HLX_JOB_BOARD.toml

# HLX Job Board (TOML)
# Master task source for the HLX project.

[[ticket]]
id = "D-238"
title = "Stdlib Prelude — Eliminate extern fn Boilerplate"
status = "complete"
tier = "red"
area = ["language", "stdlib"]
assignee = "codex"
opened = "2026-04-16"
completed = "2026-04-16"
branch = "codex/d-238-239"
# prelude.rs injects core builtin declarations automatically. Merged to main cc93427.

[[ticket]]
id = "D-239"
title = "Script Entry Point — Top-Level Expressions Without fn main()"
status = "complete"
tier = "red"
area = ["language", "runtime"]
assignee = "codex"
opened = "2026-04-16"
completed = "2026-04-16"
branch = "codex/d-238-239"
# __top_level__ acts as full entry point when fn main() is absent. Merged to main cc93427.

[[ticket]]
id = "D-204"
title = "Closures and Captured Environments"
status = "complete"
tier = "red"
area = ["language", "vm", "runtime"]
assignee = "codex"
opened = "2026-04-14"
assigned = "2026-04-19"
completed = "2026-04-19"
branch = "codex/d-204"
# fn [captures](params) { body } — explicit copy-capture, lexical scope isolation, first-class value. 11/11 tests green. Merged to main 0a6abc8.

[[ticket]]
id = "D-187"
title = "RSI Expansion Gate — Terminal Authorization UI"
status = "complete"
tier = "orange"
area = ["runtime", "rsi", "infra"]
assignee = "codex"
opened = "2026-04-03"
completed = "2026-04-21"
branch = "main"
blocked_by = []
# Human-in-the-loop authorization for capability expansion.
# Expansion gate landed across primitive promotion, strategy promotion, Meta-RSI proposals, and batch-end risk ceilings.

[[ticket]]
id = "D-188"
title = "Scalable Architecture — No Hard Dimension Ceilings"
status = "complete"
tier = "purple"
area = ["runtime", "rsi"]
assignee = "codex"
opened = "2026-04-03"
completed = "2026-04-17"
branch = "codex/d-188"
# RuntimeCapacityProfile, LyapunovConfig, RsiProposalCapacityConfig/CriteriaConfig, GpuDispatchConfig. Merged to main 76d75a3.

[[ticket]]
id = "D-235"
title = "Promotion Authority Consolidation"
status = "complete"
tier = "orange"
area = ["rsi", "runtime", "governance"]
assignee = "codex"
opened = "2026-04-16"
completed = "2026-04-16"
branch = "codex/d-235"
# ApeVerdict::CandidateCleared; two-hit threshold before Vocabulary write; scan_and_propose reads Vocabulary with delta watermark. Merged to main fb70c08.

[[ticket]]
id = "D-240"
title = "Module Namespace System v1"
status = "complete"
tier = "red"
area = ["language", "runtime"]
assignee = "codex"
opened = "2026-04-16"
completed = "2026-04-16"
branch = "codex/d-240"
# use stdlib.grid as grid; from stdlib.grid use get, set; dot/:: normalization; cycle detection; prelude conflict handling; legacy preserved. Merged to main fb70c08.

[[ticket]]
id = "D-241"
title = "Standing Housekeeping — Warnings, Test Failures, Compile Noise"
status = "complete"
tier = "yellow"
area = ["runtime", "debt", "infra"]
assignee = "gemini"
opened = "2026-04-16"
assigned = "2026-04-19"
completed = "2026-04-19"
branch = "gemini/d-241-housekeeping"
# 0 test failures, 0 warnings. Real bugs fixed: EmbeddingRepr::Binary lost dimensions,
# I8 to_f32 divide/multiply inversion, Tensor Drop overflow (random crash source), governed_path Value::Map mismatch.
# NOTE: remaining dead_code/unreachable_patterns suppressed via #![allow(...)] attributes — not fixed.
# Follow-up: audit those allow attributes in next housekeeping pass; some may mask real dead weight.
# Merged 9ec600b.

[[ticket]]
id = "D-242"
title = "Structural + Residual Signature Layer"
status = "complete"
tier = "red"
area = ["rsi", "runtime", "score"]
assignee = "codex"
opened = "2026-04-17"
completed = "2026-04-17"
branch = "codex/d-242"
# TaskStructuralSignature, TransformationSignature, ResidualSignature, ResidualTransition. compute_residual() + compute_task_structural_signature(). CausalGraph.residual_transitions capped at 1000. 424 passed, 7 baseline.

[[ticket]]
id = "D-243"
title = "Working Memory — Residual-Facet Salience"
status = "complete"
tier = "orange"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-17"
completed = "2026-04-17"
branch = "codex/d-243"
# WorkingMemory { primitive_salience, facet_salience }. record_primitive_attempt, record_residual_reduction, salience_bias (additive only), top_open_facets, reset. Non-persistent, no Serialize. 432 passed, 7 baseline.

[[ticket]]
id = "D-244"
title = "Analogical Retrieval — Residual/Task Signature Query"
status = "complete"
tier = "orange"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-17"
completed = "2026-04-17"
branch = "codex/d-244"
# AnalogicalRetrieval: structural_similarity (dims/palette/object_count/density/tag Jaccard), retrieve() positive+contrast split, CausalGraph::residual_transitions() read-only accessor. 433 passed, 7 baseline.

[[ticket]]
id = "D-245"
title = "Counterfactual Density Credit (CCA-SA)"
status = "complete"
tier = "red"
area = ["rsi", "runtime", "score"]
assignee = "codex"
opened = "2026-04-17"
completed = "2026-04-17"
branch = "codex/d-245"
# CcaEngine: Q(S,T) approx from vocab stats, leave-one-out direct credit, pairwise synergy len<=4, enablement credit, redundancy flagging, collision debt. CausalCredit + PairCredit. StrategyCandidate gains primitive_ids with name-fallback for legacy candidates. Read-only invariant. 8 tests, 440 passed, 7 baseline.

[[ticket]]
id = "D-246"
title = "Residual Annihilator Search (Abductive Gap 1)"
status = "complete"
tier = "red"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-17"
completed = "2026-04-17"
branch = "codex/d-246"
# ResidualAnnihilator: facet extraction (7 dimensions), read-only causal graph scan, Wilson-damped confidence, top_k/min_support/min_confidence filtering. ExplainsDelta with name lookup + unknown fallback. 7 tests, 471 passed, 7 baseline.

[[ticket]]
id = "D-247"
title = "Meta-Learning Control Signals (Gap 4)"
status = "complete"
tier = "orange"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-17"
completed = "2026-04-17"
branch = "codex/d-247"
# SynthesisQualityRecord: support_mass, best_delta_seen, false_positive_rate, family_reuse_score, stability_impact. 5 control knob helpers. Pure compute, read-only. 8 tests, 480 passed, 7 baseline.

[[ticket]]
id = "D-248"
title = "RCL Live Wiring — Connect RCL Stack to Solve Path"
status = "complete"
tier = "red"
area = ["rsi", "runtime", "eval"]
assignee = "codex"
opened = "2026-04-17"
completed = "2026-04-17"
branch = "codex/d-248-rcl-live"
# RclOrchestrator wired into session_harness run_learning_batch. on_task_result records ResidualTransitions to CausalGraph. on_batch_end runs full ProtoEs→Eval→CCA→SS→SynthesisQuality chain. Merged edbcce1.
# RclOrchestrator in hlx-runtime/src/rcl_orchestrator.rs. Wires WorkingMemory, ResidualAnnihilator, AnalogicalRetrieval, ProtoEsEngine, EvaluationHarness, CcaEngine, SsValidationLayer, SynthesisQualityRecord into session_harness.rs. on_task_result() records residual transitions. on_batch_end() runs full strategy synthesis chain. Makes D-219 through D-247 fire in production.

[[ticket]]
id = "D-217"
title = "RSI Partial-Credit Distance Delta"
status = "complete"
tier = "red"
area = ["rsi", "runtime", "score"]
assignee = "claude"
opened = "2026-04-15"
completed = "2026-04-16"
# 25% ARC ceiling fix: distance_delta feedback topology added to RSI; edit distance + object correspondence scoring.

[[ticket]]
id = "D-218"
title = "Promoted Strategy Demotion and Probation"
status = "complete"
tier = "orange"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-15"
completed = "2026-04-17"
branch = "codex/d-218"
# failure_count + coverage_score decay x0.85; quarantine below 0.05; rollback on success; DistanceDelta feedback hook wired. 426 passed, 7 baseline failures.

[[ticket]]
id = "D-219"
title = "proto-ES Proposal Layer"
status = "complete"
tier = "red"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-15"
completed = "2026-04-17"
branch = "codex/d-219"
# ProtoEsEngine: co_occurrence_threshold=3, candidate_limit=8. propose() read-only vocab access, deduped tag union context_signature, synthetic proto_es provenance, ExpectedDelta::zero(). 426 passed, 7 baseline.

[[ticket]]
id = "D-221"
title = "Strategy Candidate Evaluation Harness"
status = "complete"
tier = "orange"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-15"
completed = "2026-04-17"
branch = "codex/d-221"
# CandidateQualityReport + ScoredCandidate in strategy.rs. EvaluationHarness::evaluate() with read-only vocab lookup, estimated_coverage, coverage_drift, support_mass, composition_cost, false_positive_risk, family_reuse_score. 427 passed, 7 baseline.

[[ticket]]
id = "D-220"
title = "SS Validation and Compilation Layer"
status = "complete"
tier = "red"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-15"
completed = "2026-04-17"
branch = "codex/d-220"
# SsValidationLayer: quality gate → composition gate → G7 gate → PromotedStrategy in Vocabulary. Sequential HLX source generation. fnv1a32 stable name. 435 passed, 7 baseline.

[[ticket]]
id = "D-179"
title = "LLM Corpus Callosum Bridge"
status = "complete"
tier = "purple"
area = ["rsi", "runtime", "candle"]
assignee = "codex"
opened = "2026-03-20"
assigned = "2026-04-19"
completed = "2026-04-19"
branch = "codex/d-179-callosum"
# hlx-callosum crate: S2NPacket/N2SPacket/CallosumAuditRecord, NeuralMode enum.
# builtin_nearest_primitives no longer a stub — CandidateRankingHint path complete.
# HTTP transport avoids crate cycle (hlx-bond binary-only). Whitelist enforcement verified.
# LinguisticSurface partial (no text field in N2SPacket yet — Phase 2). Merged ee03b75.

[[ticket]]
id = "D-205"
title = "RSI Compounding Quality"
status = "complete"
tier = "red"
area = ["rsi", "runtime", "debt"]
assignee = "codex"
opened = "2026-04-14"
closed = "2026-04-20"
# D-205a: MERGED ae9a7ae. quality_gap_scan, GapSeverity, GapSignal, tag_quality.
# D-205b: MERGED bd884c2. eligible_for_composition, composition_purity, builtin_composition_purity.
# D-205c: MERGED (2026-04-20). on_eval_run_end() induction trigger, budget priority, 20-pair cap.
# RSI feedback loop fully closed. ARC delta: training 88/400 (+2), eval 55/400 (flat) after 205a/b.
# ROOT CAUSE of zero RSI delta across all sessions. Three broken feedback loops.
# Spine: "D-205 is not about adding more synthesis — it is about making the pressure field truthful."
# Staged rollout mandatory — measure ARC delta between each phase.
#
# GAP 1 (D-205a): gap_alarm() is binary. A shim-covered tag = fully covered = no synthesis pressure.
#   compute_coverage_score() is per-primitive global breadth, not per-tag quality.
#   Fix: quality_gap_scan() returns GapSignal{tag, severity: Hard/Soft/Covered, pressure: f32}.
#   tag_quality(tag) aggregates over active primitives by audit_class weight.
#   Scheduler ranks by pressure = demand × quality_deficit × shim_penalty × recency.
#
# GAP 2 (D-205b): likely_pattern_shim is classified but never penalized downstream.
#   Shims enter composed_of freely; shim contamination hides inside deeper macros.
#   Fix: eligible_for_composition() gate. Shims hard-blocked from promoted macros,
#   soft-blocked (lower confidence, shim-derived provenance flag) in exploratory synthesis.
#   Composition purity tracked as fraction of reusable/sharp ancestors — propagates through chains.
#
# GAP 3 (D-205c): vocab_induction_run() accumulates co-occurrence but is never triggered.
#   Fix: end-of-eval-run as primary trigger. Optional bounded mini-pass after N promotions.
#   Never continuous. Threshold: co-occurrence count + PMI-style association strength, not raw count.
#   Separate repair budget vs consolidation budget — induction cannot starve hard-gap synthesis.
#
# RUNAWAY RISK: fixing all three opens many soft gaps simultaneously. Mitigate with:
#   - synthesis budget cap per session (hard gaps first, soft gaps second, induction last)
#   - hysteresis: soft gap requires K sessions of low quality, not single-session detection
#   - dedup by tag-cluster (many failure clusters → one synthesis agenda item)
#   - staged rollout: 205a → ARC measure → 205b → ARC measure → 205c → ARC measure

[[ticket]]
id = "H-001"
title = "Warder — Primitive Synthesis Harness"
status = "open"
tier = "blue"
area = ["rsi", "harness"]
assignee = "gemini-flash"
opened = "2026-04-15"
# Primitive RSI synthesis validation (ARC). Albion realm.

[[ticket]]
id = "H-002"
title = "Cosantoir — Strategy Emergence Harness"
status = "blocked"
tier = "orange"
area = ["rsi", "harness"]
assignee = "gemini-flash"
opened = "2026-04-15"
blocked_by = ["D-215", "D-217", "D-219", "D-221", "D-220"]
# Hibernia strategy emergence validation.

[[ticket]]
id = "H-003"
title = "Vakten — Saturation and Stress Harness"
status = "blocked"
tier = "orange"
area = ["rsi", "harness"]
assignee = "gemini-flash"
opened = "2026-04-15"
blocked_by = ["H-002"]
# Midgard saturation classification.

[[ticket]]
id = "D-192"
title = "SWE-bench Evaluation Harness"
status = "dormant"
tier = "blue"
area = ["harness", "score"]
assignee = "unassigned"
opened = "2026-04-03"
# Parallel harness to ARC for general symbolic primitives.

[[ticket]]
id = "D-208"
title = "Open Release Readiness Gate"
status = "open"
tier = "yellow"
area = ["infra"]
assignee = "claude"
opened = "2026-04-14"
# Meta-ticket ensuring all release-blocking items are closed.

[[ticket]]
id = "NC-103"
title = "Rust Backend — nc-bond HLX Integration Stub"
status = "open"
tier = "orange"
area = ["neuroclaw", "runtime"]
assignee = "unassigned"
opened = "2026-04-14"
blocked_by = ["NC-101"]
# Wire rust/crates/nc-bond to hlx-runtime dependency.

[[ticket]]
id = "NC-104"
title = "Base Magos — HLX Session Loop"
status = "dormant"
tier = "red"
area = ["neuroclaw", "rsi"]
assignee = "unassigned"
opened = "2026-04-14"
blocked_by = ["NC-103"]
# Pure symbolic floor evaluation framework.

[[ticket]]
id = "D-249"
title = "Time-Travel Debugging — Reversible DAP (Phase 1: stepBack)"
status = "complete"
tier = "red"
area = ["lsp", "debugger", "runtime"]
assignee = "codex"
opened = "2026-04-17"
completed = "2026-04-21"
branch = "main"
# HLX's A2 (Reversibility) + A3 (Bijection) axioms make step-back debugging architecturally natural.
# HLX does not require a separate semantic trace reconstruction layer. In debug mode, snapshotting
# canonical VM state before each step is sufficient because the VM state is authoritative and
# bijectively restorable. (This is snapshot recording policy, not event-log reconstruction.)
#
# Term separation (important):
#   - Step-back debugging: restore previous snapshot for inspection/navigation  [Phase 1]
#   - Reverse continue: step backward repeatedly until breakpoint/origin/condition  [Phase 1]
#   - Timeline pruning: semantically discard a branch from future history  [Phase 3 only]
#   Do NOT conflate step-back with timeline pruning. They are qualitatively different.
#
# Phase 1 scope (this ticket):
#   1. VmSnapshot type in debugger.rs. Must capture ALL state for observational equivalence:
#      - instruction pointer, frame pointer, call stack
#      - operand stack, locals, registers
#      - heap / object graph / handles (all runtime-visible)
#      - any pending control-flow / branch context
#      - task-scoped runtime metadata that influences future execution
#      - RNG state if any exists in debug mode
#      - VM-owned caches that could affect behavior if restored inconsistently
#      Invariant: restoring VmSnapshot must make the next forward step observationally
#      identical to what it would have been at the original timeline point.
#   2. snapshot_ring: VecDeque<VmSnapshot> in Debugger, fixed capacity (~1000 steps).
#      Define snapshot granularity as a SnapshotPolicy enum even if only one variant is
#      implemented now. Do NOT bake "every instruction always" into the API shape.
#      Policy variants (implement EachInstruction only; define others for future):
#        EachInstruction | EachSourceLine | EachBreakpoint | EveryN(u32)
#   3. StepMode::StepBack variant. Debugger::step_back() pops ring, restores VM state.
#   4. DapServer: handle "stepBack" and "reverseContinue" DAP protocol requests.
#      - stepBack: works when snapshot history exists; returns proper "not available" when exhausted.
#      - reverseContinue: repeatedly step back until breakpoint hit or history exhausted.
#   5. Wire into VM execution loop — snapshot before each step when debugger is active.
#   6. Use fully materialized in-memory snapshots only. No content-addressed dedup yet.
#
# Acceptance criteria (proof obligations — these are where A2/A3 claims become real):
#   A. Round-trip equivalence: snapshot → restore → snapshot again yields identical debugger-observable state.
#   B. Forward replay equivalence: stepping forward from restored snapshot reproduces same next instruction/result.
#   C. Stack/locals fidelity: call stack, locals, operand stack, and instruction pointer all restore exactly.
#   D. Ring eviction: when capacity exceeded, oldest snapshots drop cleanly without corrupting later stepBack.
#   E. DAP contract: stepBack/reverseContinue behave correctly at history boundary (not just happy path).
#
# Phase 2 (future): Timeline visualization panel — navigable execution graph in VS Code.
# Phase 3 (future): Branch pruning / what-if mode. This is Timeline Pruning — qualitatively
#   different from step-back. Keep future-scoped, do not conflate with Phase 1.
#
# Why this matters: no other LSP offers this because no other language has A1+A2+A3 at the
# bytecode layer. DAP already has the verbs — other adapters return "not supported."

[[ticket]]
id = "D-259"
title = "Wire AnalogicalRetrieval into RclOrchestrator on_batch_end"
status = "done"
tier = "orange"
area = ["rsi", "runtime", "rcl"]
assignee = "codex"
opened = "2026-04-22"
completed = "2026-04-23"
branch = "main"
depends_on = ["D-258"]
# H-002 harness confirmed retrieval_wired=false: AnalogicalRetrieval works standalone
# (retrieved 4 candidates from causal graph) but is not called in RclOrchestrator::on_batch_end.
# Wire it between residual transition recording and ProtoEsEngine::propose so structural
# similarity biases candidate selection.

[[ticket]]
id = "D-258"
title = "Restore per-task crash isolation in run_learning_eval_internal"
status = "in_progress"
tier = "orange"
area = ["eval", "runtime", "infra"]
assignee = "codex"
opened = "2026-04-22"
branch = "main"
# D-252 cold-start probe broadened solve path, exposing latent OOB bugs in try_separator_reflection
# (fixed 3806708) and try_latin_square. run_learning_eval_internal removed sub-VM isolation assuming
# only 1 task crashed. Restore via new call_safe Rust builtin — wraps symbolic_solve_constrained
# per task, absorbs RuntimeError → nil, no recompile overhead. Unblocks RSI learning score.

[[ticket]]
id = "D-250"
title = "RSI co-occurrence recording — drop promotion gate, record all pairs"
status = "complete"
tier = "green"
area = ["rsi", "vocab"]
assignee = "codex"
opened = "2026-04-20"
completed = "2026-04-20"
branch = "main"
# rsi_record_success now records all op pairs whenever len(ops) >= 2, independent of macro promotion.

[[ticket]]
id = "D-251"
title = "Vocabulary name prefix match — unblock co-occurrence recording"
status = "complete"
tier = "green"
area = ["rsi", "vocab"]
assignee = "codex"
opened = "2026-04-20"
completed = "2026-04-20"
closed = "2026-04-20"
branch = "main"
# builtin_vocab_record_co_occurrence now resolves parameterized op strings by vocab-name prefix with a word boundary.

[[ticket]]
id = "D-252"
title = "RSI cold-start fix — M2 probe after M1 + threshold → 1"
status = "complete"
tier = "green"
area = ["rsi", "vocab", "solve"]
assignee = "codex"
opened = "2026-04-20"
completed = "2026-04-20"
branch = "main"
# After M1 short-circuit in symbolic_solve_constrained, probe all 6 transforms × color_map combos.
# Verified combos record co-occurrence; near-misses call record_attempt_from_eval.
# INDUCTION_THRESHOLD_COUNT lowered to 1 (set via D-205c, confirmed present).
# Cold-start probe: solve.hlx lines 3599-3618.

[[ticket]]
id = "D-253"
title = "Attempt graph — AttemptRecord + attempt_log + record_attempt builtin"
status = "complete"
tier = "green"
area = ["rsi", "vocab", "solve"]
assignee = "codex"
opened = "2026-04-20"
completed = "2026-04-20"
branch = "main"
# AttemptRecord{session,task_id,profile_key,op_a,op_b,closeness,gap_desc} in rsi_engine.rs.
# attempt_log SQLite table in session_vocab.db. load/save_attempt_store() lifecycle.
# builtin_record_attempt() + record_attempt_from_eval() in solve.hlx.
# set_attempt_context/clear_attempt_context for session/task threading.
# Current data: identity-composing pairs (rotate1+rotate3, transpose+transpose), closeness 0.25.

[[ticket]]
id = "D-254"
title = "Attempt-graph composition synthesis — LLM-free template promotion"
status = "complete"
tier = "orange"
area = ["rsi", "runtime"]
assignee = "codex"
opened = "2026-04-20"
completed = "2026-04-21"
branch = "main"
# Attempt-derived synthesis is live: session 73 reported attempt_synthesis considered=4, inducted=4, promotions=4.
# Landed tiered gate (0.60/count>=1, 0.40/count>=2), parameterized-op canonicalization, and attempt_derived two-hit bypass.
# Four attempt_composed_* primitives now persist in vocabulary.

[[ticket]]
id = "D-255"
title = "Ground-Truth Bootstrap Pass — pre-seed co-occurrence + attempt_log from training data"
status = "complete"
tier = "orange"
area = ["rsi", "runtime", "eval"]
assignee = "codex"
opened = "2026-04-20"
depends_on = ["D-253", "D-254"]
completed = "2026-04-21"
branch = "main"
# --bootstrap-training --force-bootstrap is working and idempotent via bootstrap_metadata persistence.
# After the tiered-gate fix, forced bootstrap reported distinct_pairs_above_synthesis_gate=4.
# Bootstrap now pre-seeds both co-occurrence and attempt_log state before the first eval session.

[[ticket]]
id = "D-256"
title = "Callosum MCP Adapter — expose HLX distillation tools to agent MCP servers"
status = "complete"
tier = "yellow"
area = ["callosum", "mcp", "distillation"]
assignee = "codex"
opened = "2026-04-21"
completed = "2026-04-21"
branch = "main"
# Thin adapter layer over hlx-callosum query surface. Expose semantic_context, gap recording,
# semantic vocab lookup, and distillation-observation tools to agent MCP servers.
# Implemented 4 synchronous adapter tools plus the hlx-mcp JSON-RPC stdin/stdout server binary.
# MCP_CONFIG_SNIPPET.json provides the manual agent-config stanza without editing agent configs directly.

[[ticket]]
id = "D-257"
title = "LinguisticSurface Phase 2"
status = "complete"
tier = "orange"
area = ["callosum", "taxonomy", "distillation"]
assignee = "codex"
opened = "2026-04-21"
depends_on = ["D-256"]
completed = "2026-04-21"
branch = "main"
# D-257a: closed transformation taxonomy types in hlx-callosum/src/taxonomy.rs.
# D-257b: packet/schema expansion for semantic task context.
# D-257c: downstream wiring into prompts and solve-path consumers.
# Live OpenAI smoke remains environment-blocked in this shell (`401 invalid_api_key`);
# merge gate accepted the graceful no-key path and code-level wiring instead.

[[ticket]]
id = "D-257d"
title = "Universal endpoint config — callosum.toml with priority-ordered provider list"
status = "complete"
tier = "orange"
area = ["callosum", "config", "distillation"]
assignee = "codex"
opened = "2026-04-21"
completed = "2026-04-21"
branch = "main"
# Replaced hardcoded provider selection with ~/.hlx/callosum.toml + compiled defaults.
# Supports priority-ordered OpenAI-wire and Anthropic endpoints without further code changes.

# --- CHANGELOG (Historical) ---
# D-250 and D-251 are tracked as [[ticket]] entries above; not duplicated here.

[[changelog]]
id = "D-222"
title = "JSON Builtin Naming Inconsistency"
status = "complete"
tier = "green"
closed = "2026-04-15"
fix = "Standardized on json_parse/encode. Added VM aliases."

[[changelog]]
id = "D-223"
title = "Numeric Type Comparison Crashes (i64 vs f64)"
status = "complete"
tier = "yellow"
closed = "2026-04-15"
fix = "Implemented safe numeric coercion in comparisons."

[[changelog]]
id = "D-224"
title = "Map/Object Builtin Naming Inconsistency"
status = "complete"
tier = "green"
closed = "2026-04-15"
fix = "Normalized to map_ prefix and consolidated builtins."

[[changelog]]
id = "D-225"
title = "Sandboxed File Operations Silent Failure"
status = "complete"
tier = "orange"
closed = "2026-04-15"
fix = "Returned explicit error values from APE gates to HLX."

[[changelog]]
id = "D-209"
title = "Checkpoint / Restore — Serialization Foundation"
status = "complete"
tier = "yellow"
closed = "2026-04-14"
fix = "Added Serialize/Deserialize to Vocabulary and CausalGraph."

[[changelog]]
id = "D-213"
title = "Checkpoint / Restore — CLI Flag + RSIPipeline"
status = "complete"
tier = "yellow"
closed = "2026-04-16"
fix = "Implemented full RSIPipeline serialization and --resume flag."

[[changelog]]
id = "D-231"
title = "D-213 Checkpoint Governance — Restore Validation API"
status = "complete"
tier = "yellow"
closed = "2026-04-16"
fix = "Added risk/status validation to checkpoint loading."

[[changelog]]
id = "D-232"
title = "Vocab Induction Schema Mismatch — composition_kind Missing"
status = "complete"
tier = "yellow"
closed = "2026-04-16"
fix = "Updated rsi_engine to emit composition_kind: seq."

[[changelog]]
id = "D-233"
title = "wrong_cells_frac / fraction_correct Naming Inversion"
status = "complete"
tier = "blue"
closed = "2026-04-16"
fix = "Renamed metric to fraction_correct throughout codebase."

[[changelog]]
id = "D-234"
title = "Search Portfolio — Wire or Quarantine"
status = "complete"
tier = "orange"
closed = "2026-04-16"
fix = "Wired adaptive portfolio ranking into search pipeline."

[[changelog]]
id = "D-236"
title = "Causal Graph Persistence — Save on Success"
status = "complete"
tier = "blue"
closed = "2026-04-16"
fix = "Added save_causal_graph calls on successful records."

[[changelog]]
id = "D-237"
title = "Predicate Genesis — Single Counter Authority"
status = "complete"
tier = "yellow"
closed = "2026-04-16"
fix = "Consolidated recurrence tracking into Rust runtime."

[[changelog]]
id = "D-230"
title = "RSI Signal-Path Audit — Metrics to Decision Wiring"
status = "complete"
tier = "orange"
closed = "2026-04-16"
fix = "Verified causal paths for all computed metrics."

[[changelog]]
id = "D-215"
title = "Strategy Candidate Object Type"
status = "complete"
tier = "orange"
closed = "2026-04-16"
fix = "Added typed StrategyCandidate with context signatures."

[[changelog]]
id = "D-216"
title = "APE G7 — Strategy Compression Predicate"
status = "complete"
tier = "orange"
closed = "2026-04-16"
fix = "Implemented G7 formal density inversion check."

[[changelog]]
id = "D-210"
title = "Stdlib Type Name Inconsistency"
status = "complete"
tier = "green"
closed = "2026-04-14"
fix = "Normalized type names to lowercase (string, list)."

[[changelog]]
id = "D-211"
title = "Stdlib Namespace Collision — kind Function"
status = "complete"
tier = "green"
closed = "2026-04-14"
fix = "Renamed colliding functions with unique prefixes."

[[changelog]]
id = "D-212"
title = "Dead Code Audit — gc_agent.hlx, governance.hlx"
status = "complete"
tier = "grey"
closed = "2026-04-14"
fix = "Cleaned up stubs with no exports."

[[changelog]]
id = "D-226"
title = "rsi_gate.rs — AST Visitor Consolidation"
status = "complete"
tier = "blue"
closed = "2026-04-16"
fix = "Consolidated structural recursion into RsiVisitor."

[[changelog]]
id = "D-227"
title = "z_meta.rs — Normalize L2 Helper"
status = "complete"
tier = "green"
closed = "2026-04-16"
fix = "Extracted shared normalize_l2 helper function."

[[changelog]]
id = "D-228"
title = "rsi_engine.rs — BFS Component Count Dedup"
status = "complete"
tier = "green"
closed = "2026-04-16"
fix = "Extracted shared flood_fill_component BFS helper."

[[changelog]]
id = "D-229"
title = "hil/solve.hlx — make_result Delegation"
status = "complete"
tier = "green"
closed = "2026-04-16"
fix = "Delegated make_result to make_result_full with defaults."

[[changelog]]
id = "D-163"
title = "Candle Q1_0 Binary Upstream Prep"
status = "complete"
tier = "orange"
closed = "2026-04-15"
fix = "Implemented BlockQ1_0 and dequantization logic."

[[changelog]]
id = "NC-101"
title = "NeuroClaw Rebrand Pass"
status = "complete"
tier = "green"
closed = "2026-04-14"
fix = "Global rebrand of openclaw to neuroclaw."

[[changelog]]
id = "NC-102"
title = "APE Channel Governance — Surface Audit"
status = "complete"
tier = "blue"
closed = "2026-04-14"
fix = "Audited APE insertion points in the frontend."

[[changelog]]
id = "RoC-001"
title = "Arena Agent Testbed — Headless realms of camelot"
status = "complete"
tier = "orange"
closed = "2026-04-15"
fix = "Built server-side scripting harness and verified loop."

[[changelog]]
id = "T-000"
title = "Autoping Dynamic Pane Lookup"
status = "complete"
tier = "green"
closed = "2026-04-16"
fix = "Resolved panes at inject-time via regex heuristics."

[[changelog]]
id = "T-001"
title = "Message Lifecycle State"
status = "complete"
tier = "green"
closed = "2026-04-16"
fix = "Added status/ticket_id columns and tether_close tool."

[[changelog]]
id = "T-002"
title = "TOML Job Board + Tether Bridge"
status = "complete"
tier = "yellow"
closed = "2026-04-16"
fix = "Converted board to TOML and launched watchdog bridge."

[[changelog]]
id = "D-175"
title = "Distribution and Install Story"
status = "complete"
tier = "blue"
closed = "2026-04-16"
fix = "Wired cargo install hlx and automated releases."

[[changelog]]
id = "D-206"
title = "Production Validation Matrix"
status = "complete"
tier = "yellow"
closed = "2026-04-14"
fix = "Codified trust suite into repeatable CI paths."

[[changelog]]
id = "D-207"
title = "Documentation and Discoverability"
status = "complete"
tier = "blue"
closed = "2026-04-14"
fix = "Created pitch, stdlib ref, and walkthrough."

[[changelog]]
id = "D-214"
title = "Scrub corpus.db from git history"
status = "complete"
tier = "yellow"
closed = "2026-04-14"
fix = "Scrubbed sensitive files using git filter-repo."

[[changelog]]
id = "D-200-followup"
title = "Tool Surface HARDENING"
status = "obsolete"
tier = "grey"
closed = "2026-04-14"
fix = "Functionality covered by specific tickets."

[[changelog]]
id = "D-254"
title = "Attempt-graph composition synthesis — LLM-free template promotion"
status = "complete"
tier = "orange"
closed = "2026-04-21"
fix = "Shipped tiered gating, canonicalized attempt pairs, and attempt_derived first-hit promotion; session 73 promoted four attempt_composed_* primitives."

[[changelog]]
id = "D-255"
title = "Ground-Truth Bootstrap Pass — pre-seed co-occurrence + attempt_log from training data"
status = "complete"
tier = "orange"
closed = "2026-04-21"
fix = "Added bootstrap-training pass with persistent bootstrap_metadata; forced bootstrap now yields four attempt-synthesis candidates before session 1."

[[changelog]]
id = "D-257"
title = "LinguisticSurface Phase 2"
status = "complete"
tier = "orange"
closed = "2026-04-21"
fix = "Shipped closed semantic taxonomy, Phase 2 callosum packets, and solve-path semantic bias wiring; live OpenAI smoke stayed blocked by a stale API key, but no-key fallback and Rust/HLX integration gates passed."

[[changelog]]
id = "D-257d"
title = "Universal endpoint config — callosum.toml with priority-ordered provider list"
status = "complete"
tier = "orange"
closed = "2026-04-21"
fix = "Moved semantic endpoint selection into ~/.hlx/callosum.toml with compiled DeepSeek/Grok/OpenAI/Anthropic defaults and priority-ordered fallback when the config file is absent."

[[changelog]]
id = "D-256"
title = "Callosum MCP Adapter — expose HLX distillation tools to agent MCP servers"
status = "complete"
tier = "yellow"
closed = "2026-04-21"
fix = "Implemented the 4 callosum MCP adapter tools, added the hlx-mcp JSON-RPC binary in hlx-eval, and wrote the manual MCP config snippet for external agents."

[[ticket]]
id = "D-258"
title = "Restore per-task crash isolation — CLOSE"
status = "complete"
tier = "orange"
area = ["eval", "runtime", "infra"]
assignee = "codex"
opened = "2026-04-22"
completed = "2026-04-23"
branch = "main"
# call_safe Rust builtin shipped. 8s sub-VM timeout matches hlx-eval. 89/400 confirmed by Gemini re-run.
# Duplicate entry above is the in-progress marker; this one closes it.

[[ticket]]
id = "D-260"
title = "Wire z_meta_query into ARC solve pipeline — strategy hint at solver entry"
status = "open"
tier = "orange"
area = ["rsi", "runtime", "arc", "meta"]
assignee = "codex"
opened = "2026-04-23"
depends_on = ["D-259"]
# z_meta.rs is live and functional for Bitsy's reason() loop but is NOT wired to symbolic_solve_constrained.
# Before the solver tries any strategy, it should call z_meta_query(z_brain, z_meta_agg, z_meta_bank)
# to get a StrategyHint and reorder / bias the strategy attempt sequence accordingly.
# This turns the static ordered pipeline into a meta-strategy-guided one that improves with every session.
# Wire point: symbolic_solve_constrained entry, after scene graph build, before first solver attempt.
# z_brain state should be derived from task input (embed or hash-project the test grid).
# z_meta_update must be called on solve success/failure to close the feedback loop.

[[ticket]]
id = "D-261"
title = "HLX SDK — standalone distributable language toolchain"
status = "in_progress"
tier = "orange"
area = ["infra", "distribution", "ffi"]
assignee = "codex"
opened = "2026-04-23"
# Extract clean SDK from monorepo: hlx-runtime (research modules stripped), hlx-run, hlx-lsp,
# hlx-lint, hlx-ffi (C ABI for Unity/Python/C# embedding), ape.
# New repo: /mnt/d/Language Projects/hlx-sdk/ → github.com/latentcollapse/hlx-sdk
# Full spec: HLX_SDK_SPEC.md in workspace root.
# cargo build (no features) must pass clean. No ARC/RSI builtins, no model weights, no eval harness.

[[ticket]]
id = "D-262"
title = "Strategy induction — promote inducted primitives into StrategyPortfolio"
status = "open"
tier = "orange"
area = ["rsi", "runtime", "search"]
assignee = "codex"
opened = "2026-04-23"
depends_on = ["D-260"]
# RSI currently inducts primitives into Vocabulary but they never reach StrategyPortfolio in search.rs.
# StrategyPortfolio is used by BidirectionalBeam/MCTS for strategy selection during solve.
# Pipeline to wire: VocabEntry (inducted, win_rate > threshold) → PromotionGate (Sprout+) →
# StrategyPortfolio::register() so MCTS can select and compose inducted strategies dynamically.
# Also wire strategy_seq_apply compositions from strategy.hlx into the portfolio.
# This is the primitive→strategy→meta-strategy compounding chain.

[[ticket]]
id = "D-263"
title = "Bootstrap broadening — diversify seed primitives beyond identity compositions"
status = "done"
tier = "green"
area = ["rsi", "vocab"]
assignee = "codex"
opened = "2026-04-23"
completed = "2026-04-23"
branch = "main"
depends_on = ["D-259"]
# Current bootstrap (D-255) seeds 4 primitives: rotate1+rotate3, transpose+transpose.
# These are identity compositions — they teach the system nothing about the task distribution.
# Replace with a broader seed set covering the main ARC operation families:
#   - Color mapping compositions (recolor + mask)
#   - Geometric transforms (reflect + crop, rotate + shift)
#   - Pattern fill sequences (flood_fill + border)
#   - Object isolation + placement
# Seed set should be derived from the top co-occurrence pairs observed across all 400 training tasks
# (query arc_session.db for pairs with count > 5 and positive score delta).
# Target: 20-30 diverse seed primitives spanning at least 8 distinct ARC task families.

[[ticket]]
id = "D-264"
title = "RSI gate pass rate diagnostic — measure G1-G4+G6 rejection breakdown"
status = "done"
tier = "green"
area = ["rsi", "governance", "diagnostics"]
assignee = "codex"
opened = "2026-04-23"
completed = "2026-04-23"
branch = "main"
depends_on = ["D-259"]
# We have never measured how many synthesized candidates clear each APE gate (G1-G4+G6).
# If pass rate is near zero, synthesis is structurally blocked regardless of candidate quality.
# Add per-gate rejection counters to RsiGate::evaluate(). After each eval run, emit a summary:
#   G1 (novelty): N passed / M rejected
#   G2 (safety): N passed / M rejected
#   G3 (consistency): N passed / M rejected
#   G4 (coverage): N passed / M rejected
#   G6 (human gate): N passed / M rejected
# Log to arc_session.db and surface in hlx-eval batch summary.
# This is a diagnostic ticket — no behavior change, just observability.

[[ticket]]
id = "D-265"
title = "Phase 2 dry-run mode — exercise RSI gate without plugin mutation"
status = "done"
tier = "green"
area = ["rsi", "diagnostics", "eval"]
assignee = "codex"
opened = "2026-04-23"
completed = "2026-04-23"
branch = "main"
depends_on = ["D-264"]
# Full run_learning_eval() currently leaves phase 2 disabled because rsi_promote() rewrites plugin.hlx.
# Add a dry-run path that still executes hypothesize -> synthesize -> rsi_gate_check so D-264
# counters can report real gate rejection structure without mutating the shared tree.

[[ticket]]
id = "D-266"
title = "RSI synthesis prompt hardening — inject HLX parser constraints and one valid example"
status = "in_progress"
tier = "green"
area = ["rsi", "prompting", "eval"]
assignee = "codex"
opened = "2026-04-23"
branch = "main"
depends_on = ["D-265"]
# D-265 showed the gate is not the bottleneck: 5/5 phase-2 dry-run rejections were parse errors.
# Harden the local rsi_synthesize prompt with explicit HLX parser constraints and one minimal valid
# example before changing the oracle so we can measure whether syntax failures drop on the same eval slice.

[[ticket]]
id = "D-267"
title = "Bond stub footgun — warn or error when HLX_BOND_ENDPOINT is unset in non-test runs"
status = "open"
tier = "green"
area = ["runtime", "bond", "diagnostics"]
assignee = "codex"
opened = "2026-04-23"
branch = "main"
depends_on = ["D-265"]
# hlx-run currently falls back to a fake `[Bond LLM response to: ...]` string when no bond endpoint
# is configured. That silently invalidates RSI synthesis/eval experiments by making stub text look like
# model output. Follow-up: warn loudly or hard-fail in non-test mode when HLX_BOND_ENDPOINT is absent.