Open source / deterministic writing and presentation

PlainSpeak

PlainSpeak by PCGsoft makes machine-generated text clearer and more natural through deterministic transformation and review.

Last reviewed 2 September 2026

Active development — deterministic core / next-generation architecture

Not an AI paraphraser. The current source is a public, offline-first Python project with a documented transformation and integrity architecture.

What is PlainSpeak?

PlainSpeak is a deterministic presentation and humanisation layer for AI- or machine-generated text. It aims to make output clearer, less formulaic and easier to review without using a second LLM to make semantic decisions.

How does the current architecture work?

PlainSpeak Next separates transformation and core analysis from document handling, integrity checks, reporting and adapters. Declarative rules are matched against a stable projection of the original document, and transformations are applied only after mapping and conflict checks.

What evidence is implemented?

  • Deterministic rule loading, matching, conflict resolution and application.
  • Stable source spans and reproducible document behaviour across supported formats.
  • Integrity checks that can veto a structurally valid transformation.
  • Protected terms and preservation-aware document handling.
  • Regression and characterisation coverage that keeps the layers and outputs inspectable.

What is the preservation boundary?

The repository supports deterministic, protected transformations, but it does not evidence a shipped Reticle-specific adapter. For a future structured-output path, IDs, numbers, requirement strength, selected technologies, acceptance criteria, exclusions and unresolved items must remain protected semantic material rather than being rewritten as ordinary prose.

Human review remains part of the contract.

Deterministic behaviour makes a transformation reproducible and auditable; it does not by itself prove that every proposed wording change is appropriate for a high-stakes document.

Potential downstream role

Presentation after canonical planning.

ReticleProjectSpec and compiler outputs
PlainSpeakDeterministic transformation
Human readerReadable presentation and receipt

A future integration could present a canonical Project Bible or build plan through PlainSpeak, run an integrity check, and return a human-readable presentation. This is a potential relationship, not a claim that the Reticle integration is already shipped.

Future roadmap

Make the deterministic contract usable downstream.

  1. Finish and seal the deterministic core and preservation contract.
  2. Expose PlainSpeak as a headless service.
  3. Add MCP integration so agents and tools can submit text and receive transformed text, ruleset/version, input/output hashes and an integrity/audit receipt.
  4. Add structured document support, including DOCX, where the preservation model is explicit.
  5. Add human/manual editing surfaces later.

Future exploration: Spoken PlainSpeak may later apply the same principle to speech delivery and prosody while preserving intended meaning. No speech engine or WebMCP speech service is claimed here.

Source and lineage

Keep one project identity.

Portfolio relationship

A possible presentation companion.

Reticle Systems

Potential upstream source of canonical planning outputs; integration is not claimed as shipped.

FAQ

Does PlainSpeak use a second LLM to paraphrase text?

No. The current direction is deterministic transformation and review using defined rules, preservation constraints and integrity checks.

Does PlainSpeak guarantee semantic preservation for every structured project field?

No such Reticle-specific guarantee is claimed yet. The public source documents deterministic mappings and protected material; a structured downstream contract still needs to be exposed and sealed.

Does PlainSpeak include the future speech system?

No. Spoken PlainSpeak is only a later exploration, not an implemented speech engine or WebMCP speech service.