Skip to content

COMPACT_RUNTIME_PROFILE_V1.md

Latest commit

 

History

History
172 lines (109 loc) · 3.08 KB

COMPACT_RUNTIME_PROFILE_V1.md

File metadata and controls

172 lines (109 loc) · 3.08 KB
Raw

COMPACT_RUNTIME_PROFILE_V1

Status: experimental
Date: 2026-06-26

Purpose

Define the first compact runtime profile for GLYPH.

This profile separates heavy build artifacts from deployed runtime artifacts.

The goal is to reduce runtime disk footprint while preserving:

  • exact-byte query
  • FM interval
  • match count
  • sampled locate offsets
  • byte-check/evidence compatibility

Problem

The earlier public evidence path kept dense and temporary build artifacts in the same directory:

  • fm.bin
  • sa.bin
  • corpus.sentinel.bin
  • corpus.bin
  • chunk_map.bin

This made GLYPH appear to require 8x-12x disk overhead.

That is not a fundamental requirement of query runtime.

It is a layout/profile problem.

Runtime artifact set

Compact Runtime Profile V1 keeps only:

  • bwt.bin
  • fm_core.bin
  • locate_core_s16.bin
  • manifest.json
  • optional compact_runtime_manifest_v1.json

Forbidden runtime artifacts

The following files are build artifacts and must not be required in a compact runtime directory:

  • fm.bin
  • sa.bin
  • corpus.sentinel.bin
  • corpus.bin
  • chunk_map.bin

Query path

Query/count/FM interval runs from:

  • fm_core.bin
  • bwt.bin

using:

  • query_fm_core_v1

Dense fm.bin is not required.

Locate path

Offset recovery runs from:

  • fm_core.bin
  • locate_core_s16.bin
  • bwt.bin

using:

  • locate_backend_v2
  • glyph_locate_offsets_v0.py

Full sa.bin is not required at runtime.

Local validation: Pizza 50MB

Source demo directory:

examples/public-evidence-demo/work/pizza_english_50mb

Compact runtime candidate:

/tmp/glyph_compact_runtime_pizza_50mb_c2048_s16_runtime_only

Build parameters:

  • checkpoint_step: 2048
  • sample_step: 16

Runtime files:

  • bwt.bin: 50,000,001 bytes
  • fm_core.bin: 50,008,088 bytes
  • locate_core_s16.bin: 50,000,040 bytes
  • manifest.json: 736 bytes

Runtime total:

  • 150,008,865 bytes

Ratio vs original 50MB corpus:

  • about 3.0x

Forbidden files were absent:

  • fm.bin
  • sa.bin
  • corpus.sentinel.bin
  • corpus.bin
  • chunk_map.bin

Query:

Ten Days that Shook the World

Observed compact query result:

  • bwt_bytes: 50000001
  • checkpoint_step: 2048
  • num_checkpoints: 24416
  • fm_interval: [12587658, 12587659]
  • match_count: 1
  • count: 1

Observed compact locate result:

  • offset_mode: locate_backend_v2
  • offsets: [53]
  • total_steps: 5
  • max_steps: 5
  • ok: true

Meaning

This proves that dense fm.bin is not required for query runtime.

It also proves that full sa.bin is not required for locate runtime.

The next engineering objective is to measure the size/latency tradeoff over:

  • checkpoint_step: 1024, 2048, 4096, 8192
  • sample_step: 16, 32, 64, 128

Non-claims

This profile does not yet solve:

  • cold build time
  • full compressed BWT storage
  • r-index
  • PFP construction
  • cryptographic inclusion proof
  • non-membership proof
  • verifier without corpus or commitment

Current conclusion

GLYPH's 10x+ disk bloat was not a fundamental law.

It was caused by dense runtime layout.

Compact Runtime Profile V1 is the first practical path toward a smaller industrial runtime profile.