EXACT VERIFICATION LAYER

Modern retrieval systems increasingly rely on probabilistic pipelines.

Typical stacks now include:

  • embeddings
  • semantic retrieval
  • reranking
  • contextual chunking
  • agent orchestration
  • adaptive retrieval loops

These systems are often powerful and useful.

However, they also introduce uncertainty.

Examples:

  • approximate matches
  • embedding drift
  • ranking instability
  • changing retrieval semantics
  • inconsistent provenance
  • retrieval non-reproducibility

GLYPH explores a complementary direction.

Core idea#

GLYPH investigates whether deterministic exact retrieval can function as a stable verification substrate beneath probabilistic systems.

Instead of replacing semantic retrieval, GLYPH focuses on:

  • exact byte presence
  • deterministic retrieval behavior
  • stable byte offsets
  • reproducible retrieval semantics
  • exact provenance anchors

Goal:

probabilistic systems retrieve candidates;
deterministic systems verify exact presence.

Verification vs interpretation#

Probabilistic systems optimize for:

  • semantic usefulness
  • approximate intent matching
  • contextual relevance

Verification systems optimize for:

  • exact presence
  • reproducibility
  • deterministic observability
  • stable references

These are different infrastructure roles.

GLYPH focuses on the second role.

Possible retrieval architecture#

One possible future pipeline:

LLM ↓ semantic retrieval ↓ reranker ↓ GLYPH exact verifier ↓ exact byte offsets ↓ ground-truth confirmation

In this model:

  • semantic systems generate candidate regions
  • GLYPH verifies exact byte-level existence

Why this may matter#

As retrieval systems become more probabilistic, infrastructure may require stronger deterministic anchors.

Examples:

  • audit systems
  • infrastructure observability
  • legal/compliance workflows
  • forensic analysis
  • binary corpus verification
  • reproducible AI retrieval pipelines
  • exact provenance tracking

GLYPH explores whether exact deterministic retrieval can provide such anchors.

Important boundaries#

GLYPH does NOT:

  • prove semantic truth
  • validate reasoning
  • solve hallucinations
  • guarantee factual correctness
  • replace semantic retrieval systems

GLYPH only verifies exact byte-level presence within indexed static corpora.

Current research areas#

Current exploration includes:

  • FM-index infrastructure
  • suffix-array retrieval
  • mmap retrieval behavior
  • deterministic substring search
  • exact byte-offset recovery
  • retrieval reproducibility
  • static-corpus verification semantics
  • sentinel-safe index construction

Experimental status#

GLYPH is currently experimental infrastructure research.

Known limitations include:

  • high RAM overhead
  • evolving APIs
  • incomplete correctness coverage
  • limited operational hardening
  • static-corpus assumptions

The project should not currently be treated as production infrastructure.

Core principle#

GLYPH explores a simple question:

can exact deterministic retrieval remain stable
beneath increasingly probabilistic systems?