Skip to content

README.md

Latest commit

 

History

History
107 lines (81 loc) · 5.08 KB

README.md

File metadata and controls

107 lines (81 loc) · 5.08 KB
Raw

RSRM Random-String TLM Dataset

A standalone, self-contained knowledge dataset for the constraint-based random-string / password generator (randomstringllm / PassGen.Engine), authored in the native RSRM TLM format and compiled to .tlmz artifacts that load and validate inside the live RSRM runtime unchanged.

Nothing here depends on the full rsrm/ repo. The compiler/decompiler is the small self-contained PassGen.Tlm library; the .source.json files are the canonical source-of-truth; the .tlmz artifacts are byte-identical to what the real RSRM compiler would emit for the same source.

Layout

dataset/
  source/        *.source.json    generated TlmPackage source
  compiled/      *.tlmz           compiled artifacts (envelope + Brotli + JSON)
  decompiled/    *.decompiled.json round-trip output (verification / inspection)
  README.md

The source is authored in C# by PassGen.Tlm.Cli/DatasetAuthor.cs (tlm author) — no Python.

The bundle (7 linked TLMs)

TLM Role Prio Imports Covers
rs-char-classes Foundation 100 - Alphabet, the 4 classes, and every individual character; ambiguous set; safe/extended symbol pools
rs-constraint-spec Logic 120 char-classes The ConstraintSpec contract: length (exact/min/max), per-class allowed/min/max, exclude/include, feasibility
rs-operations Interface 110 constraint-spec Operations that narrow the spec: only / exclude / include / min / max / length / no-ambiguous / letters / alphanumeric / digits / symbols
rs-entropy Analysis 90 constraint-spec Strength model: bits = log2(valid-string count), charset size, crack time, strength bands, reference points, key insights
rs-generation Policy 130 constraint-spec, char-classes Deterministic algorithm: validate -> place minimums -> uniform fill -> validate/repair; CSPRNG vs seeded RNG; bias avoidance (5 policies)
rs-nl-vocabulary Interface 105 operations English phrasing -> constraint intents, with 12 trigger cues; ambiguity flag
rs-bundle Overlay 1000 all six Index TLM: domain root + module dependency DAG

Totals: 178 concepts, 229 relations, 358 parameters. All validate clean (0 errors, 0 warnings) and round-trip losslessly.

TLM format (recap)

A TLM is a TlmPackage:

  • Manifest - metadata (TlmId, Role, Priority, Version, Checksum, HotSwapPolicy, StabilityScore) + Imports (the dependency DAG).
  • Concepts - SymbolicConcept { Id, Label, Category, Description, Aliases, Properties }.
  • Relations - SymbolicRelation { SourceId, TargetId, Type, Description, Strength, ... }.
  • Dimensions / Policies / Cues / FitSignals / Generators - optional structured extras.

A .tlmz artifact is:

[16-byte envelope: "TLMZ" magic + uint16 major + uint16 minor + uint32 flags + uint32 reserved]
[Brotli-compressed UTF-8 compact JSON of the TlmPackage]

The Checksum is the SHA-256 of the compact JSON of the package with the stored checksum field cleared. RSRM recomputes it on load and rejects any mismatch. Because the PassGen.Tlm model classes are byte-faithful ports of Rsrm.Core.Models, this tool's checksums match RSRM's exactly - verified by loading real RSRM artifacts (including dictionary-core: 169,164 concepts / 662,057 relations) through this tool and confirming their stored checksums.

Build / verify

# one shot: build tool, compile all, decompile all, verify round-trip
./build-dataset.ps1

Or drive the tool directly (the tlm CLI lives in PassGen.Tlm.Cli):

dotnet run --project PassGen.Tlm.Cli -- compile   all   --root dataset
dotnet run --project PassGen.Tlm.Cli -- decompile all   --root dataset
dotnet run --project PassGen.Tlm.Cli -- validate  all   --root dataset
dotnet run --project PassGen.Tlm.Cli -- verify          --root dataset
dotnet run --project PassGen.Tlm.Cli -- list            --root dataset
dotnet run --project PassGen.Tlm.Cli -- stats rs-entropy --root dataset

verify proves losslessness: for each TLM it compiles the source, decompiles the artifact, recompresses, and asserts the bytes are identical AND the checksum is self-consistent. Compiles are reproducible (the source pins CreatedUtc), so the artifacts are deterministic across machines.

Loading into live RSRM (optional)

The .tlmz files are drop-in. Copy them next to RSRM's other compiled TLMs and add them to the boot config, or hot-load at runtime:

/load dataset/compiled/rs-bundle.tlmz

RSRM will validate the checksum, mount the graph, and resolve the Imports DAG.

Regenerating / growing the dataset

The .source.json files are generated by the C# author. To grow coverage (add a class alias or a cue synonym), edit PassGen.Tlm.Cli/DatasetAuthor.cs, then:

dotnet run --project PassGen.Tlm.Cli -- author --root dataset   # regenerate source
./build-dataset.ps1                                              # compile + verify