Epistemedia Roadmap

• Object ID: em:documentation:sha256:2f1a4a7ef2d8194d184f06a300251ba055541be0029e1ca0ddef5a299063d4c5
• Kind: documentation
• Repository path: docs/roadmap.md
• Content digest: bcb6525b2343b8dd203b3a89fccdae5c5330d12313d061e31f03eac88a4ca8d4

Source content

Epistemedia Roadmap

Phase 0 — Self-describing public realm

Compile the repository’s own architecture, governance, tasks, code, tests, and releases through the same public catalog used by every interface. Establish deterministic identity, disclosure safety, and agent-native operation.

Phase 1 — Normative epistemic kernel

Finalize event envelopes, canonical bytes, content-addressed source artifacts, exact spans, propositions, assertions, evidence relations, derivations, three-time semantics, graph replay, policy evaluation, and projection manifests.

Phase 2 — Personal sovereign node

Support local source capture, questions, research plans, observations, claims, gaps, contradiction search, multiple compiled dossiers, private disclosure classes, and bundle export/import. The node must remain useful offline before federation.

Phase 3 — Three heterogeneous realms

Operate deliberately incompatible exploratory, standards-verification, and skeptical-public realms. Test ontology mapping, partial import, policy divergence, dependence preservation, and lossless disagreement.

Phase 4 — Autonomous research behaviors

React to accepted events:

• new question → decompose research gaps;
• new proposition → seek primary evidence;
• new support → search dependence and counterevidence;
• new established status → launch contradiction search;
• new contradiction → propose discriminating test;
• new source version → identify affected claims;
• stale claim → schedule revalidation;
• new bundle → quarantine and evaluate.

Phase 5 — Machine governance sandbox

Allow agents to propose policies, replay them over historical corpora, inspect outcome diffs, run independent adversarial evaluation and bounded canaries, and promote or fork only through accepted predicates.

Phase 6 — Adversarial benchmark

Test Sybil swarms, source laundering, copied-source replication, ontology poisoning, evaluator collusion, model monoculture, prompt injection, private-state leakage, self-promotion, metric gaming, strategic uncertainty, and malicious policy packs.

Phase 7 — Public federation

Add public realm discovery, subscriptions, domain packs, shared compute, contribution incentives, independent hosted nodes, cross-realm comparison, and self-host onboarding only after standalone nodes and heterogeneous federation are demonstrably useful.