Evidence-based assessment framework · First live instance: Odoo ERP

Structured assessments
grounded in evidence,
not impressions.

OIAR transforms multi-stakeholder interviews and documents into structured, evidence-backed reports — so your assessments hold up to scrutiny and drive real decisions. Human-in-the-loop at every critical gate.

O Orchestrate I Interview A Analyze R Report
// 00 · PROBLEM

Diagnosis fails before analysis

In enterprise contexts, information is distributed, incomplete and often conflicting. Facts, opinions and interpretations mix together. Conclusions don't hold up to scrutiny because sources are not traced.

Distributed sources

You spend weeks gathering info that's still incomplete

Every stakeholder has a partial version. Gathering them manually introduces delays, gaps and unverifiable interpretations.

Bias and diverging incentives

Dominant voices suppress the signals that matter most

Without systematic isolation and classification, dominant statements suppress minority signals that may be critical.

Indefensible reports

Reports without evidence chains don't survive scrutiny

A report without a traceable link between evidence and recommendation is opinion, not analysis. It won't hold up to scrutiny or generate decisions.

// 01 · FRAMEWORK

Four phases. One repeatable process.

OIAR is not a generic chatbot. It's a framework for orchestrating structured information gathering, isolating sources, classifying evidence and producing operational reports.

01 · O
O

Orchestrate

The Orchestrator Agent analyzes the objective, produces the Global Objective and Plan, assigns skills to Interviewer Agents and activates them in parallel.

  • Global Objective (MD)
  • Assessment Plan (MD)
  • Skill assignment
  • HITL-PLAN: plan approval
02 · I
I

Interview

N Interviewer Agents operate in parallel, each segregated in their own scope. They gather documents via MCP channels and produce a structured Interview MD.

  • Parallel interviews (N×)
  • Isolated scope per agent
  • Email · Chat · Voice
  • HITL-CLOSE: close interviews
03 · A
A

Analyze

The Analysis Agent reads all Interview MDs, classifies every statement (Fact, Evidence, Contradiction), detects gaps and assigns confidence scores.

  • Claim classification
  • Evidence linking
  • Contradiction detection
  • HITL-REPLAN: critical gaps
04 · R
R

Report

The Report Agent produces the final document according to the scenario template: Executive Summary, Finding, Evidence Map, Roadmap. Publication only after HITL-REPORT.

  • Executive Summary
  • Finding + Evidence Map
  • Prioritized Roadmap
  • HITL-REPORT: final approval
// 02 · FLOW

5 phases, 4 mandatory gates

Every critical transition requires explicit approval. You can reject with notes: the system returns to the previous phase and incorporates your feedback before re-presenting the result.

01

Conversation with the orchestrator

You start a chat with the orchestrator: define objective, scope, version and stakeholders. The system generates the assessment plan.

Objective Modules in scope Stakeholders
HITL
HITL-PLAN · Plan approval · Mandatory
Review plan, stakeholders and interview skills. Approve or reject with notes — the system incorporates feedback and re-proposes.
HITL-PLAN
HITL
HITL-INTERVIEW · Start interviews · Mandatory
Approve to send email invitations and create dedicated chats. Data Room folders are created only at this point.
HITL-INTERVIEW
02

Parallel interviews via dedicated chat

Each stakeholder receives an email link with a private isolated chat. Agents operate in parallel, each in their own scope and Data Room folder.

Dedicated chat per stakeholder Segregated scope interview.md
HITL
HITL-CLOSE-INTERVIEW · Close interviews · Mandatory
Auto-approves when all interviews are completed. You can force closure manually. Only then does the analysis begin.
HITL-CLOSE
03

MAP-REDUCE analysis of evidence

The Analysis Agent classifies each statement (Fact, Opinion, Hypothesis, Contradiction), detects module gaps and produces a sufficiency score.

Claim & Evidence Contradiction detection Sufficiency score
HITL
HITL-REPLAN · Insufficient data · Optional
If evidence is insufficient, the system suspends. Approve to reopen interviews, reject to proceed with available data.
HITL-REPLAN
04

Structured report generation

The Report Agent produces Executive Summary, Gap Analysis per module, Claim Evidence Map and remediation Roadmap. Written in Markdown, downloadable.

Executive Summary Evidence Map Roadmap
HITL
HITL-REPORT · Publication approval · Mandatory
Read the report before publication. Reject with notes: the system regenerates incorporating feedback, without losing previous work.
HITL-REPORT
05

Publication in Data Room

The approved report is published in reports/final-report.md. The audit log of every HITL decision is retained and consultable.

final-report.md Audit Log Download MD
// 03 · PRINCIPLES

Designed for conflicting information

OIAR assumes sources are partial, biased or incomplete. It doesn't seek consensus: it seeks triangulation, evidence and transparency about gaps.

Evidence-first

Every claim is classified

Declared fact, opinion, hypothesis, documentary evidence, contradiction. The classification is preserved in the data model and in the report.

Least Knowledge

Each agent sees only its scope

The Interviewer Agent receives no statements from other stakeholders. Cross-source contamination bias is eliminated structurally.

Human-in-the-loop

5 HITL gates (4 mandatory)

HITL-PLAN, HITL-INTERVIEW, HITL-CLOSE, HITL-REPLAN (opt.) and HITL-REPORT. The system never advances alone at critical transitions.

Byzantine-resilient

Numerical consensus is not truth

A minority with strong evidence can be more relevant than a majority without documentary support. Contradictions are made explicit, not suppressed.

// 04 · OIAR INSTANCE

One framework. Every scenario is an instance.

OIAR is not a single portal. It's a template instantiated once per scenario. Every instance has the same screens, panels, logic — only the color theme and loaded skills change.

OIAR Core — shared structure
menu · panels · agentic graphs · HITL gates · audit log · Data Room · Skill Repository
shared · never forked
LIVE · BETA
Odoo Assessment
Live beta — gap analysis · pre-implementation · project review
Gap Analysis ERP Maturity Remediation
Try the live instance →
Project Recovery
coming soon · IT projects · blocked or out-of-control
Root Cause Risk Map Recovery Plan
M&A / Due Diligence
coming soon · vendor assessment · technical diligence
Tech Assessment Vendor Risk Maturity
Golden rule

Shared structure, color per instance

If a change touches layout, panels, menu or behavior → it belongs to the shared layer. If it touches color or font → it belongs to the instance theme file.

Single accent

"Single signal accent" discipline

Each instance has one accent color (--signal). A second color (--accent-2) is defined but not wired — only for scenarios that explicitly require it.

Create a new instance

Duplicate · change 4 tokens · touch nothing else

Duplicate the reference scenario folder. Change only the 4 accent tokens and neutral tint in the theme file. No panels are touched.

// 05 · SKILL REPOSITORY

Skills are the instance's brain

Each agent is not hard-coded: its instructions are versioned Markdown files in the repository, selected by the Orchestrator based on the stakeholder profile and scenario. To create a new instance, write its skills.

agents/skills/ · Skill Repository
📁 agents/skills/
├── 📂 interview/
├── technical-stakeholder.mdIT Manager · DevOps
├── executive-sponsor.mdCIO · Business Owner
├── vendor-representative.mdPartner · Fornitore
└── key-user.mdUtente operativo
├── 📂 analysis/
├── contradiction-detection.md
└── confidence-scoring.md
└── 📂 report/
├── project-recovery.md
├── odoo-assessment.md
└── vendor-dispute.md

Git-versioned Markdown files

Skills are MD files in the repository, not runtime configurations. They change via PR with review, not through an interface. Versioning is free.

Selected by the Orchestrator

The Orchestrator picks skills from the Skill Repository suited to each stakeholder's profile and injects them into the assigned Interviewer Agent's prompt.

One skill per stakeholder profile

Technical stakeholder, executive sponsor, vendor representative, operational key user. Each profile has dedicated questions, signals and output format.

Creating an instance = writing its skills

New scenario (SAP Migration, Vendor Dispute, …): duplicate the most similar instance folder, write the specific skills, modify the theme file. Zero code.

// 06 · USE CASES

Every scenario is an OIAR instance

Same framework, different skills, different theme. Every use case originates from duplicating the base instance with the appropriate skills loaded in the repository.

Odoo Assessment · live

ERP Assessment (Odoo)

Gap analysis of Odoo implementation by module, ERP maturity, stakeholder misalignments, prioritized remediation roadmap.

IT ManagerBusiness OwnerKey UserOdoo Partner
Prova l'istanza live →
IT Assessment · coming soon

Project Recovery

Late, blocked or out-of-control IT project. Root cause diagnosis, risk map, dependencies, missing decisions, recovery plan with evidence.

SponsorPM · PMODev TeamVendor
Due Diligence · coming soon

M&A / Vendor Due Diligence

Rapid evaluation of a vendor, system or technical asset before a major commitment. Key risks, maturity assessment, gaps and prioritized recommendations.

AcquirenteTeam tecnicoLegalConsulenti

The first live instance is Odoo ERP assessment.

Start with a concrete use case — gap analysis and pre-implementation assessment for Odoo projects. If your use case is different, tell us.

No spam. Your data is never shared with third parties.

Request sent

We'll get back to you shortly to arrange a demo of the framework on your scenario.

OIAR · Orchestrate · Interview · Analyze · Report