Technical documentation on AnankeNet

Information layers

Mon, 01 Jan 0001 00:00:00 +0000

AnankeNet keeps six information layers separate. Each layer has a different source, meaning, and level of certainty.

Layer	Meaning	Must not be presented as
Evidence	Preserved source material and its recorded origin	An analytical conclusion
Observation	A fact extracted directly from evidence	A probabilistic inference
Derived	A value calculated from observations	A source fact
Analysis	An explainable signal or relationship	Proof of intent or conduct
Hypothesis	A possible explanation requiring review	An established conclusion
Human review	An accountable assessment of the record	Automatic ground truth

Why the distinction matters

A numerical score can appear precise while remaining uncertain. A relationship can be real while its meaning is unknown. A hypothesis can be useful while still being wrong.

Evidence integrity

Mon, 01 Jan 0001 00:00:00 +0000

Evidence preservation has priority over analytical sophistication. Analysis can be repeated; an original source that has been overwritten cannot be recovered with the same confidence.

Preservation sequence

The public integrity contract is straightforward:

capture lawfully accessible public material;
record its source and canonical UTC capture time;
calculate a cryptographic digest;
seal the original representation;
analyze a separate copy or derived representation;
verify integrity before review or export.

SHA-256 is the default digest for captured artifacts. A digest can make later modification detectable, but it does not prove that the original source was true or that collection was authorized.

Provenance and audit

Mon, 01 Jan 0001 00:00:00 +0000

Provenance answers a simple question: where did this value come from?

For an important record, a reviewer should be able to determine:

when and from where the underlying material was collected;
whether a value was observed, calculated, analyzed, or hypothesized;
which component and version produced it;
which earlier records it depends on;
whether the underlying evidence still verifies;
what actions were taken, when, and by whom.

Append-only history

Important actions create new audit events. Historical events are not silently rewritten. Corrections are represented as later events that identify what they correct, preserving the earlier record and the reason for change.

Analysis and uncertainty

Mon, 01 Jan 0001 00:00:00 +0000

Analytical output should help a reviewer understand why something may deserve attention. It must not turn an anomaly into an accusation.

Explain the components

A significant analytical result should preserve the individual observations and derived signals that contributed to it. An unexplained number is not enough.

The public contract requires output to show:

the relevant source and observation references;
the analytical component and version;
the reasons contributing to the result;
known limitations and missing information;
uncertainty or confidence where meaningful;
the fact that human review is required.

Context changes meaning

An unusual value is meaningful only relative to suitable context. A price, phrase, image, timing pattern, or relationship cannot carry the same meaning across every category, region, period, or source.

Reproducibility

Mon, 01 Jan 0001 00:00:00 +0000

A reviewer may need to understand an old result after software, configurations, or models have changed. Reproducibility begins by recording the context that influenced interpretation.

Interpretation record

Where relevant, analytical output records:

component, model, or method name;
exact version;
configuration and threshold version identifiers;
schema and dataset versions;
prompt identifiers and versions when language models are involved;
processing timestamp;
code revision;
stochastic seed when practical.

Comparable does not mean identical

Scores produced by different versions are not assumed to be directly comparable. A changed model, baseline, configuration, or input population can change the meaning of a number even when its scale appears unchanged.

Human review

Mon, 01 Jan 0001 00:00:00 +0000

Human review is not a decorative final step. It is the point where an accountable person examines the source, reasoning, limitations, and uncertainty before assigning meaning.

What reviewers need

A useful review record should make it possible to inspect:

preserved source material and integrity status;
source and capture metadata;
observations and normalized values;
individual analytical signals and their provenance;
related public observations where relevant;
analytical component and version information;
missing information and uncertainty;
earlier decisions and correction history.

Disagreement is part of the design

Reviewers must be able to disagree with analytical output, record an unknown or insufficient-data result, explain their rationale, and correct an earlier decision without erasing history.

Publication boundary

Mon, 01 Jan 0001 00:00:00 +0000

Transparency is important, but publishing operational detection knowledge or protected material could create security, privacy, and investigative risks. The public site therefore documents principles, contracts, limitations, and status—not sensitive internals.

Appropriate for publication

research purpose and current status;
evidence-first architecture principles;
distinctions between facts, calculations, analysis, and hypotheses;
integrity, provenance, audit, and reproducibility requirements;
explainability and human-review requirements;
known high-level limitations and external approval needs;
public website privacy and security information.

Not appropriate for publication

collector or parser implementation details;
platform weaknesses or source-specific operational behavior;
detection rules, active indicators, thresholds, weights, or signal combinations;
internal prompts that influence operational analysis;
operational datasets, evidence, source captures, or case material;
personal information or investigative leads;
internal governance, authorization, or legal working documents;
credentials, private infrastructure details, or repository metadata.

No approval by implication

Documentation must not describe a proposed collaboration, pilot, authorization, external review, or operational capability as if it already exists. Unknown and not assessed remain valid—and necessary—public statements.