Reopenable claim-state graph

Epistemic Audit Graph

Turns disputes into actionable evidence challenges by managing contested claims as reopenable states.

Open Claim Cascade Claim Lifecycle Workspace View repository

proposed

contested

weakened

parked

reopened

Five-minute click path

Start from this hub, then click these routes in order if you want the shortest walkthrough.

1. Open Claim Cascade Feel how a simple question becomes a parked claim and an evidence challenge. 2. Claim Lifecycle Workspace See source wording, inference, parked state, and reopening conditions as a workspace. 3. K-Pg graph Use a familiar domain example to inspect accepted, contributing, contested, and open claims. 4. Park-condition propagation See why uncertainty travels from evidence through inference into a claim.

Start with these two

Use the first demo to feel the idea, then the workspace to see the same pattern as a claim-state system.

1. First touch Claim Cascade The first-touch interactive demo: a casual Y/N question becomes a parked claim, a graph, structure cards, and concrete next actions. 2. Structure Claim Lifecycle Workspace A direct workspace view of source wording, inference, downgrade, parked state, reopening conditions, and the claim challenge board.

Examples and implementation proof

These routes are useful after the first two. They show domain examples, validation data, and deeper prototype mechanics.

Claim lifecycle dashboard Six-domain validation cases with source wording, inference risk, downgrade checks, and reopening questions. K-Pg claim-state graph Flagship source-audited graph for accepted, contested, unresolved, and parked claims. Bronze Age claim-state graph A more accessible historical route through evidence challenges and claim state changes. Lifecycle timeline prototype Timeline and relation views for claim states, challenge nodes, and non-destructive downgrading. Park-condition propagation A one-screen explanation of how parked evidence and inference assumptions propagate into a parked claim.

Expected Q&A

Short answers for common questions about prior work, AI authority, LLM use, reopening, and boundaries.

Is this completely new from scratch? No. It overlaps with defeasible reasoning, non-monotonic reasoning, truth-maintenance systems, Wikidata-style statements, ClaimReview, and argument mapping. The test is the combination as a practical claim-lifecycle workflow.

Is this just Kialo or argument mapping? It overlaps, but the focus is different: claim lifecycle management, non-destructive downgrading, reopening conditions, and evidence challenges rather than arranging pro/con positions.

Is this just Wikidata or ClaimReview? Those structure claims and references. Epistemic Audit Graph adds explicit claim states, downgrade reasons, reopening conditions, inference risk, and propagation from evidence to derived claims.

Does AI decide what is true? No. The AI may draft or structure nodes, but humans inspect the claim state. The validator checks structural rules and overclaim patterns; it is not a truth oracle.

Could this help LLM memory or retrieval? That is a future hypothesis. Machine-readable claim states could let an LLM retrieve not only a claim, but also its status, downgrade reason, uncertainty, and reopening conditions.

Can LLMs handle research papers reliably? Not as standalone truth judges. But if constrained to a retrieved source library, an LLM may help compare source wording, flag conflicts, detect overclaims, and draft evidence challenges for human review.

Who decides whether a claim can reopen? Some reopening conditions can be structurally checked, such as whether required evidence fields exist. Substantive judgment remains a human or community review decision.

Does parking preserve every hypothesis forever? No. Parking is bounded by reliable sources, no original research, due weight, and community judgment. A claim can be kept out of article space while its reason and challenge remain inspectable.

What is the WikiCred feedback ask? Which workflow should be the first serious test: disputed section audits, source-lineage maps, Talk page summaries into claim states, or synthesis/original research/undue-weight risk flags?

What question should this raise? Is there already a tool or workflow in the Wikimedia / WikiCred ecosystem that handles this kind of claim lifecycle well?

Short novelty answer: this is not a from-scratch invention; it is a practical combination of public claim-state management, non-destructive downgrading, reopening conditions, actionable evidence challenges, uncertainty propagation, source-conflict inspection, and possibly machine-readable claim-state memory for LLM-assisted workflows.

Why This Is Familiar

In frontline sales work and engineering work, people constantly separate what was actually said, what was inferred, what is still only a hypothesis, and what needs confirmation.

In sales, mixing a customer's actual words with internal assumptions can cause real mistakes. In engineering, requests, specifications, hypotheses, tests, issues, reviews, and rollback all need to be separated and managed as states.

But in public knowledge work, sources, interpretations, assumptions, counterevidence, and unresolved questions often collapse back into prose. In Wikipedia articles, papers, policy documents, and AI outputs, different knowledge states can become compressed into one fluent explanation.

Epistemic Audit Graph tries to bring the practical discipline of frontline work and engineering into the public knowledge layer: separating what was said, what was inferred, what remains uncertain, and what conditions would allow a claim to be reopened or strengthened.