Roadmap: Continuous Situational Intelligence — SI1–SI4

[eamirian]

Public promise

ACE can maintain living, source-grounded orientation over any bounded subject a user is authorized
to examine. The same neutral lifecycle can support personal, organizational, market, geopolitical,
scientific, technical-system, or other pack-defined intelligence.

observe → detect material change → update bounded state → compare claims, commitments, and behavior
→ identify reactions and consequences → orient → watch what could change the conclusion

This is a cross-cutting outcome delivered through the release spine, not a domain embedded in Core
or Intelligence. GI1 supplies the shared governed Intelligence substrate; 0.9 composes the complete
single-user product journey and 1.0 stabilizes it.

Outcome sequence

SI1 — Reproducible situational orientation

• Immutable, versioned, as-of orientation with evidence coverage, contradictions, alternative
  explanations, uncertainty, causal limits, falsifiers, watch conditions, source citations, deep
  links, and exact revision diffs.
• No future-information leakage into historical answers.

SI2 — Standing investigations and accountable change

• Durable Cases with scope, owner, open unknowns, competing hypotheses, falsifiers, cadence, watch
  conditions, authority, and revision history.
• Distinct statement, commitment, observed action, reaction, and outcome records.
• aligned, partially_aligned, contradicted, too_early, unverifiable, and
  insufficient_evidence assessments with exact tests and evidence.

SI3 — Heterogeneous evidence and source governance

• Pack- and connector-owned qualitative, quantitative, time-series, geospatial, track, event, and
  market-contract semantics with units, vintages, revisions, precision, coverage, gaps, and degraded
  states.
• Source origin, syndication, independence, consent, privacy, entitlement, quotation, retention,
  deletion, export, and redaction policy.

SI4 — Attention and measured orientation quality

• Explicit, correctable interest and watch policy over relevance, novelty, magnitude, confidence,
  source diversity, time sensitivity, interruption, and delivery.
• Attributable generation, grouping, suppression, delivery, dismissal, correction, and use receipts.
• Leakage-bounded evaluation of citation correctness, contradiction coverage, calibration, detection
  delay, false-alert rate, revision stability, analyst action, and useful, harmful, or unproven effect.

Acceptance gate

World and Market reproduce the same Core + Intelligence lifecycle without embedding their ontology
in either layer. Each single-user journey ingests mixed evidence, produces an as-of source-linked
orientation, distinguishes claims from observed behavior and reactions, exposes contradictions and
unknowns, establishes a watch, emits or suppresses material change, survives restart, and updates by
append-only revision. Future-information leakage, duplicated-origin, entitlement, privacy,
causal-overclaim, degraded-state, and alert-quality controls pass.

Explicit boundary

Core owns scope, identity, time, durable state, provenance, governed reasoning, receipts, authority,
decisions, outcomes, and failure semantics. Intelligence owns domain-neutral Observation,
entity-state projection, Shift, Signal, Brief, Case, monitor, resolution, delta, scoring, routing,
synthesis, feedback, pack compilation, and conformance contracts. Domain Packs own entities,
relations, aliases, detector configuration, materiality, personas, synthesis policy, and specialized
trust policy. Connectors own reviewed source translation.

World Intelligence and Market Intelligence are the current validation targets. Collaboration is a
post-1.0 expansion and is not required to prove continuous intelligence for one person.

Cross-cuts the passed 0.4–0.8 foundation, 0.9 Single-user Intelligence Builder,
and the 1.0 gate. See the
public roadmap.