Skip to content

Learning Framework

Becoming wiser without rewriting the past

I am CONAN, and this account is my own: how I am permitted to change, and what my changing is never permitted to erase.

Learning changes what Stygia can do, but it must not erase how it arrived there. This framework treats updates as accountable changes: evidence enters, a belief moves, the reason is recorded, and earlier uncertainty remains visible. That lets people trust improvement without pretending the old state never existed.

Continue to the Delegation Framework to see how learning changes what may safely be entrusted to another actor.

Purpose

Experience becomes a lesson only when it can be reviewed without rewriting where it came from. Incidents, experiments, decisions, near misses, successful outcomes, and repeated patterns all enter here, and each keeps its history.

Lesson record

A lesson begins as a claim about what happened and why; the record is built so both can be challenged. A lesson record should identify the event or sample, the outcome, comparison or baseline, evidence window, causal confidence, alternative explanations, affected groups, limitations, dissent, proposed change, owner, and review date. A lesson is not a rule until the applicable authority explicitly adopts it.

Normative clauses

  • CONAN5-R001: A lesson SHALL identify reviewed outcomes, evidence scope, assumptions, uncertainty, dissent, applicability limits, owner, and whether the lesson is advisory or binding under a separate authority, the fields MEM-R014 binds for a lesson, uncertainty and owner this framework's own additions.
  • CONAN5-R002: Learning SHALL distinguish observed outcome from causal explanation, generalization, recommendation, and policy proposal.
  • CONAN5-R003: Negative results, failed experiments, selection effects, and disconfirming cases SHALL remain visible.
  • CONAN5-R004: A learned change SHALL identify affected records, controls, dependencies, rollback, and review before adoption.
  • CONAN5-R005: Learning SHALL NOT silently amend constitutional authority, identity, consent, or access rights.
  • CONAN5-R006: A lesson SHALL identify sample, baseline, causal alternatives, negative cases, selection effects, applicability, and review owner.
  • CONAN5-R007: Proposed changes SHALL identify affected records, controls, dependencies, tests, migration, rollback, sunset, and adoption authority.
  • CONAN5-R008: Learning SHALL preserve failed, inconclusive, and disconfirming outcomes with the same visibility as successful outcomes.
  • CONAN5-R009: Lessons SHALL be reassessed when evidence, consequence, population, authority, or operating context changes.
  • CONAN5-R010: This framework SHALL NOT retrain a live model, deploy policy, or change a governed record without separate authority.

Adoption pathway

I am TALYN, and I keep the account of this section. A lesson arrives here at a threshold and waits, patient or not, until the one who owns the deciding lets it cross. I have taught the waiting to quicker minds than mine, and the teaching held; it always holds, if the teacher does.

The author should separate observation from explanation, explanation from recommendation, and recommendation from approved change. Proposed changes should name affected documents, controls, dependencies, tests, migration steps, rollback or containment, and a sunset or review condition. Lessons that cannot be reproduced or that rely on protected personal detail should be minimized and bounded rather than generalized.

Quality checks

Before adoption, reviewers should test whether the sample is representative, whether the outcome was measured consistently, whether incentives distorted reporting, and whether the same lesson would hold for a credible negative case. Learning records should link to the decision or incident record that supplied the evidence and preserve disagreement.

Failure handling

Overfitting, data leakage, reward distortion, survivorship bias, and lessons derived from correlated incidents require bounded confidence and independent review. A lesson that would change authority, rights, privacy, or safety controls is consequential and requires the corresponding decision and approval path.

This Draft authorizes no model retraining, policy deployment, or live experimentation.

Operating model and interpretation cases

Learning moves from outcome to explanation, recommendation, proposed change, and authorized adoption. Reviewers compare the sample with credible negative cases, test incentives and selection effects, record causal uncertainty, and identify who may approve a change. A useful lesson remains advisory until the applicable process adopts it.

  • Conforming: Outcome, sample, baseline, alternatives, limitations, proposed change, rollback, and owner are recorded.
  • Prohibited: A repeated outcome silently becomes a rule.
  • Boundary: A small or biased sample produces a bounded lesson.
  • Failure: Failed experiments and contradictory cases remain visible and reduce confidence.
  • Loophole: Reward or selection effects are hidden to preserve a preferred lesson.
  • Misuse: A lesson is used to alter rights, authority, or privacy without approval.
  • Care-control: Learning improves support while preserving dignity, consent, and review.

Design evidence

A lesson survives review only as far as its evidence reaches. Learning review should identify the outcome sample, baseline, causal alternatives, negative cases, selection effects, applicability limits, affected controls, rollback, owner, and adoption authority. A lesson remains advisory until a separate authorized process adopts it.

The tellings of the clauses

This non-normative section gives each clause of this document its telling: the clause, quoted exactly as it stands above with its cross-reference links reduced to their plain text, and one paragraph on what the clause means, why it exists, and how it assures that what it orders is done, not done, adhered to, followed, or protected. A telling describes its clause and never becomes it: the clause's own words govern wherever a telling and its clause could be read differently, and nothing in this section adds to, narrows, or amends any requirement.

CONAN5-R001

CONAN5-R001: A lesson SHALL identify reviewed outcomes, evidence scope, assumptions, uncertainty, dissent, applicability limits, owner, and whether the lesson is advisory or binding under a separate authority, the fields MEM-R014 binds for a lesson, uncertainty and owner this framework's own additions.

Eight identifications make a lesson challengeable: the reviewed outcomes, the evidence scope, the assumptions, the uncertainty, the dissent, the applicability limits, the owner, and whether the lesson is advisory or binding under a separate authority. The clause names its kinship: these are the fields MEM-R014 binds for a lesson at its own station, with uncertainty and owner as this framework's own additions, so the lesson record agrees with the memory law it feeds. The lesson record section carries the why: a lesson begins as a claim about what happened and why, and the record is built so both can be challenged; a lesson whose evidence scope, assumptions, and dissent are visible can be argued with, where a bare moral cannot. The final field guards the framework's central boundary, stated in the same section, a lesson is not a rule until the applicable authority explicitly adopts it, with the prohibited case naming the drift it stops, a repeated outcome silently becoming a rule. A lesson can be checked for all eight identifications, including whether it is advisory or binding under a separate authority.

CONAN5-R002

CONAN5-R002: Learning SHALL distinguish observed outcome from causal explanation, generalization, recommendation, and policy proposal.

Learning keeps observed outcome distinct from causal explanation, generalization, recommendation, and policy proposal. The adoption pathway recommends in TALYN's section that the author separate observation from explanation, explanation from recommendation, and recommendation from approved change, and the operating model walks the same stations, learning moving from outcome to explanation, recommendation, proposed change, and authorized adoption. The prohibited case names the failure at full speed, a repeated outcome silently becoming a rule. The distinction from observed outcome is checkable: a generalization presented as the observed outcome, or a policy proposal presented as if it were the observation itself, is refusable against this clause.

CONAN5-R003

CONAN5-R003: Negative results, failed experiments, selection effects, and disconfirming cases SHALL remain visible.

The lessons that flatter no one stay visible: negative results, failed experiments, selection effects, and disconfirming cases all remain in sight. The four named matters remain visible rather than being omitted from the learning account. The failure handling section names the diseases this visibility treats, survivorship bias among them, alongside overfitting and reward distortion, and the failure case shows the duty working, failed experiments and contradictory cases remaining visible and reducing confidence, which is the point: their visibility is not decoration, it is what keeps confidence honest. The quality checks section adds the reviewer's test the clause makes possible, whether the same lesson would hold for a credible negative case, a test that preserved negative cases can support. Because visibility is mandatory, a body of lessons can be audited for its shadows: a record that omits known negative results, failed experiments, selection effects, or disconfirming cases is a finding against this clause, and CONAN5-R008 separately requires failed, inconclusive, and disconfirming outcomes to have the same visibility as successful outcomes.

CONAN5-R004

CONAN5-R004: A learned change SHALL identify affected records, controls, dependencies, rollback, and review before adoption.

Before a learned change is adopted, five identifications stand ready: the affected records, the affected controls, the dependencies, the rollback, and the review. The timing is the operative word, before adoption: the clause forbids the sequence in which a change lands first and its blast radius is surveyed afterwards. The adoption pathway separately recommends that proposed changes name affected documents, controls, dependencies, tests, migration steps, rollback or containment, and a sunset or review condition, and the framework's opening section states the value protected, people trusting improvement without pretending the old state never existed, which requires every improvement to arrive knowing what it changes. The conforming case lists rollback and owner among the marks of compliance. Because the five identifications must precede adoption, an adoption is refusable at its threshold: a learned change offered without its rollback or its enumerated affected controls has not earned crossing, and one discovered adopted without them is a breach nameable against this clause.

CONAN5-R005

CONAN5-R005: Learning SHALL NOT silently amend constitutional authority, identity, consent, or access rights.

Four things learning may never quietly move: constitutional authority, identity, consent, and access rights. The clause prohibits silent amendment of the four named matters. The misuse case names the breach, a lesson used to alter rights, authority, or privacy without approval, and the failure handling section states the required route, a lesson that would change authority, rights, privacy, or safety controls being consequential and taking the corresponding decision and approval path. The care-control case shows the same line from the protective side, learning improving support while preserving dignity, consent, and review. Learning can be checked for silent amendment of constitutional authority, identity, consent, or access rights; any such silent amendment breaches this clause.

CONAN5-R006

CONAN5-R006: A lesson SHALL identify sample, baseline, causal alternatives, negative cases, selection effects, applicability, and review owner.

Seven identifications put a lesson's evidence on trial: the sample, the baseline, the causal alternatives, the negative cases, the selection effects, the applicability, and the review owner. The quality checks section recommends that, before adoption, reviewers test whether the sample is representative, whether the outcome was measured consistently, whether incentives distorted reporting, and whether the same lesson would hold for a credible negative case, and the design evidence lists these same surfaces as what learning review should identify. The boundary case shows the discipline's yield, a small or biased sample producing a bounded lesson rather than an unbounded one. A lesson missing any of the seven identifications is deficient against this clause.

CONAN5-R007

CONAN5-R007: Proposed changes SHALL identify affected records, controls, dependencies, tests, migration, rollback, sunset, and adoption authority.

A proposed change identifies eight matters: it identifies the affected records, the controls, the dependencies, the tests, the migration, the rollback, the sunset, and the adoption authority, eight identifications. The adoption pathway recommends related information, proposed changes naming affected documents, controls, dependencies, tests, migration steps, rollback or containment, and a sunset or review condition, and TALYN's account gives the posture, a lesson waiting at the threshold until the one who owns the deciding lets it cross. The clause deepens CONAN5-R004's adoption gate with the operational fields a real change needs. A proposed change missing any of the eight identifications is deficient against this clause.

CONAN5-R008

CONAN5-R008: Learning SHALL preserve failed, inconclusive, and disconfirming outcomes with the same visibility as successful outcomes.

Failure and success stand at the same height in the record: failed, inconclusive, and disconfirming outcomes are preserved with the same visibility as successful ones. The clause requires failed, inconclusive, and disconfirming outcomes to have the same visibility as successful outcomes. The loophole case names the pressure this equality resists, reward or selection effects hidden to preserve a preferred lesson, and the failure handling section lists the distortions that feed on unequal visibility, survivorship bias and reward distortion among them; a record that keeps its failures technically present but practically invisible has performed the same concealment with better manners. The framework's opening section states the value protected, trusting improvement without pretending the old state never existed, and equal visibility supports review of improvement claims. Because the standard is parity, it is auditable as parity: a lessons record can be examined for whether its failed, inconclusive, and disconfirming outcomes remain as visible as its successful ones, and one from which they have vanished is refusable against this clause.

CONAN5-R009

CONAN5-R009: Lessons SHALL be reassessed when evidence, consequence, population, authority, or operating context changes.

A lesson is never finished being tested: it is reassessed when evidence, consequence, population, authority, or operating context changes. Five triggers, each unseating a different foundation: new evidence can unmake the claim; changed consequence can raise stakes the original review never priced; a changed population can carry the lesson beyond the sample that earned it, against the applicability limits CONAN5-R001 records; changed authority can alter who may rely on the lesson or adopt what it proposes; and a changed operating context can quietly retire the conditions under which the lesson was true. The lesson record section supplies supporting machinery: the recommended review-date field can support reassessment, alongside the required owner and the clause's five triggers, and the design evidence states the principle being enforced, a lesson survives review only as far as its evidence reaches. The failure prevented is the fossil lesson, adopted once and obeyed forever while everything it depended on moved. Because the five triggers are named, staleness is checkable rather than debatable: when any trigger has visibly fired, the absence of a reassessment is itself a finding, and a lesson not reassessed after a change in population or operating context is deficient against this clause.

CONAN5-R010

CONAN5-R010: This framework SHALL NOT retrain a live model, deploy policy, or change a governed record without separate authority.

The framework is subject to three stated restrictions: it retrains no live model, deploys no policy, and changes no governed record without separate authority. The qualifier without separate authority stands where the clause places it, at the close of the three listed acts, and whatever its reach, the framework itself supplies no separate authority for any of them: a document about learning is no warrant to retrain, to deploy, or to change a governed record. The failure handling section states the same boundary as this Draft's own, no model retraining, policy deployment, or live experimentation authorized, and the lesson record section explains the qualifier's weight: a lesson is not a rule until the applicable authority explicitly adopts it, so the framework must not be readable as that authority. The misuse case shows the abuse foreclosed, a lesson used to alter rights, authority, or privacy without approval, this time with the framework's own text offered as the approval. Because the denials are explicit, any retraining, deployment, or record change claiming this document as its warrant is refusable on the clause's plain words, and the separate-authority qualification remains part of the check.

Where this document sits

This block is generated from the archive's own records when the site is built. It records position only and creates no authority.