The Semantic Situation and Continuity Law

1. The unit of continuity is the semantic situation

AIOS does not define continuity as retention of conversation alone. A transcript records an interaction sequence. A semantic situation records the present relationship among purpose, knowledge, decisions, artifacts, plans, time, and attention.

The semantic situation answers:

• What is this work for?
• What is known, inferred, contested, or missing?
• What has been decided, by whom, and why?
• What artifact or outcome is being formed?
• What is active, parked, stale, or superseded?
• What change is presently being considered?
• What authority and stopping conditions govern that change?

This is the meaning-bearing world reconstructed from durable domain ground and current active state for the next reasoning act.

2. Five regions of meaning

The semantic situation contains five conceptual regions. They need not live in one file or schema.

The central node names a relationship, not a container. The reciprocal ring shows that a change in any region can alter the others—for example, evidence can reopen a decision, which can change an artifact, foreground, or governing direction.

2.1 Direction

Direction carries:

• purpose and desired outcome;
• beneficiary, audience, or receiving system;
• values and governing commitments;
• scope and non-goals;
• adequacy and stopping conditions.

Direction supplies the bearing by which relevance and tradeoffs are judged.

2.2 Epistemic ground

Epistemic ground carries:

• claims and definitions;
• exact evidence and source provenance;
• interpretations and assumptions;
• counterevidence and alternatives;
• contradictions and unknowns;
• confidence basis and uncertainty;
• minority or dissenting accounts.

Its central obligation is to keep standing visible. A polished inference does not become evidence because it is memorable.

2.3 Intentional ground

Intentional ground carries:

• decisions and rationale;
• who had authority to decide;
• live and rejected alternatives;
• conditions under which a decision should reopen;
• commitments, constraints, dependencies, and accepted risks;
• grants for bounded work.

This region distinguishes what the system knows from what the person or project has chosen.

2.4 Productive ground

Productive ground carries:

• artifact purpose, audience, structure, and current language;
• plan outcome, strategy, architecture, commitments, and foreground;
• workstreams, assignments, checkpoints, and returns;
• integrated changes and repairs owed;
• completion criteria and evidence;
• the next meaningful movement.

This region makes reasoning consequential by relating it to work taking form.

2.5 Temporal and attentional ground

Temporal and attentional ground carries:

• current foreground and durable background;
• open, parked, abandoned, or completed branches;
• prior states needed to understand present decisions;
• changed, stale, or superseded ground;
• lineage and recovery routes;
• unattended work and returned results;
• the object to which current attention is bound.

This region is essential to interruption, resumption, asynchronous work, and correction.

2.6 Three context resolutions

The five regions describe kinds of meaning. Three context resolutions describe the span at which those regions are made available. The two structures are orthogonal: direction, epistemic ground, intentional ground, productive ground, and temporal or attentional ground can each appear at any of these resolutions.

• The immediate frame contains the object now in focus, the present input, the selected cognitive movement, and the authority or effect boundary that governs it.
• The active semantic situation contains the purpose, evidence, decisions, artifact state, live branches, and constraints that presently bear on the work.
• The durable domain ground contains the longer-lived terminology, sources, relationships, methods, accepted decisions, and history from which active situations can be reconstructed.

A person ordinarily directs attention through one foreground frame. AIOS can preserve multiple active situations and durable relationships across files and asynchronous branches, then compose the resolution needed for a particular movement. No single model invocation is assumed to receive or comprehend the entire field. Each invocation receives a bounded projection suited to its charge; concurrency belongs to the system as a whole.

A cross-resolution context lock is the degree to which a composed context aligns the immediate frame with the relevant active semantic situation and durable domain ground. It is not a claim of completeness. A strong lock makes the governing purpose, current state, material evidence, standing, and authority sufficiently coherent for the judgment while preserving routes to omitted ground. Its quality can vary: stale, missing, excessive, or wrongly weighted material can weaken the lock even when the context is large.

3. The continuity law

The target intelligence loop is:

durable domain ground + current active state
  → reconstructed semantic situation
    → context-visible situation brief
      → purpose-shaped reasoning ground
        → one judgment
          → semantic product + semantic delta
            → integration and person acceptance where needed
              → renewed active semantic situation
                + authorized changes to durable domain ground

The fork after judgment is essential: the thing made and the account of what it changes must remain separately inspectable. They reunite at the integration boundary, where required person authority enters; only then do renewed active state and authorized durable change shape the next brief.

Does not establish Renewal is selective rather than lossless, and physical persistence of a product or reasoning trace does not grant it accepted or future-governing standing.

The loop has two distinct outputs:

Semantic product

What the judgment creates: a finding, interpretation, draft, decision proposal, plan, edit, test result, synthesis, or return.

Semantic delta

What that product changes—or does not change—in durable ground:

• confirms;
• qualifies;
• contradicts;
• supersedes;
• proposes;
• opens;
• closes;
• leaves unresolved;
• changes an artifact;
• creates a dependency;
• requires person acceptance.

The product without a delta is difficult to integrate. The delta without the product is difficult to evaluate.

Continuity therefore proceeds through situation renewal, not indefinite extension of one reasoning trace. The system carries forward the accepted or otherwise relevant product and delta: what now bears on purpose, knowledge, commitment, artifact state, attention, or future judgment. It may preserve additional thinking physically as history, evidence, or a recovery route, but physical persistence does not give that material future-governing standing.

Renewal is deliberately selective rather than lossless. It should preserve the source paths, standing, material disagreement, and disclosed losses needed to recover or challenge the renewed account. It should not require every intermediate token, discarded route, or low-level operation to occupy the next reasoning field.

4. Standing is not inferred from storage

The architecture preserves distinctions among:

Storage, repetition, model confidence, typographic polish, metadata presence, or visibility in a user interface do not change standing.

5. A situation brief is a use-shaped projection

No single judgment needs the whole semantic situation. A situation brief is a compact, purpose-shaped view containing the regions that bear on the present act.

A strong brief names:

1. the governing purpose and present bearing;
2. the current object and requested transformation;
3. accepted, contested, and unknown ground;
4. relevant evidence, decisions, constraints, and live alternatives;
5. current artifact or plan state at the needed resolution;
6. authority and effect boundary;
7. expected product and stopping condition;
8. omitted or compressed ground and a path to deeper sources.

The brief is neither a canonical source or artifact nor accepted ground. It is a reconstructable projection from canonical sources and standing-aware durable records.

The brief should begin at the coarsest sufficient semantic altitude for the intended consequence. It descends only when a named need makes higher-level ground inadequate—for example, disputed compression, exact evidence, consequential wording, structural ambiguity, or a pending file mutation. Material recovered below must then be interpreted again at the altitude of the receiving judgment rather than allowed to displace the purpose that warranted descent.

6. Context requires a warrant

Context selection should be explainable.

For each included region, the system should be able to state:

• why it bears on the judgment;
• which source and revision supplied it;
• what standing it has;
• whether it is exact or compressed;
• what was intentionally withheld;
• when deeper ground should re-enter.

This warrant can live in preparation records and receipts. It does not need to burden the model-facing capsule with mechanical prose.

The warrant converts context selection from an opaque retrieval act into a researchable composition decision.

7. Example: correcting a strategy after new evidence

Situation before the new evidence

direction:
  outcome: Establish a credible launch strategy.
epistemic_ground:
  accepted_claim: The primary audience values breadth of capability.
  basis: Three early interviews.
intentional_ground:
  decision: Lead with breadth.
  reopen_if: Contradictory customer evidence appears.
productive_ground:
  artifact: launch-strategy.md
  current_state: messaging draft formed around breadth
temporal_ground:
  status: active

New movement

Five additional interviews indicate that the audience values depth and trust over breadth.

Semantic product

A comparison of the old and new evidence, including sample limitations and possible segment differences.

Semantic delta

challenge:
  target: accepted_claim
  status: material-counterevidence
decision_effect:
  action: reopen
artifact_effect:
  status: revision-proposed
unresolved:
  - whether the interviews represent two distinct segments

Renewed situation

The prior decision remains visible but no longer governs unqualified future context. The next capsule includes the contradiction, the reopened decision, the affected artifact, and the exact evidence needed to discriminate between alternatives.

This is correction through lineage, not deletion of an embarrassing past answer.

8. Situation reconstruction across interruption

When attention returns after an interruption, the system need not replay every exchange. It can reconstruct:

purpose and desired outcome
  + current artifact/plan state
  + last consequential movement
  + accepted decisions and active commitments
  + open tensions and unknowns
  + relevant asynchronous returns
  + current source standing
  + likely next meaningful movements
  + the person's new input

This reconstruction is a primary claim of the architecture: continuity can be maintained by durable semantic state even when conversational process is absent.

9. Different judgments require different projections

Context width is controlled by the judgment's altitude and consequence, not by a universal token budget.

This is not a one-way preference for abstraction. High-altitude ground can hide decisive evidence, while exact detail can obscure project or domain bearing. The compositional obligation is to start no lower than necessary, descend for an explicit question, and return the finding to the semantic scale at which its consequence will be judged.

The requested operation matters as much as the subject matter. Exact lookup, comparison, aggregation, contradiction resolution, temporal update, structural reasoning, transformation, and verification can require different evidence and different arrangements of the same evidence. Two movements about the same document may therefore need materially different situation briefs.

This makes a projection operation-aware as well as purpose-shaped. The system is not only asking, “What information is related to this topic?” It is asking, “What relationships must be visible for this particular transformation to be performed well?”

10. The situation is distributed, not centralized

The five semantic regions can be reconstructed from multiple owner files:

• canonical documents;
• companion Document Homes;
• project or Session Homes;
• workflow and stage files;
• source sets and exact evidence;
• decisions, annotations, relationships, and lineage;
• assignments, returns, and operation receipts;
• accepted expertise resources;
• derived maps and indexes.

No file needs to become a universal semantic object or a hidden global memory. The situation exists through resolvable relationships among durable records.

The mechanisms that maintain those records and the intelligence moving through their relationships are different layers of the same architecture. A file observer may detect a changed revision, a validator may identify malformed structure, and an indexer may rebuild a projection. Those mechanical facts do not by themselves infer purpose, change semantic standing, initiate new semantic work, or authorize a consequence. Intelligence arises as people and bounded model judgments interpret purpose and ground through the system; exact mechanisms preserve and enact only the effects their authority permits.

11. Failure modes

Transcript substitution

Treating the recent interaction sequence as the complete situation omits project purpose, standing, decisions, and artifact state.

Summary substitution

Treating one summary as canonical hides what it omitted and prevents exact descent.

Standing collapse

Treating proposals, observations, and decisions as equivalent allows speculative language to harden into future context.

Context maximalism

Including everything reduces relevance, obscures conflict, increases latency, and makes provenance harder to interpret.

Context minimalism

Removing purpose, rationale, disagreement, or source paths produces concise but brittle reasoning.

Central-state dependence

If the semantic situation exists only in an active process, interruption destroys continuity and prevents audit.

12. Evaluation criteria

A semantic-situation implementation should be evaluated on whether it can:

1. reconstruct the present work after restart from durable files alone;
2. preserve source, interpretation, decision, and proposal standing;
3. descend from high-level bearing to exact evidence;
4. show why a context block was included;
5. identify what a judgment changed;
6. prevent unaccepted output from silently becoming future truth;
7. stop using superseded ground while retaining lineage;
8. recompose different projections for different judgment forms;
9. carry a contradiction across interruption into later reasoning;
10. remain useful when derived projections or indexes are deleted and rebuilt;
11. renew a situation from the relevant semantic product and delta without replaying or promoting the complete reasoning trace;
12. show the named need that justified exact descent and how the result was reinterpreted at its receiving altitude.

13. Relationship to the whole and research

The semantic situation is the connective tissue of the architecture. The Fractal Seed supplies its developmental grammar. Cognitive Movements determine which transformation is being attempted. Context Composition turns the situation into a model-facing field. Memory Ecology explains how the situation can be reconstructed across timescales. Files and Metadata supply its durable local substrate.

Recent context research reinforces the need for operation-aware projections. NoLiMa shows that nominal context capacity does not guarantee access to semantically relevant facts as context grows. A 2025 study of context length and retrieved evidence found that additional surrounding context can impair performance even when the relevant information has been retrieved. These findings do not prescribe the five-region semantic situation. They support the more general decision to optimize the relationships visible to one judgment rather than equating available tokens with usable context.

The complete evidence and boundary analysis is available in the context-composition research brief. The public research question is direct: can reconstructable, standing-aware, operation-specific situations improve accepted outcomes and correction over transcript continuation, generic retrieval, or context-maximal baselines?