# MMRY–SensAffect Memory Atom Standard

## Shared Emotional Cluster Specification

**Publication edition 1.0 · 8 September 2026**

**Status:** design specification. This edition defines the agreed conceptual model and requirements for conforming implementations. It does not certify existing products, establish validated psychological scoring, or claim that every requirement has been implemented.

## 1. Core definition

**Memory + Feeling + Context = MMRY Atomic Unit.**

A **Memory Atom**, abbreviated **MEM**, is an identifiable MMRY record linked to a SensAffect emotional interpretation and associated context. Its **multimetric cloud** is a visual representation of that information.

| Component | Meaning | Responsibility |
|---|---|---|
| Memory | What was recorded or recalled | MMRY: content references, identity, time and provenance |
| Feeling | How the experience was felt | SensAffect: primary and supporting emotions, each with VAD coordinates and interpretation provenance |
| Context | The circumstances associated with the memory and feeling | Sourced people, places, times, situations, topics, sentiment and relationships |

Memory and interpretation remain separate. A revised interpretation must not rewrite the recorded memory. A Memory Atom may exist before its feelings or context have been identified. Unknown information must remain unknown.

The standard comprises a **shared schema** defining the information and **shared rules** defining composition, interpretation and presentation. “MUST” denotes a requirement of this specification. A **proposed method** is an implementation candidate, not an established or validated scoring method.

## 2. Hierarchy and scope

The order below describes conceptual scope, not rendered dot size or a rigid physical containment structure.

| Scope | Term | Definition |
|---|---|---|
| Reserved | Quark | No meaning assigned; not required for implementation |
| 1 | Particle / eVAD particle | One affect state with its own complete VAD triplet and optional emotion word |
| 2 | Memory Atom / MMRY Atomic Unit | Memory + Feeling + Context |
| 3 | Memory Cluster | A recollection unit, usually 3–5 related atoms associated by time and meaning; this is not a mandatory size limit |
| 4 | Narrative Cloud | Memory clusters connected through a sequence or story |
| 4 | Theme Cloud | Memory clusters connected through shared meaning, including across time |
| 5 | Nebula | The wider web of organized and unorganized memories and their connections |
| 6 | Identity | The evolving whole of a person's memories, feelings, triggers, actions and relationships; not a single score |

**Fog** describes unorganized memories or clusters, regardless of size. **Event** is a time/context association, not automatically a synonym for Memory Cluster. **Neural Map, Timeline and nebula views** are presentations of the records, not additional hierarchy levels.

A multimetric cloud depicts an atom or a clearly identified aggregate; its visual name does not make it a Narrative or Theme Cloud. Shared membership must reference the same records rather than duplicate them. Counts must distinguish memories, atoms, clusters, affect states and visual marks.

## 3. Feeling and VAD

Each interpretation can contain primary, secondary and tertiary affect states. Supporting states are optional. They are three possible feelings, not the three VAD axes and not automatically three moments in time.

| Dimension | Meaning | Canonical representation |
|---|---|---|
| Valence | Unpleasant to pleasant | −1 to +1 |
| Arousal | Low to high activation | −1 to +1 |
| Dominance | Low to high perceived control | −1 to +1 |

These signed ranges are a **representation convention adopted by this specification**, not a claim that all existing sources use them. Each adapter MUST declare its input scales. A declared unit-interval value is converted using `signed = 2 × unit − 1`. Scale must never be guessed from the sign of a value. Invalid values must be identified; missing values remain null. Zero is a valid midpoint.

Each affect state MUST carry its own VAD structure and source. Complete VAD values are necessary for a complete coordinate plot; incomplete states can remain available as explicitly incomplete information. No system may invent a secondary or tertiary state to fill a visual slot.

Emotion words support interpretation and calibration. A person's reported or selected word must remain distinguishable from a derived label. A VAD coordinate does not uniquely establish an emotion word. Primary/secondary/tertiary ranking, confidence, intensity, sentiment and activation are different concepts and must not substitute for one another.

## 4. Context as the third side of triangulation

Context makes it possible to relate what happened to how it felt and under what circumstances. It is a first-class part of the atom, not decorative atmosphere. “Triangulation” describes the joint interpretation of Memory, Feeling and Context; it does not establish a causal or statistical procedure.

Context records include people and their names, places, time, situations, topics and other relevant circumstances. Each record MUST retain its identity, source, and whether it is recorded, user-confirmed or inferred. Names alone must not be treated as unique person identities. A context record should distinguish the original occurrence from a later recall when that information is available.

### 4.1 Detection

Context may be supplied directly by a person, available in source metadata, or extracted from recorded content. An implementation may identify named people, places, temporal expressions and situations in text or media. These are permitted input routes, not a prescribed detection model.

Detection MUST preserve evidence references or source wording where available and distinguish an inferred interpretation from an explicit statement. Recording time is not automatically the event time; a mentioned person is not automatically present; a mentioned place is not automatically the event location. Ambiguous identities and times remain unresolved rather than silently merged.

### 4.2 Influence on a feeling

A context item can relate to the atom generally or influence a specific affect state. A scored influence link MUST identify the context item, affect state, interpretation revision, strength, source and any confidence assessment.

For example, “At the hospital, seeing Sarah made me feel relieved” supplies place, person and situation context and explicitly links seeing Sarah to relief. It does not establish an exact influence strength or establish that the hospital contributed equally.

Influence evidence may come from:

| Evidence | What it supports | Limitation |
|---|---|---|
| Direct user assessment | A reported contribution and, if asked, reported strength | Subjective report, not independently established causality |
| Explicit relational language | A stated context–feeling association | Often establishes a link without a numeric magnitude |
| Qualifiers such as “slightly” or “mostly” | An inference about relative strength | Numeric mappings are not established by this standard |
| Repeated context–feeling patterns | A hypothesis of association across memories | Frequency and co-occurrence do not establish causal influence |

A proposed storage scale is `[0, 1]`: zero is explicitly assessed no influence; one is the strongest influence within the declared method; null means unknown. **The scoring procedure and its calibration are not established.** An uncalibrated system must retain a qualitative assessment or null rather than invent a precise score. High confidence that a relationship exists does not imply strong influence.

### 4.3 Proximity

**Closer context dots mean stronger contextual influence on the linked emotion; farther means weaker influence.** Unknown influence must appear in an identified unscored area or listing rather than be positioned as weak influence.

Once a comparable influence score exists, a proposed display mapping is:

`distance = near + (1 − influence) × (far − near)`

Here `0 < near < far` are declared profile parameters. This formula is a display convention, not a psychological scoring model. Angular placement may aid readability; radial distance carries influence. The emotion's VAD values and coordinate position must remain unchanged.

One context item may influence multiple feelings with different strengths. A view may show the context of a selected feeling or use linked appearances of the same context record. It must not imply that a single point satisfies incompatible distances. Duplicate appearances do not create new records. Layout constraints must not silently distort the relationship being displayed.

### 4.4 Contextual similarity

Keep three comparisons distinct:

| Comparison | Basis |
|---|---|
| Memory similarity | Content or experience |
| Feeling similarity | Corresponding affect states and VAD |
| Context similarity | Shared people, places, temporal relationships, situations or topics |

Similar circumstances can produce different emotions. Similar emotions can occur in different circumstances. Contextual similarity can support suggested clusters and clouds, but must retain its evidence and comparison method. Unknown context is not a mismatch.

**No universal similarity formula, feature weights or clustering thresholds are established here.** Influence links context to feeling within an interpretation; similarity compares circumstances across memories. Neither score may stand in for the other.

## 5. Logical schema

The following is an illustrative typed contract, not a claim of an existing API or a complete machine-readable validation schema. Versioned implementations must add formal validation and define reference resolution. Scores are optional when their methods are not established.

```ts
type Provenance = {
  product: string;
  sourceRecordId: string | null;
  method: string | null;
  recordedAt: string | null; // ISO 8601 when known
};
type Metric = {
  value: number | null; // signed [-1, 1]; null means unavailable
  origin: 'observed' | 'inferred' | 'aggregated' | 'unavailable';
};
type AffectState = {
  id: string;
  vad: { valence: Metric; arousal: Metric; dominance: Metric };
  emotionWord: {
    label: string;
    origin: 'reported' | 'selected' | 'derived';
    lexiconVersion: string | null;
  } | null;
  confidence: number | null; // [0, 1], not intensity or rank
  source: Provenance;
};
type MemoryAtom = {
  schemaVersion: string;
  id: string;
  memory: {
    id: string;
    contentRefs: Array<{ id: string; mediaType: string }>;
    recordedAt: string | null;
    recalledAt: string | null;
    source: Provenance;
    relationships: Array<{ type: string; targetMemoryId: string }>;
  };
  feeling: {
    interpretationId: string;
    revision: string;
    source: Provenance;
    primary: AffectState | null;
    secondary: AffectState | null;
    tertiary: AffectState | null;
  } | null;
  context: {
    items: Array<{
      id: string;
      kind: 'time' | 'place' | 'person' | 'situation' | 'other';
      label: string; // e.g. a person's name, place or situation description
      value: string; // ISO 8601 for a known precise time; otherwise source wording
      entityRef: string | null; // stable person/place/etc. identity when available
      temporalScope: 'occurrence' | 'recall' | 'unspecified';
      interpretationRef: { id: string; revision: string } | null;
      origin: 'recorded' | 'user-confirmed' | 'inferred';
      confidence: number | null; // [0, 1] when supplied
      source: Provenance;
    }>;
    affectLinks: Array<{
      id: string;
      contextItemId: string;
      affectStateId: string;
      interpretationId: string;
      interpretationRevision: string;
      influence: number | null; // [0, 1]; null is unknown, not zero
      confidence: number | null; // independent of influence
      origin: 'user-reported' | 'inferred';
      source: Provenance; // method identifies the scoring policy when inferred
    }>;
    topics: Array<{ id: string; label: string; source: Provenance }>;
    sentiment: {
      category: string;
      score: number | null;
      source: Provenance;
    } | null;
    occurredAt: string | null;
    relationships: Array<{
      type: string;
      targetId: string;
      source: Provenance;
    }>;
    clusterIds: string[];
    eventIds: string[];
    narrativeCloudIds: string[];
    themeCloudIds: string[];
  };
};
```

Entity and content references resolve to separately identified records. Influence links must resolve to existing context items and affect states in the named interpretation revision. Confidence is not automatically a calibrated probability. A numeric influence requires a declared scoring method; inferred values require the producing method/model version in provenance. A release schema must specify how those method versions are encoded.

If context is shared across atoms, each local association retains its provenance. Raw media need not be copied into the renderer. Access to related memory content remains subject to the source application's permissions.

## 6. Multimetric rendering

The richer multimetric cloud is retained; a Memory Atom is not limited to four dots.

| Element | Meaning |
|---|---|
| Selectable MMRY reference or atom container | Opens the associated memory; a separate compulsory memory dot is not required |
| Primary emotional core | The primary feeling |
| Separate V/A/D markers | Three dimensions of the selected feeling; primary is the default |
| Supporting emotion dots | Secondary and tertiary feelings, each retaining its own VAD |
| Topic and sentiment markers | Sourced contextual fields |
| Context dots | Sourced people, place, time, situation or other context, with influence-based proximity where scored |

Dots remain round, using size and color rather than shape to differentiate visual roles. Every meaningful dot MUST resolve to its data. Context must not be fabricated to fill a fixed decorative quota. Grouped dots must expose their underlying records.

A spherical scene and a literal VAD coordinate plot are distinct. The multimetric profile may use separate dimension markers and supporting arrangements; it must not claim every dot's position is a complete VAD coordinate. A coordinate profile must declare its transform. A geometric origin `(0, 0, 0)` does not claim that the underlying memory has neutral feelings. Points need not sit on a sphere's surface.

The VAD marker palette is green for valence (`#4ade80`), amber for arousal (`#fbbf24`) and purple for dominance (`#a78bfa`). The complete emotion palette and size mappings remain release work. Each profile must declare whether size expresses rank, activation or another field; these must not be silently conflated. Confidence must not be mistaken for influence strength.

Shared profiles may include a multimetric atom view, nested Neural Map, compact thread summary, Timeline and semantic-nebula view. Equivalent data and profile versions must preserve meanings across clients. Semantic distance is not VAD distance or contextual influence distance.

Interactive views must provide accessible textual details, keyboard-operable controls, selection, recentering where applicable and reduced-motion support. Color alone must not be required to interpret data. Projection, depth and motion must preserve round particles and readable relationships.

## 7. Aggregation and change

An aggregate must identify its source atoms, interpretation selection, affect role, method and coverage. Repeated theme membership must not multiply contributions. Primary, secondary and tertiary states must not be silently averaged together.

An unweighted per-axis mean of one selected role across unique atoms is a **proposed baseline**, not an established measure of psychological meaning. Missing axes are excluded with contributing counts; no contributors means unavailable. An aggregate must not appear to be an original memory record. Revised metrics, context links and memberships must refresh their representations without changing source identity.

## 8. Methods not yet established

These items are explicitly **not established or calibrated by this publication**. A working software implementation or a bounded number alone does not validate a scoring method.

| Method or decision | Current status | Required before claiming established scoring |
|---|---|---|
| Media/text → VAD | No universal estimator prescribed | Named model/method, source scales and evaluation against an appropriate reference |
| Emotion words ↔ VAD | No unique universal mapping | Versioned lexicon, treatment of user language and evaluated label assignment |
| Primary/secondary/tertiary rank | Roles defined; assignment procedure open | Explicit user ranking or documented and evaluated inference policy |
| Context detection/entity resolution | Evidence/provenance rules defined; model open | Detection method, ambiguity handling and evaluation |
| Context influence strength | Proposed `[0,1]` storage only | Elicitation rubric or inference method, anchors and calibration; distinguish association from causal claims |
| Influence from linguistic qualifiers | No numeric mapping established | Evaluated rubric; preserve original wording |
| Influence from repeated patterns | Association evidence only | Declared method and evaluation; no causal claim from co-occurrence alone |
| Confidence | Optional separate field | Defined meaning and calibration before treating it as a probability |
| Context/memory/feeling similarity | Three distinct comparisons defined; formulas open | Versioned features, weights, normalization and evaluation |
| Cluster/cloud assignment | Typical size and relationships defined; thresholds open | Explicit manual grouping or documented, evaluated grouping procedure |
| Aggregate affect | Unweighted mean proposed | Declared population/role and evaluation appropriate to intended use |
| Sentiment and intensity | Distinct from VAD and influence; scoring open | Named scales, methods and comparison limits |
| Influence → display distance | Monotonic convention defined; linear mapping proposed | Shared profile bounds and usability validation; this does not validate the influence estimate |
| Emotion colors, particle sizes, spherical transforms | Complete cross-client mappings not finalized | Published versioned rendering profile and conformance checks |

Until these items are resolved, implementations MUST label inferred or experimental outputs, identify methods and preserve unavailable values. This document establishes a representation standard, not diagnostic validity or a validated causal model.

## 9. Conformance and versioning

A conforming release must demonstrate that:

1. Memory, Feeling and Context remain independently identifiable and linked; interpretation changes do not overwrite memory content.
2. Primary and supporting VAD triplets survive adaptation independently, including zero and unavailable values.
3. Input scales and conversions are explicit, applied once, and preserve provenance.
4. Context dots resolve to actual sourced records; no missing context generates filler dots.
5. Influence, confidence and similarity remain distinct; scored proximity is monotonic and unknown influence is unscored.
6. Context linked to multiple feelings retains its identity and each relationship's strength.
7. Context similarity can remain high when feeling differs; explanations expose supporting matches and methods.
8. Membership and visual duplication do not duplicate records or inflate aggregates.
9. Accessible details expose the same information as the visualization; selection and reduced-motion behavior work.
10. Schema, adapter, interpretation/scoring methods, lexicon and rendering-profile versions are recorded. Experimental methods and incomplete conformance are disclosed.

These are acceptance requirements, not reported test results. Adoption across MMRY, SensAffect, THRPY and Qbit requires implementation-specific verification. A published edition does not itself establish deployment or conformance.