Judge Demo · public read-only walkthrough · Golden path

Change state. Keep the history.

HydraDG makes reference → poison → antidote explicit graph transitions, then keeps migration, experiment and release states as separate FCG timepoints. The judge path separates deterministic evidence, probabilistic model interpretation and unexecuted future comparisons.

Start with reference Executed Track 03 result Models used

Public data plane

Hosted HydraDB not connected on this deployment

DEPLOYMENT NOT CONFIGURED

BackendCHECKING

DatabaseNOT_CONFIGURED

CollectionNOT_CONFIGURED

TraceabilityNOT_CONFIGURED

This card is only a live server-side connectivity/readback diagnostic for the current deployment. HydraDG's retained local HydraDB executions, graph use case, deterministic math and bounded hosted receipts are separate evidence objects. Credential values are never returned to the browser.

Why HydraDB · math · implementation Open retained evidence

01 / GOLDEN PATH

Reference → poison → antidote.

Gold marks the judge navigation path only. Violet = reference/normal, Orange = poison/mutation, Blue = antidote/restoration. Those scientific state colors do not encode pass/fail or retrieval accuracy.

01 · CURRENT

Reference

Begin with one declared current fact and its source/evidence path.

Why: Freeze the comparison state before perturbation.

02 · CONTRADICTS / SUPERSEDED_BY

Poison

Introduce one controlled conflicting state while retaining predecessor provenance.

Why: Expose the first divergent relationship without overwriting history.

03 · RESTORES / CURRENT

Antidote

Represent recovery as a new state while the perturbation remains traversable.

Why: Test restoration without deleting counterevidence.

02 / WHAT USED A MODEL?

The submitted K=5 experiment did not.

Track 03 full500 K=5

MODEL = NONE

extractor=heuristic · model=null · ollarma_url=null

Primary executed retrieval evidence.

Local analyst lane

Qwen2.5 7B + Qwen2.5-Coder 7B

Optional loopback diagnostic models after deterministic output freeze. Their responses are probabilistic model-output FCOs.

Frontier-cloud comparison

NOT CONTROLLED / NOT RUN

No local-vs-frontier superiority claim is established. Synthetic design rows are not model-execution evidence.

Open full model protocol + HF cards

03 / TIMEPOINT EVIDENCE LANES · T0–T5

Do not copy a score into a state that never declared one.

T0–T2 use declared synthetic distributions for G*, ΔG* and Cloud Drift. T3–T5 have no declared scoring distribution and instead carry bounded migration, classification and release evidence.

Timepoint	Classification	Scientific score	Evidence
T0_REFERENCE Reference / normal	SYNTHETIC_FIXTURE	*G: -0.061230** ΔG*: 0.000000 Cloud Drift: 0.0000	Declared deterministic fixture baseline.
T1_MUTATION Poison / mutation	SYNTHETIC_FIXTURE	*G: 0.572956** ΔG*: +0.634186 Cloud Drift: 40.3629	Declared controlled perturbation fixture.
T2_RESTORATION Antidote / restoration	SYNTHETIC_FIXTURE	*G: -0.027496** ΔG*: -0.600452 Cloud Drift: 1.8729	Declared restoration fixture; counterevidence remains in the graph.
T3_HOSTED_MIGRATION Hosted HydraDB migration	PRODUCTION_RELEASE_STATE	N/A BY CONTRACT	Hosted parity receipt: 36 canonical FCOs, 24 edges, zero canonical set/edge/content-hash delta.
T4_CONTEXT_VS_ENTROPY Context vs Entropy	PRODUCTION_EXPERIMENT_STATE	N/A BY CONTRACT	18,567 raw findings; 18,555 context-classified; 12 abstentions; 99.9354% coverage.
T5_FINAL_JUDGE_RELEASE Final judge release	PRODUCTION_JUDGE_RELEASE_STATE	N/A BY CONTRACT	Exact deployed Git SHA and WebsiteRelease FCO are resolved at build/runtime; no scalar G*/Cloud Drift is fabricated without a declared distribution.

The retained T3 hosted-parity statement is bounded to the historical receipt that produced it. It must not be generalized to expanded current FCG parity without a fresh reconciliation receipt.

04 / JUDGE LENSES

Primary Track 03; architecture inspected from all three graph problems.

Track 01 · Enterprise context + ontology

Identity resolution

Inspect canonical entities, aliases, provenance and contradiction handling. Core data is downloaded/hashed; real-data ontology ingestion/evaluation remains pending.

Track 02 · Repos + dependencies

Blast-radius traversal

Inspect transitive graph reasoning and fail-closed canary boundaries. Real npm/dependency evaluation remains pending.

Track 03 · Primary submission

Memory + context retrieval

500-case full500 run executed; null/negative retrieval result retained. This is the submitted empirical lane.

05 / OFFICIAL HACK HYDRA JUDGING CRITERIA

Make the evaluation surface explicit.

01Technical execution

02HydraDB + graph-native use

03Product completeness + usability

04Quality of results

05Originality

Hack Hydra evaluates submissions within their selected track first and then compares finalists across tracks; Best Use of HydraDB is a separate lens. Official hackathon page.

06 / CONTINUE THE GOLDEN PATH

Follow the evidence, not a sales claim.

Read executed result Trace one FCO Models used Verify custody Evidence ledger Static fallback

The public walkthrough is read-only. It demonstrates declared state-transition and custody contracts; it does not claim that clicking presentation states mutates a judge-visible HydraDB tenant.