Principles

Two layers on one page. First principles are the five irreducible commitments behind the project — stable, generative, rarely revised. Heuristics are the working rules that follow from them, grouped by category and each tagged with the commitment(s) it derives from. Settled heuristics are binding; candidates are under stress-test.

FP1. Structural data sovereigntyFP2. Truth is contested; provenance is notFP3. Capture friction is the existential constraintFP4. Meaning lives in arenas, not coordinatesFP5. Structure emerges from data, not from architectural foresightSettled · Platform-wide · 11Settled · Canonical layer · 9Settled · Editorial knowledge · 6Settled · Monetization · 3Candidate · under stress-test · 11

The five commitments

Irreducible and generative: nothing else in the project derives them, and most other principles can be traced back to them. If one of these is wrong, the project is wrong.

1

Structural data sovereignty

The user owns their data — both the bytes and the meaning — structurally, not as a promise. Portability written into ownership beats a promise of portability; membrane architecture beats both.

This is the project's answer to who owns what. It generates the membrane abstraction, the cell architecture, the insistence that raw input remain authoritative, and the requirement that no architectural decision foreclose the path to Tier 3. Platform-holds-the-data commitments are conscious tradeoffs against this principle, not defaults.

2

Truth is contested; provenance is not

The system does not claim authority on truth. Knowledge, facts, and meaning are continuously contested and revised — by individuals, by families, by the canonical layer, by history. Remoir's job is to capture what was said, by whom, when, with what confidence, and to surface that transparently. It is not to adjudicate.

The underlying stance: no authority on truth. Wisdom, knowledge, reason, ethics, truth, morality are valued highly because they are continuously being worked on, not because the system can deliver them. This is why raw input must be preserved as authoritative (you cannot redo provenance later) and why every derivation is a re-runnable projection. It is also why family memory takes precedence over canonical fact in the family's own record — the family is the authoritative narrator of its own life, not because family memory is more accurate, but because the system declines to overrule it.

3

Capture friction is the existential constraint

At end-user ingestion, any design that adds steps is rejected on those grounds alone. If capture is not instant and frictionless, nothing else in the system gets to matter, because there will be nothing to operate on.

This generates the "capture now, make sense later" workflow, the "everything is optional at capture time" rule, the photo-primary capture model, and the design veto that complexity must stay invisible to the user. Later phases — editorial, governance — may deliberately introduce friction; this principle scopes to the moment of capture by the end user.

4

Meaning lives in arenas, not coordinates

The unit of meaning in a memoir is the arena (the bounded transformation of a place by what happens in it) — not the GPS pin, the calendar event, the photo file, or any other digitally-determined structure. Geography, time, and media are services to meaning, not the substrate of it.

This generates the Moment as the experiential atom, the threads-vs-tags distinction, the principle that tags grow into threads (and threads into richer objects) through use rather than upfront ontology, and the layered object model. It is the project's rejection of database-shaped thinking about lived experience.

5

Structure emerges from data, not from architectural foresight

Build only the layer the data actually demands. Let proliferation justify each new entity, table, or abstraction. The pull toward modeling completeness is to be resisted; the pull toward additive, minimal structure is to be trusted.

This generates the additive data model, the four-layer object model (where layers reflect how an object came to exist), the tag-graduation principle, the deferred-architecture stance (ADR-001), and the general posture that today's correct abstraction is tomorrow's lock-in if we commit too early. It is epistemic humility applied to the schema rather than to the content.

---

Heuristics

Working rules grouped by category. Each card shows the first principle(s) it derives from as a → FPn link. A placeholder marks heuristics that don't trace cleanly to any current commitment.

Settled · Platform-wide

11
Capture now, make sense later
Input must be instant and frictionless. Organisation and insight happen asynchronously, when you have time.
product → FP3 04-2026?
Capture friction is the existential constraint (at end-user ingestion)
Any design that adds steps at capture time is rejected on those grounds alone. Applies to end-user capture; later phases (editorial, governance) may deliberately introduce friction.
product → FP3 05-2026?
Complexity stays invisible to the user
If Remoir ever asks a user to disambiguate ontological categories the system uses, the design has failed. Used as a hard veto test.
product → FP3 04-2026?
Structural data sovereignty
Data belongs to the user structurally, not as a promise. Portability written into ownership > promise of portability; membrane architecture > both.
architecture → FP1 04-2026?
No architectural decision forecloses cell/membrane
Decisions compatible with all three sovereignty tiers (promise / structural / membrane) are preferred. Platform-holds-the-data commitments are conscious tradeoffs, not defaults.
architecture → FP1 15-04-2026
Meaning lives in arenas, not coordinates
The memoir's unit of meaning is the arena (Huizinga's magic circle) — the bounded transformation of a place by what happens in it. Geography is service to meaning, not the other way around.
data-model → FP4 04-2026?
Threads vs. tags — the narrative-continuity test
Threads are infinite (carry a story across time); tags are experiential (color individual moments). The test: can you tell its story across time? If yes, thread; if no, tag. Guards "thread" from becoming a catch-all.
data-model → FP4 04-2026?
Tags grow into threads through use, not upfront design
Promotion from tag → thread is suggested by usage patterns; the user decides. No upfront ontology imposed on the user.
data-model → FP4 04-2026?
Raw Input as primary; derivations are projections
The original, unprocessed input (a WhatsApp message, a photo with EXIF intact) is the authoritative object. Every structured entity is a re-runnable derivation. Gates every storage decision and is load-bearing for Tier 3 portability.
architecture → FP1→ FP2 04-2026?
Inaction is the correct default on contested governance
The federation is structurally biased toward stasis on contested decisions. For a multi-decade memory archive, this is a feature.
governance 04-2026?
Structure emerges from data, not from architectural foresight
Build only the layer the data actually demands. Let proliferation justify each new entity tier. Overrides the pull toward modeling completeness.
process → FP5 04-2026?

Settled · Canonical layer

9
Everything is optional at capture time
A memory can arrive with just a text body. No field blocks capture.
data-model → FP3 04-2026?
Entities exist independently
Location, Event, Person are first-class. Can be created and linked before any memories reference them.
data-model → FP4 04-2026?
The model is additive
Phase 2 features bolt on without modifying existing tables. Nullable columns + new tables only.
data-model → FP5 04-2026?
Canonical records are reference data, not user data
Users never create canonical records directly. Admin-curated reference layer.
data-model → FP2 04-2026?
Match, don't merge
User-created rows are never rewritten or deleted when mapped to a canonical. They are joined through the canonical id at query time. Every mapping is reversible.
data-model → FP2 04-2026?
Mapping is reversible and auditable
Every canonical mapping has a documented revert path and an audit trail.
architecture → FP2 04-2026?
User experience is unchanged by canonicalisation
The user does not perceive that canonicalisation happens; the user-facing model stays simple.
product → FP3 04-2026?
Reversibility is mandatory for automated decisions
Anything the matcher writes autonomously must have a documented one-action revert path. The audit shape of an auto-confirm is identical to a human confirm.
architecture → FP2 04-2026?
Privacy by default for cross-family discovery
Cross-family discovery is opt-in. Default is private.
product → FP1 04-2026?

Settled · Editorial knowledge

6
Lazy enrichment at memoir-generation time
External knowledge is pulled in only when memoir is being generated, not at capture or storage time.
architecture → FP3 04-2026?
Verification, not authorship
The editorial role verifies external knowledge; it does not author family memory.
product → FP2 04-2026?
Knowledge is not identity
Separate tables for canonical knowledge vs. canonical identity. They have different lifecycles and trust models.
data-model → FP2 04-2026?
Public vs. private canonicals is structural, not policy
The distinction is enforced by schema, not by access rules.
architecture → FP1 04-2026?
Trust must be visible to the reader
Provenance and confidence are surfaced in the reader UI; no opaque "the system says so".
product → FP2 04-2026?
Family memory wins conflicts
When family-stated memory and external knowledge disagree, family memory takes precedence. Family memory is never silently corrected.
product → FP2 04-2026?

Settled · Monetization

3
Not all data value is equal — the value stack
Four types of data value (collection, maintenance, interpretation, custom metrics) have different monetization stances. Lower-stack = commons-protective; higher-stack = more freely commercial.
strategic 05-2026?
Provenance and attribution are foundational
All monetization presupposes the ability to track who did what, when, with what quality. Not a feature — a system property.
architecture → FP2 05-2026?
Activation economics per dataset, not user-count thresholds
Commercial-contribution viability is determined per-dataset (activation rate × payout per activation), not by global user count.
strategic 05-2026?

Candidate · under stress-test

11
Audit transparency as a sovereignty primitive
Visibility into what the system has captured and processed is structural proof that the user owns their digital cell. The audit layer must be glanceable by default and exhaustive on demand.
product → FP1 12-05-2026
Moment is the experiential atom
Remoir's anchoring unit is the Moment — the lived experience — not the Calendar Event, photo, or any other digitally-determined structure. Calendar events, photos, listening history, location pings are data points that belong to a Moment.
data-model → FP4 12-05-2026
Build the membrane abstraction from day one; defer the cryptography
Phase 6 cryptographic separation is end-state. Implementation deferred; the abstraction is not. Tier 3 readiness checklist binds all Phase 1–5 work.
architecture → FP1 11-05-2026
"Fed but not read" — canonical layer as one-way valve
Canonical layer can be contributed to but not directly read or queried. Data surfaces only via legitimate user encounter with sufficient match specificity. Person-data-scoped; world-data scope TBD.
architecture → FP1 05-2026?
Canonical layer regime split (world-data vs person-data)
Two regimes (at minimum) with different access and consent rules. Don't pre-commit to strict two-regime taxonomy — leave room for additional regimes.
architecture → FP1 05-2026?
Three-axis data architecture (scope × stage × character)
Data has three orthogonal properties: scope (private / connection-gated / public), stage (raw / curated / canonical), character (factual / contested / opinion / promotional). Character is a tag, not a layer.
data-model → FP5 05-2026?
Digital cell = private row across all stages
The digital cell maps to the private scope across all three data stages (raw, curated, canonical), not just one of them.
architecture → FP1 05-2026?
Capped returns over governance gymnastics
Removing the financial reason for extraction is more durable than building governance defenses against extraction. Belongs at the capital-structure layer.
strategic 05-2026?
Object model is layered by origin and function
Objects are organised in four layers — Plumbing (substrate), Structural (bounded, nameable), Emergent (AI-surfaced from patterns), Deliberate (user-authored). The layer reflects how an object comes to exist, not just what it is. Orthogonal to the three-axis data architecture (which is about data scope/stage/character).
data-model → FP4→ FP5 13-05-2026
Tags graduate to richer objects through pattern persistence
Tags are provisional markers. A Tag can graduate to Topic or Thread (stays Emergent) or to Event / Organization / Location (promotes to Structural) when the pattern clarifies. Extends the settled "Tags grow into threads through use, not upfront design."
data-model → FP4→ FP5 13-05-2026
Insight as deliberate authoring
An Insight is a deliberate user-authored act asserting meaning — composed of an authoring input (itself stored as a Raw Input) plus zero or more bindings to existing objects. Distinct from Moment by being categorically human-authored.
data-model → FP1→ FP2→ FP4 15-05-2026