Alefba compiles source context into typed, source-grounded commitments, builds a budgeted candidate for a declared model and harness profile, and reports what was selected, omitted, transformed or remains unverified.
Alefba—الفبا—means alphabet in Persian: a shared structure that lets different voices become legible without making them identical.
01MARK
02BRICK
03ALPHABET
04CODE
05APIR
THE PREMISE
Build the commons before intelligence becomes a monopoly.
Alefba begins with a possibility, not a prediction: as machine capability reaches decisions once reserved for human judgment, society will need more than larger compute. It will need open semantic infrastructure that keeps meaning inspectable, portable and accountable across models.
Founded by Danial Samiei (دانیال سمیعی), Alefba is an open international development program. It invites researchers, builders and institutions to test a layer where diverse human knowledge can meet machine intelligence without being owned by one set of weights or one legal regime.
A
Access without a technological caste.
Reduce dependency on any single model, lower the cost of useful AI and keep advanced capability from hardening the divide between those with access and those without it.
B
Ethics without a single moral center.
Carry tolerance, restraint, uncertainty and plural moral provenance into systems of growing capability instead of allowing one model, company or jurisdiction to become the default conscience.
C
Culture without reduction.
Preserve language, context and irreducible human values as first-class commitments, so machine-readable does not become culturally flattened or falsely universal.
THREE PARTS / ONE OPEN STRUCTURE
The mark / a working interpretation
A triangle built from memory, imperfection and measure.
01
The first mark
The form evokes the bovine figures among humanity's earliest surviving cave images: a reminder that representation began long before computation.
02
Imperfect parts, coherent whole
Its local asymmetry resists sterile perfection; its global order suggests natural systems, bounded intelligence and a larger field of awareness.
03
Three building units
Three brick-like fragments resolve into a triangle: geometry, construction and the smallest repeatable units from which shared structures can grow.
The mark is an invitation to interpretation, not a historical claim of firstness.
01 / The compilation path
How it works?
Alefba does not merely shorten a prompt. It creates a traceable intermediate representation, renders a profile-bound candidate and records verification limits. Safety, model uptake and outcome remain separate evidence questions.
01TRANSFORM
Sense
Locate goals, constraints, decisions, evidence and unresolved ambiguity.
02TRANSFORM
Structure
Compile meaning into canonical, source-linked APIR commitments.
03TRANSFORM
Pack
Select a budgeted candidate under declared profile, risk and omission rules.
04RECEIPT
Assure
Report structural coverage and omissions; route critical unknowns to a declared fallback.
INTERACTIVE TRACE
One compiler. Different operational pressure.
ACTIVE TARGETCoding model / repository harness
SOURCE BUFFER03 LINES
01
Return one runnable HTML file.
GOAL
02
No external libraries.
CONSTRAINT
03
Canvas selected; SVG rejected.
DECISION
EXTRACT→
APIR COMMITMENTS3 / 3 SELECTED
C-01required
OUTPUT_CONTRACT
single-file / runnable
C-02critical
CONSTRAINT
dependency_count = 0
C-03required
DECISION
renderer = canvas
VERIFY→
TARGET INVARIANTNOT YET PROVEN
DESIGN TARGET
Target invariant: rejected implementation choices remain explicit after rendering.
IF LOST
The model can reopen SVG or silently add dependencies.
SAFE FALLBACKCanonical commitment + source span
02 / Assurance fabric
One score cannot carry four kinds of truth.
Alefba separates what is structurally intact from what a model actually uses and what the final outcome proves. Unknown remains unknown.
I
Integrity
Do the bytes, fingerprint and commitment IDs agree?
V0 / referenceP
Representation
Can the commitment be recovered from the rendered pack?
V1 / plannedU
Uptake
Does this model use it across order and wording changes?
V2 / empiricalO
Outcome
Does the result actually satisfy the commitment?
V3 / observed
PASSFAILUNKNOWN ≠ PASS
03 / Claim contract
Every claim earns its tense.
Reference behavior is stated in the present. Product milestones are stated in the future. Research stays conditional until a frozen run, strong baseline and full-cost result exist.
NOWREFERENCE
Make the transformation inspectable.
Tagged extraction, canonical APIR, deterministic candidate selection, three renderers and a V0 structural receipt.
The echo API does not prove provider parity, semantic equivalence, token savings or task improvement.
Alefba represents, compiles and verifies semantic commitments. It does not grant permissions, execute workflows or claim that compression is universally lossless. APIR can travel through MCP or APIs; it does not replace their transport or authority model.
Canonical APIR / profile-bound codecs
MCP and API interoperability
Metadata-first observability
Reproducible evidence and explicit unknowns
FIELD INDEX / 01-06
Explore the open program
The Semantic Atlas
Six routes through Alefba: why it exists, what it specifies, what it can prove, and where the work remains open.
