposed Fix

Impact
Users affected, revenue impact, frequency of occurrence
Required Changes
Numbered list of file-level changes with descriptions
e.g., “ventures-controller.ts:89 — Return empty array instead of null”
Effort & Risk
Small / Medium / Large · Low / Medium / High risk

07: Review & Approval

Reviewer
Core team member who reviews the fix
Decision
Approve Fix / Approve with Modifications / Request More Investigation / Reject
Priority
Hotfix (deploy immediately) / Next sprint / Backlog

Template: A1 — User Experience Finding

Use when a feature works but the experience is poor — users get lost, confused, or can’t find what they need.

Meta

Document ID
A1-YYYY-NNNN
Source
Usability Test / Think-Aloud / Expert Review / Agent Observation
Feature / Page
Which feature and page this relates to

01: Observation + UX Dimension

Observation
What happened from the user’s perspective, 2–4 sentences
UX Dimension
Navigation / Cognitive Load / Visual Hierarchy / Information Architecture / Feedback & Affordance / Consistency / Error Recovery
Select one primary dimension, optionally one secondary

02: Evidence

Session Data
Task completion rate, average time on task, misclick rate, backtrack count
User Quotes
Direct quotes from think-aloud and debrief
e.g., “I feel like I’m staring at an empty filing cabinet.”
NPS Impact
Score, sentiment, and key NPS reason

03: Analysis

Current State
Why the current design creates friction
Heuristic Violations
Which usability heuristics are violated (Nielsen’s 10)
e.g., “Violates Recognition over recall — users must guess what to do”
Mental Model Gap
User expects X, system provides Y

04: Impact & 05: Recommendation

Impact
Segments affected, task completion impact, downstream effects, frequency
Proposed Change
Specific UX improvement with before/after design direction
Code Impact
Files and components that need modification
Effort & Risk
Small / Medium / Large · May require design review

Template: A2 — Output Quality Finding

Use when the system produces output that is inaccurate, irrelevant, or incomplete. The feature works mechanically but quality is insufficient.

Meta

Document ID
A2-YYYY-NNNN
Feature / Output
Which feature and what kind of output

01: Observation + 02: Quality Breakdown

Observation
What the user received vs what they needed
Accuracy
Factual correctness, source attribution, logical consistency, hallucinations
Each check gets a Pass/Fail with detail
Relevance
Matches user intent, correct scope, actionable, right depth
Completeness
All sections present, sufficient detail, edge cases, formatting

03: Evidence + 04: Root Cause

Output Comparison
Table of expected vs actual for each key aspect
User Feedback
NPS, debrief quote, think-aloud excerpt, usefulness rating
Root Cause
Bad prompt / Wrong data source / Model limitation / Missing context / Pipeline error / Stale data
Each cause is checked Yes/No with specific detail
Pipeline Trace
Step-by-step trace showing where quality degraded
e.g., Step 1: Input parsing dropped “SaaS” qualifier → Step 2: Wrong industry mapping

05: Impact & 06: Fix Strategy

Impact
User trust, decision risk, frequency, affected segment
Fix by Layer
Prompt / Data pipeline / Data source / Validation layer / Model config
Each layer gets a specific change description
Code Changes
File paths and specific changes needed
Eval Suite
Whether a quality evaluation suite is needed before deploying

Template: A8 — Reliability & Trust Finding

Use when results are inconsistent, users can’t predict what they’ll get, or confidence levels are miscalibrated.

01: Observation + 02: Reliability Assessment

Observation
What inconsistency or trust issue was observed
Consistency Test
Multiple runs with identical inputs — how much did output vary?
Track variance in structure, key data, recommendations, tone, and length
Confidence Gap
System-reported confidence vs actual accuracy
Over-confident / Under-confident / Calibrated

03: Evidence + 04: Root Cause

Trust Impact
User trust rating, would they act on the output, manual verification time
Root Cause
Non-deterministic output / Inconsistent data sources / Missing validation / Race conditions / Cache issues / Uncalibrated confidence

05: Recommendation

Changes by Target
Determinism (temperature, seed) / Validation (schema enforcement) / Confidence (calibration) / Caching / Transparency (show indicators)
Code Changes
Config and code files that need modification

Priority Framework

All templates use the same 4-level priority scale. Assign based on severity and impact on the DP’s ability to get value from the feature.

P0
Critical
Blocks core workflow, no workaround. DP cannot complete their task.
P1
High
Significant friction, workaround exists. Degrades trust or output quality.
P2
Medium
Noticeable but manageable. Feature works but experience is confusing.
P3
Low
Minor, cosmetic, or edge case. No significant impact on core workflows.

What Every Template Shares

Regardless of template type, every fix document follows the same lifecycle and includes these common elements.

Common Structure Across All Templates

Meta
Document ID, title, priority, status, source, session ID, feature
Observation
What happened — the finding described from the user’s perspective
Evidence
Session data, user quotes, NPS scores, technical data supporting the finding
Analysis / Root Cause
Why this happened — code-level, design-level, or pipeline-level explanation
Impact
Who is affected, how severely, and how often
Recommendation
Proposed fix with specific code/config changes, effort, and risk assessment
Review & Approval
Core team reviewer, decision (approve / modify / defer / reject), sprint priority

Data Sources AI Uses to Draft Fixes

When you click “Generate Fix,” Testiment’s AI pulls from all available session data to populate the template.

Recording
Session video, screen capture, camera feed
Console & Network
JavaScript errors, failed API calls, action steps
Transcripts
Think-aloud with word-level timestamps
NPS & Debrief
Scores, reasons, and post-session answers
User Steps
Recorded action sequence with timestamps
Code Analysis
Repo, file, and line-level context from the codebase

Writing a Strong Fix — Tips

  • Always quote the DP directly when describing the issue — “I don’t know why this scored 72” is more compelling than “scoring felt opaque”
  • One fix per issue. If you find 3 problems in one session, generate 3 separate fix documents
  • Let AI draft first, then refine. Don’t start from scratch — AI handles the data assembly, you add domain expertise
  • Include timestamps when referencing session recordings so the core team can jump to the exact moment
  • Be specific about code changes. “Fix the bug in canvas” is vague. “Add null coalesce in canvas-summary.tsx:142” is actionable
  • When in doubt, over-document. It’s easier for the core team to skim than to ask for missing context
Assignments
SME Assignments

Each SME owns specific features and generates fixes for their domain. Find your assignment below.

SME-01
Fakhar Abdullah, Aqib Zafar (AZ)
Output Quality & Methodology
P1-IN-F1, P1-IN-F7
Prompt tuning, agent output quality, stage-gating
SME-02
Fakhar Abdullah, Aqib Zafar (AZ), Emir Arat
Market Intelligence & Data Accuracy
P1-IN-F2, F3, F4
Research output accuracy, data completeness
SME-03
Alemsah
UX & Onboarding
P1-IN-F5, P2-AN-F6
Onboarding flows, learnability, first-impression clarity
SME-04
Bassalat Sajjad, Fakhar Abdullah, Aqib Zafar (AZ)
AI / Agent Quality
P1-IN-F6, P3-AC-F2, F3, F4
HITL, agent trust, directive quality
SME-05
TBA
Knowledge Mgmt & Stage-Gating
P1-IN-F7
Folder structure, agent-to-folder mappings
SME-06
TBA
Analytics & Data Visualisation
P2-AN-F1, F2, F5
Charts, data density, cross-theatre consistency
SME-07
TBA
Strike Intelligence & Battle Zones
P2-AN-F3, F4, F7, P3-AC-F1
AAARRR mapping, lifecycle transitions
SME-08
TBA
Outreach & Messaging Quality
P3-AC-F5 (Outreach PowerUps)
Copy quality, A/B signal interpretation
SME-09
TBA
Konnect & Outreach Intelligence
P4-KO-F1 through F5
Contact enrichment, scheduling, synthesis
SME-10
TBA
Build & Prototype Quality
P5-BD-F1, F2
SDD accuracy, scaffold success rate
SME-11
TBA
Integrations & Data Connectivity
P2-AN-F8, F9, P3-AC-F5
Connector coverage, event mapping accuracy
← Back to Briefing