Soulforge

In plain words: Model trainer overview

Designed 2 min read 6 glossary terms 12 connected pages
The short version written for the Snayu team

The planned training area (Model Trainer, or Soulforge) that keeps an unbroken record from admitted data and base model through a training run to evaluated candidate weights, in four separate passages.

Why read it: Read this before any fine-tuning or LoRA work, to understand the required lineage and who decides promotion.
  1. 1Four passages: corpus admission, dataset compilation, training run, candidate handoff; none inherits permission from the last.
  2. 2Independent evaluation (Drift) shows eligibility; policy and human approval decide promotion; the serving owner routes it.
  3. 3Rollback cannot remove what weights already learned or undo completed effects.

Designed No operational training or serving path yet.

Words on this page, decoded click one for its card

EvaluatorLocal ModelModel Trainer
Where it sits in LychD

Model Trainer Soulforge

Training or fine-tuning models from approved data. A human decides promotion.

It's like: A training academy that hands graduates to HR for hiring decisions.

Click a dot to open that part's card.

Glossary page In the network

In this section 4 pages

Corpus Training data admission How a training purpose and nominated records become one admitted, immutable data snapshot and then a Dataset Manifest, with consent, license, privacy, lineage-safe splits and a sealed holdout. open → Training Run Running a training job How a training job takes exact pinned inputs (dataset manifest, base model digest, objective, recipe, resources) and produces a frozen candidate bundle plus a receipt, without touching serving or promotion. open → Candidate Handoff Evaluating and promoting trained models How a frozen trained candidate is independently evaluated, how an external decision may promote it with explicit limits, and how it is watched, rolled back or invalidated later. open → Discernment Training Training judgment from experience How approved experience from software work and human corrections could become training examples that teach judgment, such as when to ask versus act, and why training is the last resort after memory, context or procedure fixes. open →
The original text Underlined words open their plain-English card.

Long before the hammer falls, the metal has a history.

Training changes weights, but a useful result must also show what was taught and how the change was judged. SoulforgeModel Trainer keeps that history immutable: the admitted corpus and base-model digest lead through an objective, Recipe, and Training RunJob Record to evaluated candidate weights. Its tools may change; the evidence linking those stages must survive the strike.

SoulforgeModel Trainer is Designed; the passages below have no operational training or serving path yet. State of Work records delivery, and ADR 33 owns the law and complete record schemas.

SoulforgeModel Trainer may run during an admitted idle resource window, but sleep supplies no training or consolidation contract. The forge instead follows the explicit passage below.

The Four Passages

Route Receives Leaves behind Refuses
Corpus admission Training IntentRequest and nominated records Corpus Admission, exact membership, negative ledger, and sealed holdout Missing provenance, authority, privacy clearance, relevance, or a viable uncontaminated holdout
Dataset compilation One admitted snapshot Immutable Dataset Manifest with preserved lineage and splits Unadmitted material, changed dataset splits, or holdout leakage
Training run Manifest, base digest, objective, Recipe, and resource request Training RunJob Record receipt and frozen Candidate Bundle Unpinned inputs, a refused resource window, or unknowable retry state
Candidate handoff Frozen bundle and evaluation contract DriftEvaluator Outcomes and an externally owned Promotion Decision Trainer self-grading, mutable custody, or unowned serving

No passage inherits permission merely because the previous one succeeded. Orchestrator decides the physical transition needed for resources. Independent Drift evaluation can establish eligibility, after which owning policy and required MagusYou, the Operator/HitL authority decide promotion. Only the serving owner can register and route the exact promoted digest.

Observation may trigger rollback or lineage invalidation. It cannot rewrite an earlier record, erase completed effects, or remove learned influence from weights already produced.

Discernment training applies these passages to SDLC experience and human corrections: preserve the reason and conditions of a choice, then test transfer through an independent evaluation. Ouroboros connects that slower return to ordinary ContextBriefing Pack, memory and workflow improvement.