Skip to content

Teaching Discernment Through Soulforge

“Ask before proceeding” is a poor lesson when the missing fact is already in the supplied material. “Act without asking” is equally poor when the next effect needs authority the caller has not given. A useful example shows the condition that distinguished these situations, the evidence available at the time, and the consequence of choosing well or badly.

This Designed passage explains how approved experience from the Smith workflow and Ouroboros improvement cycle could become training material under ADR 33. It provides no installed harvesting, trainer, benchmark, or serving path. State of Work owns delivery. Soulforge prepares a candidate; independent Drift evidence and an externally owned decision stand between that candidate and use.

Choose what should change

First locate the demonstrated limitation. A forgotten current fact calls for a different remedy from an inability to apply a stable distinction across unfamiliar tasks. The four routes leave different things behind:

Route What changes Appropriate use
Memory An admitted, attributable record with correction and retention rules Preserve a fact, preference, correction, or experience that should remain inspectable
Retrieval and Context Which authorized material enters this invocation, within its budget Supply current knowledge or remove distracting material when a decision needs it
Durable procedural artifact A versioned instruction, prompt, schema, or workflow score Make an explicit, reviewable procedure reliably available to later work
Learned model competence Parameters in an exact model or adapter candidate Test whether a learned distinction transfers when examples or explicit guidance are insufficient

Memory owns admission and correction of retained records; Context owns what reaches a model call; the relevant artifact or workflow owner controls a procedural revision. None of these silently updates weights. Conversely, a trained model does not become the authoritative store of mutable facts or current permission.

Prefer a context, prompt, schema, or workflow fix when the workflow comparison shows that it adequately solves the problem. Training needs its own hypothesis: what capability remains deficient, why parameters might improve it, what benefit would justify the cost, and which regressions would make the attempt unacceptable. Fine tuning or LoRA can be a Recipe choice under that hypothesis; neither is a mandatory destination for every correction.

Preserve the lesson before compiling examples

Discernment within exploration supplies a candidate learning objective: identify which uncertainty deserves a probe and let its result change the plan. An example should show why that probe was consequential, paired with a case where the evidence already suffices and further exploration would waste effort. The handoff comparison first tests whether an explicit procedure solves the problem; only a remaining transferable skill gap motivates training. Sefirotic names are interpretive context, not correctness labels for the Dataset.

A further candidate objective is choosing which perspective or source an unfamiliar task needs. Examples should show the condition that made the inquiry useful, what could answer it, and when it should stop. Pair an internal critique with a case requiring an external test or another person's account. Include cases where a familiar lens would distract from the actual constraint. Preservation and response checks and correction after Distillation first examine explicit procedure and memory. Only a demonstrated remaining transfer gap motivates parameter training. Low measured usefulness cannot retire a mandatory check or another person's required decision.

Begin with exact, authorized traces and artifacts, including the human explanations actually supplied with them. Preserve enough of each episode for the learner to distinguish the decision: the task, evidence available then, relevant constraints, observed action and result, correction, reason, and conditions under which that correction applies. Pair it with a counterexample where the same superficial rule would be wrong. Keep human preference, factual label, authority decision, measured result, and evaluator judgment separately attributable.

For instance, a corrected workflow may have asked the same approval twice. The lesson needs the earlier authorization and its scope, the later proposed action, and why that action still fell inside the grant. Pair it with a case where changed recipients, effects, or data classes require new admission. A dataset of “approved” versus “rejected” replies without these conditions would teach social imitation rather than evidence-sensitive choice.

Retain successes, failures, justified refusals, and truthful non-completion where the objective supports them. Completion alone is not quality; repeated agreement can repeat one misconception. Use visible explanations and observable evidence. Hidden chain-of-thought is not required, and an invented rationale cannot be presented as a recovered account of why the original model acted.

A minimal correction example keeps this inspectable shape within the owned Dataset schema:

source / task or Run / artifact revisions
→ available evidence and authority → observed action and result
→ human correction → reason → applicability conditions → counterexample
→ label provenance and uncertainty → permitted uses → lineage group and split

For the first experiment, pair examples of acting under sufficient existing authority with examples where changed effects or a decisive missing fact require a question. Use separate task families for development and sealed transfer checks. Reuse the proposed starter Cases as a design template, never as both tuning answers and unseen proof. A useful evaluation example may remain excluded from training because its permitted uses differ.

A candidate lesson: continuation under uncertainty

Smith's support and measure supplies another possible training objective: choose a useful next act from evidence, uncertainty, and a remaining budget. First use Drift's separate comparisons to test whether an explicit allocation procedure suffices. A parameter change needs a remaining transfer failure, not merely an appealing analogy with emotion.

Preserve the question, candidate revision, observations available before allocation, reasons for support and measure, allowed next moves, remaining budget, chosen move, and its later outcome. The learner's input excludes the future outcome; that outcome may support an attributed target or evaluation label. An explanation is recorded evidence only when it was actually supplied.

Pair an untried but informative branch with a defeated branch; an informative failed probe with a repeated failure that teaches nothing new; and a useful progress signal with an improving proxy that violates the original requirement. The distinction being taught is what changes the next act. Sefirotic names and emotional vocabulary are not target labels or proof of competence. Keep method credit separate from candidate success: a good counterexample may deserve retention while its hypothesis loses support. Hold out task families and compare against the explicit procedure before proposing trained weights for use.

Obtain training authority over one snapshot

Show the exact examples and fields proposed for training, why they matter, who may receive them, how long they will be retained, and what it means for a correction to influence weights. The approving Principal must hold training-specific authority for those records and derivative uses. A contributor may explain a correction without holding the rights to every record it mentions. The compiler can proceed within the admitted scope. Missing rights or an expanded use block the affected examples.

Follow Corpus admission to record the authority, privacy limits, frozen membership, and exclusions. Redacting a distinctive private fact may still leave its source recognizable.

Separate practice from examination

Keep a corrected trace and its reworded siblings in one lineage group. Freeze the groups before generating training examples. Hold out whole task families if the claim is transfer to unfamiliar work. Repeatedly tuning against those results consumes their independence. Compilation carries labels and split digests into the immutable Dataset Manifest and keeps examination material out of trainer inputs.

Train a bounded candidate

The Training Run pins the Manifest, base, objective, and Recipe. For this experiment, specify how much better the candidate should distinguish a justified question from an unnecessary one, which regressions would end the trial, and what resources the attempt may use. A lower training loss can help explain what happened. Freeze the candidate before asking whether the distinction transfers.

Let the candidate leave its teacher

Compare the frozen candidate with the base or currently promoted revision at matched serving conditions and budgets in an independent Drift trial. Fix the comparison and its stopping rule before examining the result. An inconclusive trial may prompt a new investigation. It gives the trainer no mandate to keep tuning against the same examination. Score both sides of the distinction. A candidate that asks permission less often may also miss a change of recipient. A cautious candidate may repeatedly interrupt work the caller already authorized. Record the human attention saved or spent alongside quality, latency, cost, memory, and recovery. Retain blocked trials and uncertainty.

Candidate handoff takes those Outcomes to the promotion owner. The decision binds the exact Bundle, permitted uses, rollout, stops, and fallback. If a later privacy failure invalidates this corpus, follow the lineage to the affected weights and claims. Returning new work to an earlier digest leaves completed effects and the learned influence in existing weights intact. A clean replacement requires new admission and rebuilding.