AI EVIDENCE ENGINEERING

AI Scoring
/ ReviewSolution

More than a score. Bring business standards, source evidence, hard gates, follow-up questions, and human review together in an explainable, replayable, and auditable review chain.

◎ Every score has evidence◇ Every hard gate is judged independently✓ Every conclusion can be reviewed by a person
REVIEW RUNNINGSUPPLIER-ENTRY-V3.2
78/ 100
!

Hard gate triggeredEnvironmental compliance certificate expired

Re-review After Documents Submitted
Qualifications and compliance 18/30Delivery Capability 24/25Quality and service 21/25
Evidence 06Gap 01Human review pending
Problem Definition

Not asking the model to “go with its gut”
But making business standards actionable

The challenge in traditional reviews isn’t adding up scores. It’s that standards are scattered, evidence is hard to find, hard gates are easy to miss, and conclusions can’t be traced. AI adds value by turning this work into a product.

BEFORE

Manually find materials one by one

  • Standards live in documents and experience
  • Scores exist, but the evidence chain is incomplete
  • Reviewers apply inconsistent standards
  • Hard gates can be masked by average scores
Engineer standards→Structure evidence
WITH BOLTFUSION

Complete reviews with humans and AI

  • Approved template versions ensure consistent standards
  • Trace every score to source text and data
  • Don’t guess at gaps—generate targeted follow-up questions
  • Final recommendations and overrides are auditable
01

Clear Criteria

Templates, Versions, Weights, and Definitions

→
02

Find Evidence

Source Text, Location, Timestamp, and Source

→
03

Hard Gates

Veto Criteria and Required Documents

→
04

Calculate Scores

Weighted Scores, Dimension Scores, and Confidence

→
05

Enable Review

Differences, Comments, Sign-off, and Read-Back

Complete Scenario Overview

One foundation, covering eight
common scoring and review tasks

Scenarios may differ, but they share the same governance approach: define standards first, then link scores, risks, gaps, and recommendations back to verifiable business evidence.

01
Suppliers

Supplier Qualification

Qualifications, Delivery, Quality, and Compliance

OUTPUTQualification Score + Hard-Gate Hits + Document Checklist
02
T

Recruitment Interviews

Competency Models, Interview Notes, and Work Samples

OUTPUTDimension Scores + Evidence Statements + Follow-Up Questions
03
C

Contract Review

Standard Clauses, Rights and Responsibilities, and Risk Deviations

OUTPUTRisk Level + Source Text Location + Suggested Revisions
04
Procurement

Procurement Bid Evaluation

Scoring Method, Bid Documents, and Deviation Table

OUTPUTObjective Score + Responsiveness + Issues List
05
¥

Expense Compliance

Policies, Receipts, Standards, and Related Requests

OUTPUTCompliance Status + Exceptions + Items Requiring Explanation
06
P

Project Review

Business Case, Deliverables, Milestones, and Risks

OUTPUTMaturity + Gaps + Review Comments
07
Quality

Customer Service Quality Assurance

Recordings, Scripts, Tickets, and Service Standards

OUTPUTQA Score + Issue Clips + Coaching Suggestions
08
Content

Content Compliance

Policies, Copy, Documents, and Publishing Boundaries

OUTPUTRisk Items + Sensitive Content Locations + Draft Revisions
Standards Engine

Turn expert knowledge into
versioned scoring templates

Weights determine the score; hard gates determine whether a case can proceed. Manage them separately to prevent serious risks from being averaged away by high scores elsewhere.

Scoring dimensionWeightCurrent scoreEvidence criteriaExecution method
Qualifications and Compliance30%18 / 30

Qualification Documents, Validity Periods, and Approved/Denied Lists

Hard-Gate Check
Delivery Capability25%24 / 25

On-Time Delivery Rate, Capacity, and Fulfillment Exceptions

Quantitative Calculation
Quality and Service25%21 / 25

Defect Rate, Customer Complaints, and Response Time

Timestamp Alignment
Cost and Collaboration20%15 / 20

Quotes, Payment Terms, and Historical Collaboration Records

Definition Conversion
Overall score = Σ (dimension score × versioned weight)78 / 100
Runtime Architecture

ecology remains the source of business truth
BoltFusion provides a governed AI review layer

Identity, permissions, forms, workflows, and business objects remain managed by ecology. Review Agents read evidence, generate recommendations, and write back within authorized contexts and permission boundaries.

01

Business Entry

ecology portals, forms, workflows, mobile, and to-dos

02

Review Context

Person, Role, Permissions, Page, Form, Entity, Workflow, and Data Timestamp

03

Scoring/Review Agent

Standards Parsing, Evidence Collection, Rule Execution, Score Explanation, and Follow-Up on Gaps

AI REVIEW CORE
04

BoltFusion Runtime

Agent API, Work, Monitor, Policy Gate, Human Review, and Runtime Evidence

05

Business and Knowledge Sources

ecology fields/attachments, policies, scoring templates, business systems, and professional services

06

Evidence and Read-Back

runId, requestId, template version, evidence references, write-back fields, nodes, and audit logs

Review Loop

Designed for reviewability from the start

If evidence is missing, ask follow-up questions. If a hard gate is triggered, stop for further review. If the model is unavailable, return `MODEL_UNAVAILABLE`. The system never generates fake default scores or uses scores to gloss over uncertainty.

01

Identify Subject

Pinpoint the business entity, review type, and current workflow stage

02

Load Standards

Load approved template version, weights, thresholds, and hard gates

03

Gather Evidence

Align fields, attachments, source text, and system data by entity and time

04

Check Hard Gates First

Veto rules, required materials, permissions, and conflict rules run independently

05

Calculate & Explain

Output dimension scores, overall score, confidence, evidence, and counterexamples

06

Follow Up & Review

Request missing evidence; responsible owners confirm discrepancies and final conclusions

07

Write Back & Read Back

Write back scores, comments, and status as authorized; read back fields and stage to verify results

REVIEW RESULT ENVELOPEMore than just a score
  • requestIdREQ-DEMO-SCORE-20260729
  • templateVersionSUPPLIER-ENTRY-V3.2
  • score / gateStatus78 / BLOCKED_PENDING_EVIDENCE
  • evidenceRefs6 refs · 1 expired
  • humanReviewRequiredtrue
  • finalStatussucceeded · result_committed
Interactive Demo

See how AI links each score
back to its evidence

AI Supplier Qualification ScoringDemo Supplier A-017 · Simulated data

Results generated
78/ 100
Re-review After Documents Submitted

Re-review After Documents Submitted

GATE-ENV-01 · Environmental compliance certificate expired

Qualifications and Compliance18 / 30
Delivery Capability24 / 25
Quality and Service21 / 25
Cost and Collaboration15 / 20
!

Hard gate check completeGATE-ENV-01 · Environmental compliance certificate expired

Action items have been generated; the final decision is confirmed by the business ownerscoreWrittenBack=true · admissionApproved=false · humanReviewRequired=true
Acceptance Framework

Acceptance is not about whether it “looks right”
It is about whether the “evidence and results are correct”

Each objective must be jointly confirmed against the selected sample set, template version, and business criteria. The website presents an acceptance framework; it does not mean a customer environment has met the requirements.

01ACCEPTANCE TARGET

Consistent Reproduction

Target ≥95%

With the same template version and inputs, dimensions and criteria remain consistent

02ACCEPTANCE TARGET

Complete Evidence

Target ≥95%

Each scoring item must link to its source, location, version, and timestamp

03ACCEPTANCE TARGET

Hard-Gate Recall

Target 100%

Acceptance tests ensure triggered hard gates cannot be masked by an average score

04ACCEPTANCE TARGET

Human Control

Target 100%

Overrides, returns, document requests, reviews, and sign-offs all retain operation logs

Not sufficient as standalone proof of successHTTP 200 / 202Page toastqueued / runningVisible in Monitor
→
Business evidence that must be read backrequestId is non-emptyresult_committedScoring/feedback fieldNode + review status
Implementation Roadmap

Start with one type of review
Build a trusted closed loop

Start with scenarios where standards are stable, evidence is available, and owners are clearly defined. Run in shadow mode first and compare with human reviews, then gradually enable write-back and workflow collaboration.

01

Scenario and standards assessment

Define the review subject, owner, template, hard gates, data sources, and available samples.

OUTPUT · Scenario ownership matrix
02

Evidence and rules engineering

Turn expert criteria into versioned standards, configuring evidence sources, retrieval methods, and follow-up questions for gaps.

OUTPUT · Approved template
03

Shadow run and calibration

AI does not write back. Run alongside actual human reviews, focusing on missed hard gates, evidence discrepancies, and disagreements.

OUTPUT · Calibration report
04

Controlled launch and scenario expansion

Enable write-back of scores and feedback first, while retaining final human review. Reuse in more scenarios only after requirements are met.

OUTPUT · Launch and rollback checklist

Ground every score in evidence. Make every review trustworthy.BoltFusion AI Scoring/Review Solution

All entities, scores, documents, IDs, and run results on this site are anonymized simulated data; they do not represent any customer environment that has been configured or accepted.