Platform · the decision compiler

Decisions arecompiled,not generated.

Ask any AI for a price and it will give you one. Auctorian refuses — unless the evidence resolves, the policy passes, and the decision compiles into an artifact someone is allowed to sign.

The inversion

The model is notthe authority.

Most AI products put a model at the centre and bolt controls on afterwards. Auctorian is built the other way round: a deterministic compiler holds authority — signals, policy, candidates, artifacts — and language models are permissioned workers inside it.

Models may
Interpret intentturn a question into a governed request
Plan analysischoose which registered tools to consult
Explain decisionsarticulate what the compiler concluded, with citations
Models can never
Invent evidenceevery figure must resolve from a signal, model or policy
Execute actionsonly the execution layer acts, inside its envelope
Bypass a guardrailpolicy is enforced below the model, not requested of it
Dispatch to live systemsnothing ships without a signature — today, nothing ships at all
The compile path

From a questionto a signature.

Nine stages stand between raw data and a decision anyone is allowed to act on. The right-hand column is not a mock-up: it is what the real engine produced for the first shipped episode — including the stage it has not exercised yet. The machine is industry-blind; retail is simply the first industry it has compiled for real.

01ScopeA question becomes exact coordinates — which asset, which place, which week — bound in the organisation's own vocabulary, or refused as unanswerable.DEMO-SKU-001 × DEMO-STORE-001
02SignalsEvidence resolves with grade and lineage. Every input is source-backed and attributable — or it is marked absent, out loud.11 resolved · 5 absent
03ReadinessThe gate asks whether this decision is safe to make at all. Missing evidence caps authority before any recommendation exists.YELLOW · 5 gaps named
04CandidatesFeasible actions are generated and compete. Doing nothing is always one of them.3 actions · hold included
05GuardrailsPolicy prunes before intelligence recommends. Most of what could be suggested never survives this stage.floor 25% · raise cap 15%
06SimulationWhat survives is tested against futures that don't go to plan — and against the other decisions in flight.not exercised in the shipped path
07Score & routeQuality is arithmetic; authority is the strictest cap. A high score can still route to a human — and here it did.DQS 96 = 100 − 4 · human approval
08ArtifactThe decision compiles into a hash-bound record: evidence, alternatives, policy results, and who is allowed to sign it.artifact:7c07fd589…
09OutcomeWhat happened is measured against what was predicted. Authority is earned from the result — or taken back.advisory · nothing dispatched
Observed values · first shipped episode · payload 0507dfb5f3f5…Audit every figure
The machine

Nine layers between raw dataand a decision you can sign.

The compile path is what a single decision travels; this is the machine it travels through. The signal layer and the cartridge system carry the moat.

L09Outcome Learning
L08Execution Preview
L07Artifact & Ledger
L06Governance & Routing
L05Candidate & Simulation
L04Decision Runtime
L03Cartridge System
L02Signal Layer
L01Connectivity Layer
Architecture principle

Auctorian separates conversation from authority. The chat interface coordinates the work; deterministic, source-backed infrastructure decides what is supportable.

The Signal Layer and Cartridge System are the moat — the shared substrate every decision is compiled from, and the two rails drawn solid in the stack.

One command surface

Ask it anything.It answers in decisions.

Answer, decide, plan, simulate, audit, learn — six modes on one surface, each with a ceiling on what it may claim. Six industries below, one surface: the grammar is identical in every domain, and only the nouns change.

Claim ceilingSource-backedExplain evidence, signals, or runtime state
Manufacturing

Yield Loss — Weekly Scan

Across three lines, eleven stations show recoverable yield loss.

Scrap at station seven$612K
Rework loops$438K
Unplanned downtime$295K

Yield pressure concentrates on three lines, driven by tool wear, incoming material variance and changeover time.

Source · plant telemetry, three signalsIllustrative · not measured

Six industries, one surface. Nothing above was requested twice — Auctorian surfaces what is drifting before anyone thinks to ask — and every mode answers from the same compiled evidence, under a ceiling on what it is permitted to claim. Only the nouns change between domains.

Agent runtime

One request.
Five governed agents.

One business request activates coordinated agents that understand context, retrieve evidence, generate actions, enforce governance, and render an auditable decision.

The model does not invent the decision. The runtime compiles it from evidence, candidates, constraints, approvals, and outcome logic.

Business request

Should we move next quarter’s maintenance window on turbine four?

01Understandingintent · mode · scope · context

Interpret the user's intent, selected mode, business scope, active decision context, and follow-up questions.

02Evidencesignals · lineage · freshness · quality

Retrieve decision-grade signals with lineage, freshness, quality, and missing-evidence status from enterprise data.

03Decisioncandidates · baseline · simulate · rank

Generate feasible action options, compare them against a baseline, simulate tradeoffs, and rank candidates.

04Governanceguardrails · permissions · risk limits · approvals

Apply guardrails, permissions, risk limits, approval paths, and autonomy ceilings before any action is prepared.

05Presentation & Learningbrief · evidence trail · artifact · outcomes

Render the decision brief, evidence trail, review artifact, dry-run packet, and predicted-versus-actual outcome view.

Generic AI copilots produce answers. Auctorian coordinates governed agents to produce one auditable, review-ready decision.

Evidence-backed · guardrailed · approval-ready
Cross-decision orchestration

Decisionsdo not live alone.

A price move changes demand. Demand strains logistics. Running an asset harder moves its maintenance window. Auctorian coordinates decisions as one system — six sensible calls can still add up to a mess.

Eight domains · five couplings shown

01PricingInventoryvia expected unit velocity

A price move changes how fast a unit leaves, so every replenishment point downstream of it is answering to a demand curve that no longer holds.

02DemandLogisticsvia localised demand spike

Committed demand lands in particular places, which makes a distribution split that was balanced yesterday the wrong one today.

03InventoryDemandvia coverage shortfall

When cover falls below the horizon, demand generation is spend against supply that will not arrive — so the correct demand decision is to stop creating it.

04ProcurementPortfoliovia supplier reliability

A fill rate or defect rate that breaks its band is evidence about the supplier, and it changes which lines deserve a place in the portfolio at all.

05CapacityMaintenancevia utilisation strain

Running an asset harder brings its service interval forward, so a maintenance window booked for next quarter becomes this quarter’s ceiling on output.

See it compilea decision of yours.

Advisory and dry-run first. Every figure traceable to source, nothing dispatched without you.