Every unfamiliar object resolves backward.

Fractal Custody Object: a content-addressed logical custody object with explicit evidence and claim boundaries.

How to: Inspect an FCO ID, its single canonical object SHA-256, payload, evidence class, claim ceiling, and source/dependency edges rather than treating a hash as truth.

knowledge_fco=fco:63031d39a118f3fee64c3834db9997c724ea86a696c8a9e11040daabb65e7f10

object_sha256=63031d39a118f3fee64c3834db9997c724ea86a696c8a9e11040daabb65e7f10

graph_query=FCO

Fractal Custody Graph: the dependency and transformation graph connecting sources, evidence, derived state, claims, and artifacts.

How to: Follow the canonical project relationship vocabulary to locate the source and transformation route behind a result; do not infer correctness merely because a path exists.

knowledge_fco=fco:78fe5323bd6ba1b1a7d0e36180ff86cf1152a44c35c3e3d783da72aa90d89bbc

object_sha256=78fe5323bd6ba1b1a7d0e36180ff86cf1152a44c35c3e3d783da72aa90d89bbc

graph_query=DERIVED_FROM

Governed source/evidence intake layer used to preserve content identity, provenance, and structured atomization before downstream projection.

How to: Use SeedGraph through its governed intake/ledger surface and preserve exact-source identity before downstream transforms.

knowledge_fco=fco:7e6e233a1c2f533b3eb9af61ff66b33b76e9621cdc4af2a3ff66eefdb092872d

object_sha256=7e6e233a1c2f533b3eb9af61ff66b33b76e9621cdc4af2a3ff66eefdb092872d

graph_query=SeedGraph

A bounded context-bearing evidence fragment produced from an exact source span or deterministic transform.

How to: Open the source/evidence node first, then follow its derivation into the atom; do not collapse source bytes and semantic interpretation into one identity.

knowledge_fco=fco:e7b75364b6ba51c11947057d12d96fada44309bf0d5d8d41a1e40656d3ea1f7e

object_sha256=e7b75364b6ba51c11947057d12d96fada44309bf0d5d8d41a1e40656d3ea1f7e

graph_query=KnowledgeAtom

A bounded admitted proposition supported by explicit atoms/evidence; not an absolute-truth label.

How to: Follow support dependencies back to admitted atoms and inspect the claim ceiling before reusing the seed.

knowledge_fco=fco:3ef3e40d84440693677cc4784794c40650e765a42525d1d9293bafc010b4fb51

object_sha256=3ef3e40d84440693677cc4784794c40650e765a42525d1d9293bafc010b4fb51

graph_query=SeedOfTruth

Mechanical Scientific Method: executable observation → hypothesis → experiment → measurement → comparison/falsification → update.

How to: Use declared controls and perturbations so the next experiment is a mechanically inspectable state transition and preserve null/negative outcomes.

knowledge_fco=fco:edba2f10ead9255ca2a776abcaf4309e9c622df8c98e282ddf6adda155c5e6ea

object_sha256=edba2f10ead9255ca2a776abcaf4309e9c622df8c98e282ddf6adda155c5e6ea

graph_query=StateSnapshot

Independent SELF/NONSELF × SAFE/NONSAFE + UNKNOWN admission/classification surface.

How to: Treat classification as a scoped event. It may admit, challenge, quarantine, abstain, or reject without overwriting provenance.

knowledge_fco=fco:994abc0e6927470000e22eae1d7a4a6f280eda0e0225670f9ed7eb0e22528090

object_sha256=994abc0e6927470000e22eae1d7a4a6f280eda0e0225670f9ed7eb0e22528090

graph_query=ClassificationReceipt

Temporal edge preserving an older authentic state while pointing toward a later state.

How to: Traverse the declared current-state chain rather than deleting the predecessor; the historical state remains part of custody.

knowledge_fco=fco:ee88b6fb9adc2d3cc224a7e29f059fc5252607902aa2364f7cd3dae8f8e1844a

object_sha256=ee88b6fb9adc2d3cc224a7e29f059fc5252607902aa2364f7cd3dae8f8e1844a

graph_query=SUPERSEDED_BY

Explicit relation between incompatible values or claims retained simultaneously in the graph.

How to: Use contradiction as a retrieval/reasoning signal and preserve both sides with their source/provenance context.

knowledge_fco=fco:6a546f3d6a5bb64370458f688e186ab1cb32a054262e0f03f2a997008290b956

object_sha256=6a546f3d6a5bb64370458f688e186ab1cb32a054262e0f03f2a997008290b956

graph_query=CONTRADICTS

The state selected by the declared temporal/supersession rules at a specific evaluation time, not an erasure of prior states.

How to: Resolve current state from the relevant state-transition path and preserve the predecessor, contradiction, and source chain.

knowledge_fco=fco:4fc3fb9b9dafbd9dd0c6934b6015ac68246e777630781ef7732a05993701f2e5

object_sha256=4fc3fb9b9dafbd9dd0c6934b6015ac68246e777630781ef7732a05993701f2e5

graph_query=current_state

A declared intervention that changes a load-bearing input, value, edge, configuration, or state for a falsification test.

How to: Record the target, before/after state, operator, scope, expected effect, and FCG delta. A perturbation is not automatically poison or unsafe.

knowledge_fco=fco:96417c0db002e84e99ae63067f9663607220ebbbc36e0aecb69fe8c1d54a35a4

object_sha256=96417c0db002e84e99ae63067f9663607220ebbbc36e0aecb69fe8c1d54a35a4

graph_query=Perturbation

A declared recovery intervention intended to restore a reference-compatible state without deleting perturbation history.

How to: Apply restoration as a new state, then compare the current-state and outcome metrics against the frozen reference while retaining the poison path.

knowledge_fco=fco:dac354d92fe2f156cd751df875f6d25431acbddf0b0c3013903fda00d029cd76

object_sha256=dac354d92fe2f156cd751df875f6d25431acbddf0b0c3013903fda00d029cd76

graph_query=restoration

HydraDG presentation model that places ΔG* and Cloud Drift at the visible tip while retaining the full FCO/FCG, provenance, statistical and receipt decomposition below the waterline.

How to: Treat the hero as a read-only projection. Rotate x/y/z, scrub time, inspect node halos, then follow the selected FCO into its custody route rather than treating the visualization as the evidence source.

knowledge_fco=fco:6ae3f5b4d2cb879048be066852699420422fd542c8e4a05b5234706b699bfa41

object_sha256=6ae3f5b4d2cb879048be066852699420422fd542c8e4a05b5234706b699bfa41

graph_query=StateSnapshot

Bounded context-distribution drift magnitude defined in v1 as 100 × Jensen-Shannon divergence from a frozen reference distribution.

How to: Use halo width/size to visualize magnitude only. Do not interpret a larger or smaller cloud as better accuracy; compare hit/recall and the preregistered nulls separately.

knowledge_fco=fco:2b60bce138cf35e15a6b2780e57449681eca66283610a090a9718ccbbd733369

object_sha256=2b60bce138cf35e15a6b2780e57449681eca66283610a090a9718ccbbd733369

graph_query=cloud_drift_0_100

Symmetric bounded divergence between two probability distributions; HydraDG uses base-2 logs so the v1 value lies in [0,1] before scaling to Cloud Drift 0–100.

How to: Verify the reference distribution, bucket vocabulary and scorer/config roots were frozen before comparing states. Source lineage: Lin (1991), Divergence measures based on the Shannon entropy.

knowledge_fco=fco:7fa58bb310fedd8a90a5214d9127fca7acc5d596f1dae4e105b3d28028d8c409

object_sha256=7fa58bb310fedd8a90a5214d9127fca7acc5d596f1dae4e105b3d28028d8c409

graph_query=js_divergence

Shannon entropy H = -∑ p log₂ p: information-state uncertainty / distribution breadth across the declared state distribution.

How to: HydraDG also uses Hnorm = H / log₂(n) in the synthetic G* scorer. Source lineage: Shannon (1948), A Mathematical Theory of Communication.

knowledge_fco=fco:6dce815c214c19579a610493d1b04e91da72ab1b30824b533bb419d350de6d03

object_sha256=6dce815c214c19579a610493d1b04e91da72ab1b30824b533bb419d350de6d03

graph_query=StateSnapshot

HydraDG synthetic-fixture scalar: G* = U* - 0.35 × Hnorm, where U* is the declared burden input and Hnorm is normalized Shannon entropy.

How to: Treat G* as an application-defined dimensionless information-state diagnostic. Enßlin & Weig (2010) is design-rationale lineage for the information/free-energy analogy; it does not define HydraDG's exact formula. No physical Gibbs-energy units are claimed.

knowledge_fco=fco:92f319f1b3229895d587f6b4fb1b1f856104dd3fbb2624ce07821bd75686b56b

object_sha256=92f319f1b3229895d587f6b4fb1b1f856104dd3fbb2624ce07821bd75686b56b

graph_query=StateSnapshot

Signed stepwise change in the HydraDG synthetic-fixture scalar: ΔG*(t) = G*(t) - G*(t-1).

How to: Interpret sign/direction only within the declared scorer contract. ΔG* is not physical Gibbs free energy and is not an accuracy score; Lin (1991) applies to JSD/Cloud Drift, not to this ΔG* formula.

knowledge_fco=fco:17443bb87fa0d7f0304387d6e7c534a201e751dcf5f8f9241b3a4212d058a4d9

object_sha256=17443bb87fa0d7f0304387d6e7c534a201e751dcf5f8f9241b3a4212d058a4d9

graph_query=delta_g_star

Synthetic-fixture mutation distance is total-variation distance between the frozen reference distribution and the current distribution.

How to: Keep this lane separate from Cloud Drift/JSD. In the checked scorer, total-variation distance is 0.5 × Σ|p_ref - p_t| after normalization.

knowledge_fco=fco:f0d40a084e8d4dfa98e1f50aa328e004d5467824af25247a34317854c5266d8e

object_sha256=f0d40a084e8d4dfa98e1f50aa328e004d5467824af25247a34317854c5266d8e

graph_query=mutation_distance

Synthetic-fixture restoration gain = max(0, previous total-variation distance - current total-variation distance).

How to: A positive value means movement back toward the frozen reference in the declared TV-distance lane. It is not ΔJSD and does not erase the poison state.

knowledge_fco=fco:c423544fb6a03abea6304f58742bf67d8e90eb5ebf803097208e4af9172de2fb

object_sha256=c423544fb6a03abea6304f58742bf67d8e90eb5ebf803097208e4af9172de2fb

graph_query=restoration_gain

Declared normalized burden input U* ∈ [0,1] used by the synthetic G* fixture contract.

How to: For the current fixture, U* is explicitly supplied for each state; do not claim it was inferred from contradiction-edge density unless a separate governed scorer and receipt establishes that mapping.

knowledge_fco=fco:67d7eb8f8cb9d49dd2eac9f7330255169b21fec8c5e583598046032be9d637ec

object_sha256=67d7eb8f8cb9d49dd2eac9f7330255169b21fec8c5e583598046032be9d637ec

graph_query=burden

Explicit classification of how evidence was obtained, such as directly supplied, externally retrieved, recomputed, deterministic transform, probabilistic model output, or inference.

How to: Keep evidence class attached to the object/result so downstream claims cannot silently promote a model output into a measured finding.

knowledge_fco=fco:2d6264ee14e215979d85a937bfdc9d323a5891424d7d4a19d7cf286d9a8d467f

object_sha256=2d6264ee14e215979d85a937bfdc9d323a5891424d7d4a19d7cf286d9a8d467f

graph_query=evidence_class

Maximum assertion strength supported by the current load-bearing evidence dependencies.

How to: A hash, replay, model output, or signature can only promote the specific property it actually establishes.

knowledge_fco=fco:f81b0bdca0c26b03161230758ec61f6e79dacbfb547c2cfe7a7e8635180937af

object_sha256=f81b0bdca0c26b03161230758ec61f6e79dacbfb547c2cfe7a7e8635180937af

graph_query=claim_ceiling

Deterministic commitment over a declared ordered leaf set; distinct from signing, correctness, and live HydraDB graph state.

How to: Inspect leaf ordering, odd-leaf rule, root SHA-256, and exact scope before describing a checkpoint as committed.

knowledge_fco=fco:9f7320fa4e7decb810e91f7726baedc73088f6dfdf2e5e1ee18e899e83fdf472

object_sha256=9f7320fa4e7decb810e91f7726baedc73088f6dfdf2e5e1ee18e899e83fdf472

graph_query=MerkleCheckpoint

Operational graph/context substrate used by HydraDG for typed state, relations, traversal, retrieval, connector-backed context and hosted projection/readback.

How to: Use the pinned/local lane for reproducibility and the hosted hydradg database for separately evidenced production readback; canonical FCO SHA-256 identity remains backend-independent.

knowledge_fco=fco:1bd50750c2a6b73f55779f77325dae25308f52ed93318b778ae02e6a8967dd8c

object_sha256=1bd50750c2a6b73f55779f77325dae25308f52ed93318b778ae02e6a8967dd8c

graph_query=HydraDB

Primary public Track 03 memory benchmark used by the completed full500 run.

How to: Use answer_session_ids for evaluation only; never graph construction or ranking.

knowledge_fco=fco:ff894ecb2967e9bf0ee69e0e9c348a0a684ce97b50f1fd8fa523e9746dd6178a

object_sha256=ff894ecb2967e9bf0ee69e0e9c348a0a684ce97b50f1fd8fa523e9746dd6178a

graph_query=LongMemEval

Primary Track 01 synthetic enterprise retrieval/reasoning corpus.

How to: Model source artifacts, entities, conflicting/stale claims, current state, and evidence paths; keep benchmark data evaluation-only.

knowledge_fco=fco:950142db44c8063d40a9983b3823473611a1ad991c439d6ef4dc3204bb631f3c

object_sha256=950142db44c8063d40a9983b3823473611a1ad991c439d6ef4dc3204bb631f3c

graph_query=EnterpriseRAG-Bench

Heterogeneous enterprise deep-search benchmark used as the Track 01 stress dataset.

How to: Use as a private/local evaluation source unless the intended public use passes the CC-BY-NC-4.0 release-license review.

knowledge_fco=fco:8f5e36ea184ff852eed61453948081a9d9b33fecf04693e9a0e8ec31c31c7938

object_sha256=8f5e36ea184ff852eed61453948081a9d9b33fecf04693e9a0e8ec31c31c7938

graph_query=HERB

Track 03 environment-experience memory benchmark with trajectories, haystacks, and optional screenshots.

How to: Start with released question/trajectory/haystack objects; pull screenshot archives only when the visual evidence lane is needed.

knowledge_fco=fco:760165c3f7d08e40b9e0a980ed1d43723284f8fbdb64c8031c5e1baaef27c992

object_sha256=760165c3f7d08e40b9e0a980ed1d43723284f8fbdb64c8031c5e1baaef27c992

graph_query=LongMemEval-V2

Long-context memory benchmark across 100K/500K/1M and separately 10M-token-scale conversations.

How to: Keep probing questions/ideal answers evaluation-only and ingest conversation turns as the memory substrate.

knowledge_fco=fco:31b87dcb8787709c0fcc4f805899bb4dcf09310f0b1c14305569abb5746a3a6a

object_sha256=31b87dcb8787709c0fcc4f805899bb4dcf09310f0b1c14305569abb5746a3a6a

graph_query=BEAM

Track 01 HydraDG experiment for enterprise context, identity resolution, conflicting evidence, and ontology state.

How to: Use bounded entity-resolution canaries first; only promote to real benchmark claims after the corresponding EnterpriseRAG/HERB receipts exist.

knowledge_fco=fco:da497052c6158548663924c4358481e292cb527719b65873803cb6cde57a4c22

object_sha256=da497052c6158548663924c4358481e292cb527719b65873803cb6cde57a4c22

graph_query=HydraOntology

Track 02A HydraDG experiment for dependency blast radius and poison → partial repair → full repair traversal.

How to: Compare HydraDB reverse dependency traversal against an independent deterministic closure before using real package/advisory evidence.

knowledge_fco=fco:3ff94610b84e6545cd7dadd43c7d619f35709dfca46efab3348f82d5aed4f8e6

object_sha256=3ff94610b84e6545cd7dadd43c7d619f35709dfca46efab3348f82d5aed4f8e6

graph_query=HydraBlast

Track 03 HydraDG experiment for longitudinal memory, supersession, contradiction, retrieval, and recovery.

How to: Preserve the completed null/negative full500 retrieval result while separately executing fresh state-transition golden paths and future governed treatment comparisons.

knowledge_fco=fco:b9ee524e8438b608c01f3efd147a1412f90a5ec3c8ee6123c93921b05d8ff432

object_sha256=b9ee524e8438b608c01f3efd147a1412f90a5ec3c8ee6123c93921b05d8ff432

graph_query=HydraMemory

Content-addressed submission hero media FCO with sponsor/track annotation regions for Immersive Commons judge navigation.

How to: Open /submission, inspect the SubmissionHeroMediaFCO by hash, and follow golden-path steps without treating the image as canonical FCG append evidence.

knowledge_fco=fco:47aa0f3a9f4313043fe93f0ce6512436f164c3772a54b7669910a5b23802b850

object_sha256=47aa0f3a9f4313043fe93f0ce6512436f164c3772a54b7669910a5b23802b850

graph_query=SubmissionHeroMediaFCO

HydraDG software is Apache-2.0; FCO/FCG research content and designated submission media are CC BY-NC-ND 4.0.

How to: Read LICENSING.md before reusing artifacts. License metadata is rights declaration only — not a hash, signature, or scientific verification result.

knowledge_fco=fco:7a3a6448b05c7c26429894b8ee722e351f8828a9465153578b82ee0d0c1e2bbf

object_sha256=7a3a6448b05c7c26429894b8ee722e351f8828a9465153578b82ee0d0c1e2bbf

graph_query=license