The Charter
Midcore is the platform every NeuroBazar product is built on. It is also, by deliberate design, the dual instrument behind our position on the AI era. Each of the 20 modules below is published with what it makes possible (enables) and what it makes harder (regulates). We expect to be measured against both columns.
Enables — the consenting AI
What this module makes possible for users, teams, and societies.
Regulates — the extractive AI
What this module makes harder, or impossible, by default.
Status
Shipping today · in beta · on the 2026 roadmap.
Foundation modules
No. 01
Models the organisation’s real state so AI never speculates beyond it.
Enables — Good AI
Organisational self-knowledge: a continuously updated mirror of your processes, accounts, and constraints. AI decisions are grounded in what *is*, not what a prompt suggested.
Regulates — Bad AI
Speculative overreach. AI cannot recommend what it cannot map to a real state in the twin. Eliminates a class of confident-sounding hallucinations at the workflow boundary.
No. 09
Knowledge retrieval gated by policy, citation by default.
Enables — Good AI
Grounded answers. Models cite their sources because retrieval feeds them sources — not vibes.
Regulates — Bad AI
Confident hallucination. If a claim has no retrievable source, it cannot be presented as fact.
No. 12
Policy enforcement across every tenant, model, and surface.
Enables — Good AI
Multi-tenant, multi-region governance without bespoke spreadsheets. A single panel for who, what, where, when.
Regulates — Bad AI
Unilateral AI action. Nothing ships, accesses, or decides without an explicit policy permission.
No. 18
Sorting that knows where it can and cannot be applied.
Enables — Good AI
Accurate categorisation across tickets, transactions, content, and customers.
Regulates — Bad AI
Discriminatory bucketing. Classifiers carry a domain of validity and refuse to label outside it.
Evidence & explanation modules
No. 06
Five agentic personas that pressure-test decisions before they go out.
Enables — Good AI
Structured pre-mortems for founders without a board, councils without quorum, ops leaders without a peer.
Regulates — Bad AI
Echo-chamber decisions. Disagreement is surfaced as a feature, not a bug.
No. 10
Hash-chained audit of every AI action.
Enables — Good AI
Auditability that respects time. A regulator, a customer, or a future-you can reconstruct exactly what the system did, when, and why.
Regulates — Bad AI
Untraceable AI actions. No log gap, no convenient amnesia.
No. 14
Suggestions disclosed as suggestions, with the user always in charge.
Enables — Good AI
Helpful surfaces — products, articles, next steps — that improve session value transparently.
Regulates — Bad AI
Dark-pattern manipulation. Recommendations are labelled, their reasoning is visible, and the user can switch the engine off.
No. 15
Forecasts built from named drivers, not opaque models.
Enables — Good AI
Principled FP&A: every plan is a tree of named drivers your CFO can argue with.
Regulates — Bad AI
Black-box algorithmic budgeting that nobody can challenge.
No. 20
Your brand on the surface; sovereign over the substance.
Enables — Good AI
Brand sovereignty. Ship AI features under your name, your terms, your hosting if needed.
Regulates — Bad AI
Vendor lock-in dressed as innovation. Full handover. Code custody. Exit-ready by design.
Human-facing modules
No. 05
Every AI decision arrives with its reasoning attached.
Enables — Good AI
Trust the user can verify. Recommendations, classifications, and forecasts ship with the chain of reasoning visible — not buried.
Regulates — Bad AI
Black-box decisions affecting people. If we cannot explain it, we will not ship it.
No. 07
AI that grows the human, not replaces the human.
Enables — Good AI
Skill compounding. Each interaction makes the user more capable — not more dependent.
Regulates — Bad AI
Cognitive offloading dressed as productivity. Coaching surfaces the *why* alongside the *what* so judgement remains the human’s.
No. 13
OCR, classification, extraction, and authenticity scoring.
Enables — Good AI
Paperwork dignity. Forms get understood, claims get processed, contracts get summarised — without a back-office sweatshop.
Regulates — Bad AI
Document fraud and forgery. Tamper heatmaps with confidence, cross-document linkage, regulator-grade evidence.
No. 17
Calibrated projections with confidence intervals you can take to a board.
Enables — Good AI
Realistic outlooks across cash, demand, capacity, and risk — anchored in the digital twin, not in vibes.
Regulates — Bad AI
Hype-driven projections. Confidence intervals are mandatory. Best-case-only forecasts get flagged.
Discipline modules
No. 02
Learns each actor’s legitimate pattern so deviations become legible.
Enables — Good AI
Consent-respecting personalisation. Each user, vendor, approver, and cost centre carries a learned fingerprint that makes legitimate behaviour fast and frictionless.
Regulates — Bad AI
Manipulation and coercion. Anomalies in spending, approval, or access patterns surface with confidence scores — before money moves, not in next quarter’s audit.
No. 08
Discipline for the model lifecycle.
Enables — Good AI
Registry, training, drift detection, deploy gates — accessible to teams that don’t want a platform contract.
Regulates — Bad AI
Shadow models shipping into production without controls, versioning, or rollback.
No. 11
Workflow autonomy with cost ceilings and evidence chains.
Enables — Good AI
Ops teams build agentic workflows in a visual canvas — production-grade from day one.
Regulates — Bad AI
Rogue automation. Every action carries a budget, an audit trail, and an off-switch.
No. 16
Models trained without centralising your data.
Enables — Good AI
Privacy-preserving AI. Hospitals, banks, and family offices share insight without sharing rows.
Regulates — Bad AI
Centralised data harvesting. The data stays where it belongs; only the gradients travel.
Policy modules
No. 03
A policy DSL where ethical limits are code, not a poster on the wall.
Enables — Good AI
Author-once, enforce-everywhere ethics. Codified principles run on every model call, every tool use, every workflow.
Regulates — Bad AI
Ad-hoc ethical drift. No quiet bypass, no rushed exemption, no "just this once."
No. 04
Runtime guardrails on every model call.
Enables — Good AI
Safe rollout of agentic features into production. Teams ship LLM surfaces knowing the floor — not hoping.
Regulates — Bad AI
Unsafe outputs at runtime. Tool calls outside policy, data exposure across tenants, prompt injection from upstream content.
No. 19
Consent paths that AI can read, follow, and never bypass.
Enables — Good AI
Clear paths for who approves what, when, with what evidence.
Regulates — Bad AI
Bypassed controls. No agent can "just go ahead" past an approval gate, no matter what the prompt says.
Charter, in one line