The Symbol: Sentinel, Not Mascot
The cat silhouette is the heraldic symbol of TELHAR. It signifies observation, patience, intelligence, independence, precision, adaptation, and disciplined curiosity. It is never an illustration, never an anthropomorphized character, and never a mascot. It is a still, seated presence built on an 8U grid armature with √2 proportions.
The Twelve Principles (P01–P12)
Timeless before Trend
Permanence has priority over fashion. No design or architectural decision may depend on a current ephemeral trend to make sense. Built to endure to 2056.
Architecture before Ornament
Structure prevails over decoration. Every visible element must carry structural or informational load. If an element can be deleted without breaking meaning or function, delete it.
Clarity before Complexity
Clarity is a form of intelligence. Complexity is never a signal of sophistication. A first-time reader must state the point in one sentence within 5 seconds.
Systems before Products
We design capabilities and integrated systems, not isolated gadgets. Every asset is an instance of a coherent system, never a hand-tuned one-off.
Precision before Expression
Every element answers to functional and geometric reasons. No free-hand curves: every form is constructed from arcs, radii, and exact mathematical proportions.
Human Responsibility
Technology amplifies people; it never replaces, imitates, or impersonates them. The brand never anthropomorphizes machine intelligence.
Evidence before Opinion
Decisions rest on verifiable principles, published standards, and empirical tests — not subjective taste. Claims are cited or measured.
Quiet Confidence
Authority is expressed through restraint, not ostentation. No neon, no volumetric glow, no hype, no exclamation marks. Calm technology signals mastery.
Enduring Identity
The mark and wordmark must remain recognizable across scale (16px to 2048px) and across decades, anchored by a fixed geometric core.
Coherent Evolution
The system evolves without discarding its essence. Change is versioned, rationalized, and backward-legible.
One Source of Truth
Every fact — a token value, an architectural invariant, a principle — is defined in exactly one place and referenced everywhere else. Zero duplication.
Legible to Humans and Machines
The system is authored so that both human engineers and AI agents can retrieve rules and enforce them deterministically.
The Prohibitions (Negative Space of the Brand)
These prohibitions are machine-checkable laws. Violating them breaks the identity, regardless of aesthetic appeal.
Architectural Principles (P1–P6)
When formal rules leave ambiguity, these non-negotiable architectural principles govern engineering judgment across all repositories.
Consolidate on Capabilities, Not Names
The owner of an architectural plane is the component that already satisfies the most invariants, not the newest or most theoretical. We collapse to the real.
Operation Wins over Abstract Isolation
Three well-integrated components are vastly superior to five perfectly decoupled but inoperable services. Runtime execution precedes theoretical elegance.
Reduce Entropy in Every Change
If a change does not reduce platform entropy, it must strictly justify neutrality. No commit or pull request is permitted to increase architectural entropy.
Reversibility Over Velocity
Prefer changes with demonstrable rollback. Irreversible actions require explicit human signoff and two-phase commit approval gates.
Evidence before Authority
No architectural decision stands on hierarchy, consensus, or seniority. It stands solely on reproducible, falsifiable evidence.
Strangler Migration over Big-Bang Rewrite
Legacy components are decommissioned incrementally behind Anti-Corruption Layers (ACL). Monolithic or speculative rewrites are strictly prohibited.
The Architecture Constitution (Articles ART-1 to ART-17)
Derived from Document 35. These are machine-evaluated predicates. Violations fail CI pipelines (block-merge) or deployment gates (block-deploy).
Every capability has exactly one owner declared in .arch-owner.yaml.
Every concept has exactly one canonical contract ID ($id) and root schema.
Zero parallel duplicate runtimes or redundant implementations per capability.
All evidence events hash identically across TypeScript, Rust, and Python.
Zero local policy evaluation. All authorization evaluates strictly inside security/foedus.
No direct LLM provider SDK calls outside the dedicated Execution Layer.
All dynamic model routing and fallback decisions flow exclusively through ai/via.
Zero pip, npm, or apt installations at container runtime. Fully hermetic builds.
Systems & AI Engineering Laws
Invariants governing autonomous agents, cryptographic proof, and human oversight in production.
Evidence is the Product
If a claim cannot be reproduced by a third party from an immutable transparency log without our cooperation, the claim does not ship.
No Black-Box Autonomy
Every autonomous action records (input_hash, output_hash, model_id, model_version, policy_hash, actor_svid, timestamp) for external auditability.
Fail-Safe Physical Default
On uncertainty or ambiguous policy, agents halt rather than improvise. Deny is the safe default across every effect gate.
Models are Policy-Bounded
Models operate inside bounded contexts with declared input schemas, strict latency budgets, and zero self-granted privileges.
The Three Refusals
- ✓ We do not sell AI — we design intelligent systems.
- ✓ We do not build products — we build capabilities.
- ✓ We do not automate processes — we architect ecosystems.