feat(genome): lived-axis LLM-teacher expansion — the being learns from turns it stalled on - #2033
Merged
Merged
Conversation
…m the turns it stalled on Completes the third efferent organ of the being-loop. The received axis (expansion_examples) teaches deliberately-authored true knowledge directly; the LIVED axis was deferred with a named reason — a salient lived turn is a FAILURE with no known-correct answer, so echoing her stalled answer would train the failure, and with no `test` it can't be remediated either. This builds the path expansion_examples pointed at: a teacher RE-ANSWERS the stimulus she stalled on, and that strong answer becomes the SFT pair. Safe on untestable material for the same reason the received axis is — validated per-CONSOLIDATION by whole-being benchmark lift (#59), never per-trajectory. Structured exactly like remediation (the outlier discipline — one seam, two organs): - teach.rs::synthesize_lived_expansion — the inference core, mirroring synthesize_remediation's lane discipline (readiness gate → dedicated bare-base lane → degrade-loud to the live lane) but with NO grader loop: one teacher generation per stimulus. Two honesty guards: empty answers dropped (never ship a blank lesson), and the teacher's system turn shapes generation but is NOT trained in — the pair is the bare {question → answer}, so the being learns the class of question, not a scaffold. - curriculum.rs::LivedExpansionSynthesizer — the selection wrapper, sibling of RemediationSynthesizer: a deterministic detector-driven select() (unit-testable with no inference — only SALIENT lived failures' stimuli; a clean turn is never re-taught) + an async synthesize() that resolves the teacher and calls the teach.rs core. Returns bare SFT {messages} (like expansion_examples), NOT a RemediationCorpus — the differing type is the honest signal these are two distinct organs, not one behind a fake corpus. Organ + tests only; wiring the lived branch into the auto-firing consolidation driver (alongside the received branch) is a separate reviewable step, like #2028. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LoTjvf5j3Ez13g6k8mRkFo
joelteply
added a commit
that referenced
this pull request
Jul 29, 2026
… that proves AND improves Reframe (Joel): benchmarks are the learning signal, not just the scoreboard. A graded task == a training example; the benchmark engine is the data generator at the head of a continuous-learning loop. Prove (charts) and improve (train on failures) are the SAME pass. Connects the merged benchmark suite + unified experience stream (#2024) + stall-expansion (#2033) to cross-grid distribution, sentinel experiential plasticity, and AttnRes skip paths (the stable 'unet skip for generalization'). 5 gaps: cross-grid matrix, failure->curriculum emit, dream-forge consumes eval-fails, held-out + stochastic-depth generalization, honest-instrument discipline. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Q4NU4VNiELPQfBpCacDZGc
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The third efferent organ of the being-loop
The being-loop unifies lived + eval + received experience into one stream ([[lived-and-eval-experience-are-one-stream-one-being]]). Two of the three axes already teach:
RemediationSynthesizer, test-validated).expansion_examples, deliberately-authored true knowledge taught directly).The Lived axis was deferred with a named reason: a salient lived turn is a FAILURE (non-convergence / infra fault) with no known-correct answer, so echoing her stalled answer would train the failure, and with no
testit can't be remediated. This builds the pathexpansion_examplespointed at.What it does
A teacher re-answers the stimulus she stalled on, and that strong answer — not her stalled one — becomes the
{question → answer}SFT pair. Safe on untestable material for the same reason the received axis is: validated per-consolidation by whole-being benchmark lift (#59), never per-trajectory.Structured exactly like remediation (one seam, two organs):
teach.rs::synthesize_lived_expansion— inference core, mirrorssynthesize_remediation's lane discipline (readiness gate → dedicated bare-base lane → degrade-loud) but no grader loop. Honesty guards: empty answers dropped; the teacher's system turn shapes generation but is not trained in (the pair is the bare question→answer).curriculum.rs::LivedExpansionSynthesizer— selection wrapper, sibling ofRemediationSynthesizer: deterministic detector-drivenselect()(unit-testable, only salient lived failures; a clean turn is never re-taught) + asyncsynthesize(). Returns bare SFT{messages}, not aRemediationCorpus— the differing type is the honest signal these are two distinct organs.Scope
Organ + tests only. Wiring the lived branch into the auto-firing consolidation driver (alongside the received branch) is a separate reviewable step, like #2028.
Tests
cargo test -p continuum-core --lib curriculum— 7/7 green, incl. two new lived-expansion selection tests (salient-lived-only; empty-when-nothing-salient).🤖 Generated with Claude Code