Feizhen follows a structured seven-stage process to turn your request into a trustworthy, evidence-backed recommendation. Every stage is transparent and every decision is explained.
1
User Request
You describe what you are looking for and optionally set a budget. Feizhen identifies all candidate products in its database that match your request category.
↓
2
Evidence Collection
For each candidate product, Feizhen gathers all associated evidence records — laboratory test results, manufacturer specifications, editorial reviews, marketplace data, and regulatory records. Each record carries a full source profile including type, date, methodology, and commercial relationship.
↓
3
Evidence Evaluation
The Evidence Engine (EE-1.0) runs Source Trust Assessment on every evidence record. Five criteria are scored independently: Authority, Verifiability, Independence, Recency, and Corroboration. The combined score determines whether evidence is rated Strong, Moderate, or Limited — and caps what overall trust scores are possible.
Manufacturer evidence is always rated Limited due to commercial interest. Independent laboratory evidence can achieve Strong.
🔎 EE-1.0
↓
4
Constitution Engine
The highest authority in the Feizhen system. CE-2.0 evaluates every candidate against 10 constitutional rules covering commercial disclosure, evidence-based reasoning, balanced representation, uncertainty disclosure, trade-off disclosure, and unsupported claims.
Seven rules are mandatory. Any mandatory failure immediately withholds the recommendation — no score, no override, no exception. Commercial interests can never change this outcome.
⚖ CE-2.0
↓
5
Trust Score Engine
TSE-1.0 scores six trust categories: Evidence Quality, Source Independence, Transparency, Risk Disclosure, Uncertainty Disclosure, and Commercial Neutrality. Scores range from 0 to 100. Four categories are mandatory — a score below 50 in any mandatory category withholds the recommendation, regardless of the overall score.
The evidence base quality also applies a ceiling: a small or weak evidence base limits the maximum overall trust score achievable.
📈 TSE-1.0
↓
6
Recommendation
Only products that pass every mandatory constitutional rule and every mandatory trust category receive a published recommendation. Products that fail are listed as withheld, with complete reasons for each failure.
Published recommendations show the top match and the best alternative. Every score, every rule result, every evidence record, and every claim is fully visible.
↓
7
Final Decision by User
Feizhen shows its complete reasoning. Every claim is linked to the evidence that supports it. Every score is broken down and explained. Every withheld product comes with a full explanation of why it was withheld.
Feizhen never makes the decision for you. The recommendation is a starting point — the final decision always belongs to the user.
The Constitution is the highest authority in the Feizhen Decision Engine. No commercial interest, no financial relationship, and no external pressure can override a constitutional rule. Every recommendation is subject to it.
⚖ The highest authority
The Feizhen Constitution defines the non-negotiable principles under which every recommendation is made. It exists to protect the user. Seven of the ten rules are mandatory — failure on any mandatory rule immediately withholds a recommendation, with no exceptions and no overrides.
The Constitution was written before any product was assessed. It applies equally to all products, all categories, and all circumstances. It cannot be amended to benefit a commercial partner.
10 constitutional rules
1
Trust Before RevenueMandatory
A commercial disclosure must be present and must confirm that no financial relationship influenced the assessment. The user's interest is placed before any commercial consideration — always.
2
Evidence-Based ReasoningMandatory
Every recommendation must be backed by at least one traceable evidence record. A recommendation made without evidence cannot be published. At least one record must be rated Strong for a full pass.
3
Source IndependenceNon-mandatory
Evidence should come from sources independent of the manufacturer. If all evidence originates from the manufacturer, the recommendation is flagged for review. Independent sources reduce — but must be sought to replace — manufacturer claims.
4
Recency of EvidenceNon-mandatory
Evidence must be reasonably current. Sources that are outdated or have exceeded their expected validity window are flagged for review, prompting a refresh before the next assessment cycle.
5
Balanced RepresentationMandatory
Both advantages and disadvantages must be documented. A recommendation that presents only advantages is misleading. Every product has trade-offs — the Constitution requires they be stated before a recommendation can be published.
6
Uncertainty DisclosureMandatory
Known limits of confidence must be explicitly stated. Every recommendation must include an uncertainty level and a written explanation covering what is known, what is estimated, and what the available evidence cannot confirm.
7
Trade-off DisclosureMandatory
Trade-offs, disadvantages, and risks must all be documented. A recommendation that omits any of these three sections cannot be published. Material information must never be withheld from the user.
8
No-Commercial-Benefit TestMandatory
The recommendation must be identical whether or not a commercial benefit exists. If this test cannot be confirmed, the recommendation is withheld. A commercial disclosure is required before the test can be run.
9
User-Centric PurposeNon-mandatory
Every recommendation must be addressed to a defined intended user. Recommendations without a stated user profile are flagged for review. Feizhen recommends to people, not to markets.
10
Unsupported Mandatory ClaimsMandatory
Every mandatory recommendation claim must be linked to at least one evidence record. A claim that cannot be traced to evidence is unsupported. Unsupported mandatory claims immediately withhold the recommendation — the user cannot be asked to trust a claim that has no evidence behind it.
⚠ Commercial interests never override constitutional rules
No commercial relationship, affiliate arrangement, advertising agreement, or financial incentive of any kind can change the outcome of a constitutional check. Mandatory rules apply in full to every product, every time. A product that fails a mandatory rule is withheld — no exceptions, no partial overrides, no commercial workarounds.
The Constitution was designed this way deliberately. Trust is only meaningful if it cannot be purchased.
Feizhen only makes claims it can support. Every statement in a recommendation traces back to a specific evidence record with a named source, a date, a type, and a trust rating. Evidence is the foundation of trustworthy advice.
What is evidence?
In Feizhen, evidence is any recorded, traceable piece of information used to support a recommendation claim. It is not an opinion, an assumption, or a general impression. Each evidence record carries: the source name and type, the date the information was gathered, the methodology used (where applicable), the strength rating, and the commercial relationship between the source and the product manufacturer.
Evidence records are the building blocks of every recommendation. A claim without a supporting evidence record cannot appear in a published recommendation.
Why evidence matters
Most recommendation systems tell you what to buy without telling you why — or what assumptions they are making. Feizhen is different. By requiring that every claim trace to an evidence record, the system makes its reasoning visible. You can see what the recommendation is based on, how reliable that basis is, and where the gaps are.
Evidence also limits what can be claimed. If a product has no strong independent evidence for a performance claim, that claim cannot be rated as reliable — and a product with only manufacturer evidence for its most important claims will fail constitutional checks.
How evidence quality is assessed — Source Trust Assessment (STA)
Every evidence record is scored on five criteria. The resulting STA score determines the record's reliability rating: Strong (85+), Moderate (60–84), Limited (35–59), or Very Limited (0–34).
1
Authority
Does the source have the expertise, credentials, or institutional standing to make this claim? A specialist laboratory carries higher authority than a general-purpose publication for a technical measurement.
2
Verifiability
Can the claim be independently checked? Published methodology, raw data, and reproducible test conditions all increase verifiability. Marketing copy with no methodology scores low.
3
Independence
Is the source free from commercial relationships with the manufacturer? Sources funded by subscriptions or grants, with no manufacturer financial ties, score the highest. Manufacturer-produced evidence scores 0 on independence.
4
Recency
Is the evidence current? Product performance can change with firmware updates, component revisions, and market changes. Evidence older than 18–24 months for fast-moving product categories is flagged as potentially outdated.
5
Corroboration
Is the claim supported by more than one independent source? A claim confirmed by multiple independent records is more reliable than one supported by a single source. Corroboration is especially important for mandatory recommendation claims.
Evidence source types
🏠
Manufacturer
Specifications, press releases, and product pages produced by the company selling the product. The source has a direct commercial interest in presenting the product favourably.
⚠ Limited
🔬
Independent
Test results and analysis from laboratories or researchers with no financial relationship with the manufacturer. Funded by subscriptions, grants, or public bodies. The gold standard for evidence reliability.
✓ Strong
🛒
Marketplace
Product listings, pricing data, and user reviews from retail platforms. The platform has a commercial interest in completed transactions. User reviews vary greatly in reliability and are prone to manipulation.
⚠ Limited
⚖
Government / Regulatory
Safety certifications, compliance records, recall notices, and regulatory assessments from statutory authorities. No commercial interest. The highest authority rating for safety and compliance claims.
✓ Strong
📺
Editorial
Reviews and analysis from journalists and publications. Generally independent but methodology varies. May be influenced by advertising relationships. Useful for qualitative assessments alongside laboratory data.
ℹ Moderate
👥
User Feedback
Reports and opinions from consumers who have used the product. Highly variable in quality. Useful for identifying real-world issues not captured in controlled tests, but unreliable as a primary evidence source.
Browse, search, and explore every evidence record in Feizhen's knowledge base. Filter by category, tag, or linked constitution rule. Inspect version history and cross-references in full detail.
Loading evidence records…
Loading…
TII-1.0 · Trust Intelligence Platform
📊 Trust Intelligence
Aggregate patterns across all decisions — trust scores, constitution outcomes, anomalies, and engine drift. All signals are for review, never proof of causation.
Helping people make better decisions through honesty, transparency, evidence and trust.
Our approach
Most recommendation platforms use hidden algorithms — you receive a result but never see how it was reached, what evidence was used, or what assumptions were made. Conflicts of interest are embedded in the business model and invisible to the user.
Feizhen is built on a different principle: explainability before convenience. Every recommendation shows its complete reasoning. Every score is broken down. Every claim is linked to evidence. Every withheld product explains exactly why it was withheld. The user can audit the entire decision process.
Principles
🔎
Explainability over convenience
A recommendation that cannot be explained cannot be trusted. Feizhen shows every step of its reasoning — not because it is required to, but because that is what trust requires.
⚖
The Constitution is the highest authority
No commercial interest overrides a constitutional rule. The rules were written before any product was assessed and they apply equally to every product, every time. This is not a policy — it is the architecture of the system.
📈
Evidence-first reasoning
Claims without evidence are not claims — they are opinions. Feizhen treats the two differently. Every mandatory claim in a recommendation must trace to a named source, a date, and a trust rating.
👤
The final decision belongs to the user
Feizhen does not decide for you. It gives you structured, evidence-backed analysis and then steps back. You see what the system knows, what it is uncertain about, and what it cannot confirm. Then you decide.
Developer Preview 1.4 — Release Notes
Developer Preview 1.4 adds the Account, Consent and Privacy Experience (ACP-1.0): a demonstration prototype of how future Feizhen users will access the service, manage privacy choices, understand behavioural analytics, and view their information.
👤
Account experience prototype
Sign In, Create Account, and Guest Mode demonstration flows. No real authentication or data storage.
📈
Decision history
Fictional decision history records with status, trust scores, constitution results, and data limitations.
🔒
Privacy and consent centre
Five consent categories (Essential, Personalisation, Analytics, Voice, Marketing) with real toggles. Optional analytics off by default. No preselected optional consent.
👥
Privacy dashboard
Demonstration view of data held, consent history, active sessions, and data management actions including download, clear, and delete.
🔎
Aggregated behavioural analytics
Internal admin analytics using fictional aggregated data. No personal identifiers. No individual surveillance. Clearly labelled as demonstration data.
▹
Behavioural event registry
11 approved demonstration events with consent requirements, retention categories, and governance notes. No keystroke logging, microphone recording, or cross-site tracking.
⚙
Public and internal navigation separation
Technical modules (Evidence DB, Architecture, Arch Registry, Data Sources, Workflow) moved to Admin Console. Public navigation simplified to Home, Start Decision, How It Works, The Constitution, Evidence, About, Contact, Sign In.
Developer Preview 1.5 — Release Notes
Developer Preview 1.5 adds the Decision Experience DX-1.0: a full human-centred Decision Wizard that guides users through choosing a product, understanding the evidence, and seeing explainable fictional recommendations backed by the Feizhen Constitution.
▶
Human-centred Decision Wizard
Six-step guided wizard: Decision Type → Context → Priorities → Constraints → Review → Recommendation. Progress indicator on every step. Back, Continue, Save Draft, and Exit available throughout.
🎯
Guided context and priority collection
Structured inputs for product category, intended user, main use, budget, and timing. Priority ranking with a five-priority maximum. Users are told priorities do not override constitutional safety or evidence requirements.
📈
Explainable fictional recommendations
Three ranked fictional options (Best Match, Strong Alternative, Lower-Cost Alternative) each with match summary, priority fit, trust score, confidence level, evidence count, and main limitations.
🔓
Constitutional result display
Each recommendation shows a full constitutional check across all rules. PASS, NEEDS REVIEW, and BLOCKED results are shown. A BLOCKED result prevents a positive recommendation from being shown.
📄
Evidence-linked recommendation view
Each recommendation links to its supporting evidence records. Evidence IDs, source categories, trust ratings, independence status, and data freshness are all disclosed.
▶
Product comparison and feedback
Side-by-side comparison of all three options across 14 dimensions. Optional post-recommendation feedback. Decision History integration for signed-in users. Voice Interaction System VIS-1.0 added to Architecture Registry as Planned.
Prototype notice
Feizhen Developer Preview 1.5 is a testing prototype using entirely fictional demonstration data. All products, prices, evidence records, test results, scores, account data, and decision wizard results are invented for the purpose of demonstrating the decision architecture, privacy experience, and recommendation explanation flow. No real products are assessed. No external APIs are connected. No user data is collected. No real AI is used. DX-1.0 produces fictional demonstration recommendations only.