Public Method

How DAEPOLIS Thinks

DAEPOLIS publishes its accountability outputs, not the full internal recipe. This page explains the public trust surface: how records become traceable, how predictions become falsifiable, and where the licensed operating method begins.

Method boundary

Public enough to audit. Protected enough to remain defensible.

The public site shows commitments, ledger behavior, calibration outcomes, and example records. It does not publish the full scoring manuals, pattern criteria, prompt stack, or workstation operating playbooks. Commercial or institutional use of those materials requires a methodology licence.

Current public operating surface: GST v1.3; CJI v1.4 where the CJI layer is applied.

Methodology licence

Public Method Commitments

These are the visible constraints every DAEPOLIS output is expected to satisfy. They are governance commitments rather than implementation instructions.

Traceability

Every substantive claim must preserve a source path, evidence status, and review state before it can become institutional intelligence.

Claim trace audit

Existing prose can be inverted into a belief architecture: material claims, warrants, implied priors, belief debt, adversarial objections, falsifiers, and questions before reliance.

Institutional dependency intelligence

Risk-bearing relationships are mapped as dependencies only when a material pathway exists, with reputational, commercial, legal/regulatory, operational, evidentiary, and decision materiality kept separate.

Falsifiability

Forecasts need a probability, resolution date, observable criteria, and falsifier. Claims without those conditions remain monitoring claims.

Pre-registration

CJI class predictions can be sealed before resolution with a commitment hash, panel hash, payload hash, probability, and source-of-record criterion.

Calibration

Resolved forecasts are scored over time. Qualitative triggers are tracked separately so they do not pollute numeric Brier performance.

Contestability

Major outputs keep rival hypotheses, evidence gaps, misuse risks, and repair paths visible to reviewers.

Accountability Layer

The audit surface is deliberately concrete. Users should be able to inspect what was claimed, what would prove it wrong, what sources carried weight, and how the forecast record changes over time.

Prediction register

Public commitments, resolution windows, falsifiers, evidence notes, and Brier outcomes.

Brief forensics

Reverse Bayesian Claim Tracing reconstructs consultant reports, AI memos, ESG briefs, policy papers, board papers, and investment notes into claims, warrants, priors, evidence weights, belief debt, and a five-part adversarial reliance audit before an institution relies on them.

Dependency mapping

Institutional Dependency Intelligence uses Dependency Mapping & Materiality Analysis to separate exposure from dependency, end-user sensitivity from commercial materiality, known evidence from missing evidence, and orientation-level claims from decision-ready risk.

Source discipline

Source proximity, provenance, reliability, bias risk, and degradation signals are treated as first-class review objects.

Ledger memory

Analyses, claims, patterns, forecasts, and framework-validity observations become versioned records instead of loose prose.

CJI validation

Retrospectives expose method limits; sealed CJI pre-registrations are the predictive-credit surface.

Method versioning

Public outputs name the operating method version while detailed scoring rubrics remain controlled implementation material.

Open prediction ledger Open CJI validation Open CJI commitments

Method-To-Interface Contract

The frontend should expose the backend spine.

DAEPOLIS is no longer just an analysis form. The backend now produces record objects: ingestion reports, source states, claim ledgers, prediction distributions, decisive unknowns, review events, outcome memory, trust capsules, and Surface deltas. The product surface has to make those objects visible, inspectable, and usable.

The quality rule is simple: if the backend can prove, warn, score, repair, or remember something, the frontend should not hide it behind a generic text box.

01Intake Integrity

Documents enter as untrusted evidence, not authority.

PDF/Office extraction integrityUntrusted-content boundarySource packsHardening tasks

Show extraction quality, provenance, suspicious content, and review work before any audit treats the document as usable evidence.

02Claim And Evidence Core

Claims are separated from prose and kept provisional until supported.

Canonical claim ledgerSource-hardening statesSemantic collisionsEvidence gaps

Expose the support state, missing source class, contradiction path, and next hardening action beside the claim itself.

03Judgement Memory

A record keeps learning after the first answer.

Decision RecordsAppend-only eventsReview workflowOutcome ledger

Let users see who changed what, why the record moved, what remains disputed, and what later happened.

04Forecast And Calibration

Probability is a public object, not a sentence.

Prediction distributionsReference classesRed-team sidecarsCalibration surface

Show the distribution, outside-view class, adversarial disagreement, decisive unknowns, and update history instead of only one midpoint.

05Monitoring And Repair

The system should wake when reality changes.

Decisive unknownsMonitoring triggersLineage decayError memory

Turn uncertainty into watchable signals: what would change the judgement, what has gone stale, and what prior mistake should be reused.

06Surface Instrument

The graph is an operating surface over the record spine.

Trust capsulesRole-specific viewsSurface eventsSignal intake

Every node should answer why it matters, why it can be trusted, what is missing, what changed, and what role should do next.

Probability And Calibration

DAEPOLIS separates live analytical judgement from later scoring. The point is not to sound certain. The point is to leave a record that reality can punish.

Before resolution

Commitment

Forecasts require observable conditions, time bounds, probability distribution, confidence, reference-class context, red-team disagreement, decisive unknowns, and explicit falsifiers before publication.

At resolution

Scoring

Numeric predictions receive Brier scoring and calibration-bin treatment. Distribution quality and qualitative mechanism validations remain inspectable instead of being collapsed into a single confidence line.

After resolution

Repair

Resolved misses become error-attribution memory for source weighting, base-rate discipline, lineage decay, Bayesian feedback, and future quality gates.

Protected Method Layers

The following materials are not published as open implementation notes. They are controlled because they encode the operating advantage of the workstation.

Full CJI indicator operationalisation and scoring rubrics

Complete structural-pattern taxonomy and acceleration rules

Prompt architecture, repair loops, and adversarial audit mechanics

Base-rate reference classes and calibration thresholds

Brief-production playbooks, templates, and recipient-calibration manuals

GST v1.3 closed-loop reliability, lineage, and entity-resolution internals

What Remains Public

DAEPOLIS still makes the parts that matter for public accountability visible.

Public trust surface

Ledger, briefs, and calibration

Readers can inspect published briefs, prediction commitments, resolved outcomes, evidence summaries, and retrospective corrections.

Open public briefs

Licensed surface

Operating method and workstation logic

Institutional users can licence the full methodology, governance workflow, scoring manuals, and controlled implementation materials.

Discuss licensing