How to use HydraDG

One path from context change to custody.

Follow the numbered path. Every public step is read-only against the judge release; each canonical FCO resolves to one SHA-256 identity.

Check deployed version + release FCO

1 · Overview

Read the Context Iceberg

Violet is reference/normal, orange is poison/mutation, blue is antidote/restoration. Read ΔG* and Cloud Drift separately from Hit@K and Recall@K.

Open Overview

2 · Judge Demo

Reference → poison → antidote

Use the walkthrough table to confirm exact distributions, G*, ΔG* and Cloud Drift for the synthetic T0–T2 fixture.

Start Judge Walkthrough

3 · Graph

Trace one canonical FCO

Select a node, open its FCO inspector, and verify the FCO ID matches its single object SHA-256.

Open 4D Graph

4 · Atom

Resolve a KnowledgeAtom

Start from exact source/evidence, follow DERIVED_FROM into the atom, and keep semantic interpretation distinct from source bytes.

KB: KnowledgeAtom Graph atoms

5 · Seed

Resolve a SeedOfTruth

Follow SUPPORTED_BY edges back to admitted atoms/evidence and inspect the claim ceiling before reuse.

KB: SeedOfTruth Graph seeds

6 · Results

Read the empirical outcome

Inspect cases, graph scale, Hit@K, Recall@K and the retained null/negative result under its declared claim ceiling.

Open Track 03 Results

7 · Knowledge

Resolve project terminology

Use the governed Knowledge Base for FCO, FCG, Context Iceberg, G*, ΔG*, Cloud Drift, total-variation mutation distance and restoration gain.

Open Knowledge Base

8 · Release

Check version and custody state

The global release stamp shows the deployed Git SHA and WebsiteRelease FCO hash. Signature and Merkle states remain separate.

Release JSON Version history Eligibility

Hosted path · GitHub → HydraDB → Vercel

Connect once. Query the indexed graph from the judge site.

The hosted demo uses HydraDB as a server-side context service. Credentials never reach the browser. Canonical FCO/FCG identity remains backend-independent.

A · Source

GitHub repository → database hydradg

The GitHub connector indexes repository data in the hosted hydradg database. The verified canonical projection/readback receipt uses collection default.

B · Vercel

Server-side configuration

Use HYDRA_DB_API_KEY, HYDRADB_DATABASE=hydradg, HYDRADB_API_URL=https://api.hydradb.com, and the verified collection scope when required.

C · Query

Graph context + connector apps

The v2 query route uses graph_context=true and query_apps=true by default so connector-backed GitHub context can be returned alongside graph context.

D · Verify

Separate parity from service drift

Canonical FCO/edge/content roots are compared independently from retrieval ranking, latency, graph paths, Hit@K and Recall@K.

Live HydraDB status Collections readback

Interpretation rules

Cloud Drift = 100 × base-2 Jensen-Shannon divergence from the frozen reference distribution.
G* = U* - 0.35 × normalized Shannon entropy for the declared synthetic fixture.
Mutation distance and restoration gain use the separate total-variation lane.
G*/ΔG* are dimensionless information-state abstractions, not physical Gibbs free energy.
T3–T5 project/release states receive scalar score N/A when no distribution is declared; T2 is never copied forward.
HydraDB is a queryable projection/context substrate; FCO/FCG is the canonical custody/provenance layer.