Skip to content

katopz/katgpt-rs / .agents/skills/substrate-first

substrate-first

Verified

Pre-implementation DRY gate + existing-code drift audit for the multi-repo workspace. Use BEFORE writing any new System impl, trait, perception/cognition/emotion pipeline, state management, spatial query, or vocabulary type — to verify you're consuming existing substrate, not duplicating it. Also use to AUDIT existing code for parallel-system DRY violations (code that re-implements substrate under different names). The canonical defense is vocabulary translation; concepts ship under operator names (`GenericSpatialBelief`, `decay_confidence`), not English names ("threat field", "spatial hash") — a single-vocabulary grep returns ZERO hits even when the substrate fully exists. Sibling to boundary-guard + goat-audit + feature-gate-audit + doc-sync.

Substrate-First — DRY gate + drift audit

The workspace — 18 repos carrying a root BOUNDARY.md as of 2026-09-01, derived by the snippet in Step 2 rather than typed here — has a recurring failure mode: an agent receives a task ("add threat perception"), jumps to implementation without checking existing substrate, and builds a parallel system that duplicates functionality already shipped under a different name. The user then has to catch it manually.

This skill prevents that. It runs in two modes:

  1. Pre-implementation gate — run BEFORE writing code
  2. Existing-code audit — scan for already-shipped DRY violations

Canonical failures (the pattern this skill prevents)

Failure 1 — ThreatField (Issue 047, riir-mmorpg-examples, 2026-08-01)

Built: ThreatField — a spatial hash grid (HashMap<(i32,i32), u32>) for threat perception. Deposited monster positions into cells; NPCs sampled the 3×3 neighborhood.

Already existed: GenericSpatialBelief<T> + target_within_visible_radius()

  • decay_confidence() — the full fog-of-war → belief → decay pipeline in riir-games-shared/src/game_traits/spatial.rs.

Why the grep missed it: The agent searched for "threat field" / "spatial hash". The substrate ships as GenericSpatialBelief / SpatialBelief / confidence_decay. A single-vocabulary grep returns ZERO hits even when the substrate fully exists.

Resolution: Reverted. The plain scan (tick_swarm_emotions) is a simpler POC-scale simplification. The belief-based system is deferred until fog-of-war becomes a gameplay feature.

Failure 2 — Orchard + Motivation in SDK src/ (Issue 490 + Issue 493)

Built: NpcReasonSystem, AppleGrowSystem, OrchardGoal, EmotionField, EmotionAxis, tick_feeling_brain directly in the SDK facade's src/.

Already existed: The boundary rule (riir-game-sdk/AGENTS.md) says "no game logic in src/" + Proposal 019 excludes emotion from the SDK. The substrate belongs in riir-games (in riir-ai).

Why the grep missed it: The agent didn't check whether the types violated the domain classification rule (latent semantic emotion ≠ raw physical vocabulary).

Resolution: Extracted to riir-games (orchard) and riir-games::motivation (emotion). The SDK re-exports them.


Mode 1: Pre-implementation gate (BEFORE writing code)

Run this checklist before implementing ANY of these:

  • New impl System or tick function
  • New trait + impl (perception, cognition, emotion, state management)
  • New spatial query / index / hash / grid
  • New vocabulary type / DTO / config struct
  • New "helper" function that does math (distance, sigmoid, projection)
  • New pipeline (perception → emotion → behavior, freeze → sync → thaw)

Step 1 — Vocabulary-translate your search

The concept you're building probably already exists under a different name. Before grepping, write down 3+ name variants for the concept:

You're building...Also search for...Likely substrate names
"threat field" / "spatial hash"belief, perception, spatial cognition, fog-of-war, visibilityGenericSpatialBelief, SpatialBelief, confidence_decay, target_within_visible_radius
"emotion" / "feeling" / "mood"affect, drive, motivation, fear, desireEmotionField, EmotionAxis, AffectField, DriveSpecSet, tick_feeling_brain
"state sync" / "delta" / "snapshot"replication, gossip, cache, commitmentSyncBlock, ZoneDelta, PlayerStateCache, GossipDelta, SyncRegistry
"position" / "movement" / "physics"spatial, coordinate, force, velocityMapPos3D, ForceVector, SpatialIndex, GridSpatialIndex
"save" / "persist" / "freeze"thaw, snapshot, serialize, storefreeze_avatar_delta, LocalKvStore, ShardIndex, NeuronShard
"validate" / "anti-cheat" / "check"verify, guard, proofAvatarAntiCheatValidator, AdaptiveModConfig, make_validator_predicate
"tick" / "update" / "loop"system, schedule, game coreSystem, TickCtx, World, GameCore, FrameSnapshot
"knowledge" / "relationship" / "graph"triple, semantic, KGKgTriple, KgTripleTemplate, DualSignalEvidence
"attack" / "damage" / "combat"hp, health, dex, fightHp, Dex, CombatConfig, combat_tick, attack_interval_ticks
"npc" / "swarm" / "crowd"agent, bot, foragerSwarmState, ForagerSwarmSystem, ForagerAi, BotThought
"decay" / "fade" / "forget"sigmoid, heal, baseline, confidencedecay_confidence, tick_feeling_brain, sigmoid, EmotionBaseline
"embedding" / "vector" / "latent"direction, projection, HLA, shardNeuronShard, HlaCacheProxy, compute_animal_emotions

This is the same technique as the paper→code vocabulary translation in AGENTS.md §"Manifold Geometry". The R296 canonical failure applies internally too: a concept-name grep returns zero hits because the math ships under operator names.

Step 2 — Grep the CODEBASE (not just docs)

# Grep ALL repos for the substrate names from Step 1.
# Use multiple variants — the concept may exist under any of them.
#
# DERIVE the repo set; never type it (fixed 2026-09-01 — the hard-coded
# 7-repo brace list this replaced could not see 13 of the 20 contract repos,
# INCLUDING two product-set ones: riir-armageddon consumes
# `GenericSpatialBelief` in 2 files that the canonical DRY grep was structurally
# unable to find, and riir-dapps was equally invisible. A gate that cannot see a
# repo cannot tell you whether it consumes substrate or duplicates it.)
cd /Users/katopz/git
grep -rn 'GenericSpatialBelief\|SpatialBelief\|confidence_decay' \
    --include='*.rs' --exclude-dir=target --exclude-dir=.git \
    $(ls -d */ | while read -r d; do
        [ -f "$d/BOUNDARY.md" ] && [ -d "$d/.git" ] && printf '%s ' "$d"
      done)

Three deliberate details, each measured on this pattern (2026-09-01):

detailwhy
-d "$d/.git", not -ea git worktree has a .git file. ai-perfwt (a riir-ai worktree) otherwise contributes 43 duplicate hits — a DRY gate reporting one implementation twice is the failure it exists to prevent
--exclude-dir=targetthe old form had none, and spent nearly all of its 23 s inside build dirs
unquoted $( … ), not $VARzsh does not word-split an unquoted parameter expansion (grep would get one giant argument and fail) but it does split an unquoted command substitution. Verified in this shell

Net: 227 hits in 1.4 s vs the old form's 221 in 23 s — strictly wider coverage, no worktree duplicates, 16× faster.

# Also grep .research/ and .proposals/ for design rules that apply:
grep -rn 'two-brain\|fog-of-war\|domain classification\|sync boundary' \
    /Users/katopz/git/*/.{research,proposals,docs}/ 2>/dev/null

If you find existing substrate → STOP. Consume it. Do not build a parallel system. Document why you're consuming it (in the plan/issue).

Step 3 — Check AGENTS.md architectural rules

Before building, verify your design doesn't violate these:

RuleSourceWhat it means
Domain classificationAGENTS.md §"Latent vs Raw Space Rules"Physical = raw exact; Semantic = latent dot-product + sigmoid; Social = KG triples
Two-brain modelAGENTS.md §"Spatial Cognition"Info brain (synced ground truth) ≠ think brain (per-NPC beliefs, fog-of-war gated)
Sync boundaryAGENTS.md §"Sync Boundary Rule"Through SyncBlock → quorum → Cold = raw + deterministic; Local = latent
Bridge patternAGENTS.md §"Bridge Pattern"raw → latent = dot+sigmoid; latent → raw = clamp; zero-alloc, gateable
KG triple emissionAGENTS.md §"KG Triple Emission"Semantic encounters → KG triple; Physical events → TxDelta with raw values
Facade constraintriir-game-sdk/AGENTS.mdSDK = re-export facade, no engine deps; vocabulary in riir-games-shared
Boundary ruleriir-game-sdk/AGENTS.mdNo game logic in consumer src/; game systems in riir-games

If your design violates any of these → STOP. File an issue. Rethink.

Step 4 — Decide: consume vs. build

SituationAction
Substrate EXISTS and fitsConsume it. Wire via trait/config. Zero new substrate code.
Substrate EXISTS but wrong shapeExtend the substrate (in the right repo). File a plan.
Substrate DOESN'T existFile an issue in the right repo FIRST. Then build.
You're not sureSTOP and file an issue. Don't guess.

Never build new substrate inside a consumer. The consumer provides data + wiring only. If you're writing loops/math/constants in a consumer's src/ → you're building substrate in the wrong place.

Step 5 — Record the decision

In the plan/issue, document:

## Substrate check (substrate-first skill)
- Searched for: [concept names + variants]
- Found: [existing substrate or "none"]
- Decision: [consume / extend / build new]
- Architectural rules checked: [list which rules apply + verdict]

Mode 2: Existing-code audit (scan for drift)

Run this when reviewing code, when you suspect a parallel system, or quarterly as a DRY-hygiene gate (alongside boundary-guard).

Audit Step 1 — Inventory substrate primitives

For each domain, identify what substrate exists:

# Perception / spatial cognition
grep -rn 'GenericSpatialBelief\|SpatialBelief\|target_within_visible_radius' \
    --include='*.rs' /Users/katopz/git/*/  | grep -v '/tests/' | grep -v '/target/'

# Emotion / affect
grep -rn 'EmotionField\|EmotionAxis\|tick_feeling_brain\|AffectField\|DriveSpecSet' \
    --include='*.rs' /Users/katopz/git/*/

# State sync
grep -rn 'SyncBlock\|ZoneDelta\|PlayerStateCache\|GossipDelta\|SyncRegistry' \
    --include='*.rs' /Users/katopz/git/*/

# Spatial
grep -rn 'SpatialIndex\|GridSpatialIndex\|OctreeSpatialIndex\|MapPos3D' \
    --include='*.rs' /Users/katopz/git/*/

Audit Step 2 — Grep for parallel systems

For each substrate primitive found in Step 1, grep consumer code for reimplemented versions.

Derive the consumer set — this block hard-coded {riir-mmorpg-examples,riir-game-sdk} until 2026-09-01 and so could see 2 of the 20 contract repos. That is the same defect the vocabulary step above was fixed for, in the same file, one section down: 60655c48 corrected the Step 2 named "Step 2" and left this one, which is the step that actually looks for duplicate implementations. riir-armageddon — a product-set repo that consumes GenericSpatialBelief in 2 files (6 sites) — was invisible to every grep below.

cd /Users/katopz/git
consumers() {
  ls -d */ | while read -r d; do
    [ -f "$d/BOUNDARY.md" ] && [ -d "$d/.git" ] && printf '%s ' "$d"
  done
}
# Look for inline distance math (should use MapPos3D methods):
grep -rn '(dx.*dx.*dy.*dy).*sqrt\|distance_2d.*fn\|fn.*distance' \
    --include='*.rs' --exclude-dir=target --exclude-dir=.git $(consumers)

# Look for inline sigmoid/exp (should use substrate sigmoid or tick_feeling_brain):
grep -rn '1\.0\s*/\s*(1\.0\s*\+\|sigmoid\|exp(' \
    --include='*.rs' --exclude-dir=target --exclude-dir=.git $(consumers)

# Look for HashMap-based spatial structures (should use SpatialIndex substrate):
grep -rn 'HashMap.*i32.*i32\|spatial.*hash\|cell.*grid' \
    --include='*.rs' --exclude-dir=target --exclude-dir=.git $(consumers)

# Look for parallel belief/perception types (should use GenericSpatialBelief):
grep -rn 'struct.*Belief\|struct.*Perception\|struct.*Visibility\|last_known' \
    --include='*.rs' --exclude-dir=target --exclude-dir=.git $(consumers)

# Look for parallel emotion types (should use EmotionField/AffectField):
grep -rn 'struct.*Fear\|struct.*Mood\|struct.*Emotion\|fear.*f32' \
    --include='*.rs' --exclude-dir=target --exclude-dir=.git $(consumers)

Audit Step 3 — Classify findings

For each hit, classify:

ClassificationMeaningAction
False positiveThe code is legitimately consumer-specific (e.g., MonsterThreatSource impl)No action — document why
POC simplificationDuplicates substrate but produces identical behavior at POC scaleDocument as known debt; fix when scale changes
DRY violationRe-implements substrate under a different nameFile issue; extract to substrate
Architectural violationViolates two-brain model / sync boundary / domain classificationFile issue; redesign

Audit Step 4 — Report

Summarize findings:

## Substrate-first audit (date)
### Substrate inventory
- [domain]: [primitive] at [location]
### Findings
- [file:line] — [classification] — [description]
### Clean
- [domain] — no violations found

The vocabulary-translation defense (why this skill exists)

The hardest failures to catch are the ones where the substrate exists but ships under a name that doesn't match the concept you're searching for. This is the R296 canonical failure (documented in AGENTS.md §"Manifold Geometry"), applied internally:

You think: "I need a threat field"
You grep:   "threat field" / "spatial hash"  → 0 hits
Substrate:  GenericSpatialBelief + decay_confidence  → exists, fully functional

You think: "I need emotion decay"
You grep:   "emotion decay" / "fear fade"  → 0 hits
Substrate:  tick_feeling_brain + DecayRates + EmotionBaseline  → exists

You think: "I need state persistence"
You grep:   "save state" / "persist"  → 0 hits
Substrate:  LocalKvStore + freeze_avatar_delta + ShardIndex  → exists

The defense: always search 3+ vocabulary variants. The translation table in Mode 1 Step 1 is the canonical reference. Extend it when you discover new mismatches.

Variants hide at THREE depths (measured by the SpatialBelief-family census, ../riir-ai/.docs/05_reasoning/spatial_belief_family.md): the struct name (SpatialBelief ×3 vs the canonical GenericSpatialBelief), the fn name (decay_confidence vs confidence_decay — word-order variants), and the result name (confidence(&self, current_tick) — a method named for what it returns, not its mechanism, in a file named for something else entirely; invisible to every mechanism grep). A single-depth grep returns confident zeros.


When NOT to use this skill

  • Pure refactoring that doesn't add new concepts (renaming, reorganizing)
  • Bug fixes in existing code (the substrate is already consumed or not)
  • Test-only code (tests can define inline helpers)
  • Build/config changes (Cargo.toml, scripts)

Relationship to sibling skills

SkillWhat it checksWhen
substrate-first (this)"Does the substrate already exist? Are you duplicating it?"Before writing code + audit
boundary-guard"Is this code in the right repo? Is the consumer too fat?"After writing code + audit
feature-gate-audit"Do feature-gate claims match source wiring?"Before promoting/demoting flags
goat-audit"Has the katgpt-rs primitive been cherry-picked to riir-*?"Cross-repo cherry-pick tracking
doc-sync"Do docs match git history?"After landing plans/issues

substrate-first is upstream of boundary-guard: if substrate-first catches the drift before it ships, boundary-guard has nothing to find. They're complementary — substrate-first is the prevention, boundary-guard is the cure.


Filing violations

When the audit finds a DRY violation or parallel system:

  1. File an issue in the repo where the violation lives
  2. Reference this skill + the substrate it duplicates
  3. Include the vocabulary translation (what you searched for vs. what the substrate is actually called)
  4. Propose the fix (consume substrate / extract to substrate / revert)
  5. Classify (POC simplification vs. DRY violation vs. architectural violation)

Do NOT fix in the same commit as detection — separate detection from fix so other agents can review the violation independently.


Run log

Compacted 2026-09-11 (user-directed verbose-history pass; the file was 74 KB).

Pruned to the newest 15 rows 2026-10-12 (the 81920B skill-size ceiling; 39 older rows dropped, and two 10-12 rows that had been mis-inserted into the Mode-1 vocabulary table moved back where they belong). git log -p -- .agents/skills/substrate-first/SKILL.md recovers every dropped row verbatim. Rows carried full verification narratives until this compaction; git log -p -- .agents/skills/substrate-first/SKILL.md recovers any of them verbatim (git log -S '<date>' -- <this file> for one row). New rows append ONE line each.

Standing lessons (distilled from the compacted rows — load-bearing audit classes)

  • Body-read before filing a sigmoid/def hit. A local fn sigmoid that DELEGATES to katgpt_core (e.g. a ±50-clamp wrapper) is the SANCTIONED delegation pattern (ndb Issue 611's own fix shape; riir-auth session.rs; chain runtime.rs) — a definition-grep hit is a lead, not a finding. Corollary: katgpt_core::sigmoid is Cephes (~1 ULP), NOT bit-identical to libm-exp forms — delegation is a semantic change where bit-exactness is committed (freeze-commitment paths carry a version-boundary caveat).
  • Documented false-positive classes for distance/sqrt greps: squared-distance comparisons (dx*dx+dy*dy <= r*r — the CORRECT no-sqrt idiom; "fixing" it is a pessimization); wire/snapshot [f32;N] DTO operands (no MapPos3D to delegate); grid-space heightfield math; statistical/financial sqrt (std-dev, AMM LP checked_sqrt); disc radial sampling (r*sqrt(u)); screen-space view-plane placement; #[cfg(test)] sites; closed-form test oracles (PSD max-eigenvalue).
  • Copy-gate convention: a justified copy needs in-source rationale + a divergence-failing test (chain Issue 139 — a doc-only mirror with only self-consistency tests is NOT enough). Documented copy families: BLAKE3(pubkey)[..16] (5 gated instances — consolidate the 6th onto riir_wallet_signer::player_id_hash); splitmix64 finalizer (~20 substrate-side module-locals, all bit-identical — consolidate at the 4th CONSUMER-side arrival, currently 1); cross-repo message formats single-source by IMPORT (dapps imports chain's canonical signing_message).
  • Feature-isolated local math is REQUIRED, not drift — when the feature pulls no katgpt-core (chain forensic f64, congestion inline, the lora bridge keeping katgpt out of the Docker context). Production-dead surface carrying latent duplication under an available dep → delegate-or-prune when consumers materialize (curator_bridge class).
  • Translate CRATE names too, not just type/fn names — mmorpg vs riir-games-mmorpg missed a substrate consumer the issue itself cited (Issue 870: the proposed additive StatMods was refuted by the EXISTING clamped StatModifier, Plan 411).
  • $(var) executes a command named by the variable — the repo-set helper must be a shell FUNCTION called as $(consumers), or the grep silently scans the whole tree.
  • Sanctioned model-consumer shapes (not drift): injected-verifier traits (GmSignatureVerifier pattern); parameterized general form + env-toggle special case (quest_center "one filter home"); feature-neutral placement for gate-union decoupling (HeroStats); Arc<AtomicU8> config seams; two-brain occlusion divergence between substrate and chain verifier (AOI — symmetry pinned on the CHAIN type); plane-split crypto with a hard pub(crate) bar + documented nonce-tradeoff rationale (ndb CLI seal/unseal).

| Date | Scope | Verdict | Record | | 2026-11-17 | Mode 2 mini-audit — the full post-11-13 delta (amo idle lane, riir-refine idle loop zed-edc1a3ad; anchor = the 11-13 row's coverage edge, UTC --since 2026-11-13 00:00 +0000; 792 unique .rs across 20 contract repos [sge 135, remake 115, refine 103, ai 93, infer 90, katgpt-rs 53, seal-std 35, render 34, zed 31 external carrier, train 25 READ-ONLY routing-restricted, reflex 19, chain 19, rethink 14, mmorpg 7, grams 6, ndb 4, sdk 3, dapps 3, instinct 2, ledger 1]; signature greps over the union [fn sigmoid/softmax/log_softmax + inline 1/(1+exp) + new_v4 + SplitMix64 + thread_rng/gen_range + Mutex/RwLock-HashMap + Belief/Perception/Threat structs + blake3[..N] + OsRng + coordinate-keyed maps + fn distance/sqrt ×333] with body-reads on every named-fn hit + pickaxe provenance on both in-window sigmoid candidates; zero-cargo, the box QUIET between the sibling i1226 exclusive cells) | CLEAN — ZERO FILINGS; 2 HARDENING NOTES (both in-window, both the missing-copy-marker class): riir-infer lens/ops.rs:14 pub fn sigmoid — the RelP lens port (Research 666, stdlogits/neuron-lens @ 849bfe16 MIT provenance in module doc), op-level jvp/vjp activation-derivative constants where the one-branch libm-exp form mirrors the upstream estimator's own semantics (katgpt-core IS a direct dep — exact_sigmoid reachable, but delegation would be a numerics change on a fit-instrument lane); missing only the one-line in-source copy-rationale marker — the dspark_drafter.rs:129 class. riir-refine multiplicity_gate.rs:226 f64 gate — IN-WINDOW 57211749 11-14 (i232 T3, opt-in report-only feature), the law in-doc verbatim ("Sigmoid, never softmax") + the named test pin (gate_is_a_counter_sigmoid_never_softmax), exact_sigmoid_f64 delegation reachable, far-tail −710 underflow unreachable at counter-scale z — the locomotion_gait class, marker-only. ADJUDICATED CLEAN (classes): katgpt-core ternary_circuit_forward.rs:754/760 = the i1225 Arm A parity port (brianbell-x/codehygiene @ SOURCE_PIN; "The source's _sigmoid" clamp(−30,30) IS the source's form — Cephes delegation a semantic change; G1-A band contract + Issue-940 fold-order law in-doc); diffusion_sampler.rs:1142 = PRE-window 120339a6a 07-24 DELEGATION (fast_sigmoid, in-doc); lens/fit softmax_into + fkl.rs log_softmax_f64 (i1208 T6 KL audit arm, Issue-940 sequential fold in-doc, KL definitionally log-softmax) + dory/forward.rs softmax (nGPT paper-parity forward, fixed order in-doc) = instrument/forward classes; metal.rs softmax_rows + riir-gpu softmax.wgsl pipelines = GPU kernel numerics; deltanet attention_softmax_row_into = PRE-window 10-25 ONE-home doc holding; gemma2_train backward = trainer math, katgpt-rs-1220-aligned (READ-ONLY restricted, no action); katgpt-types exact_sigmoid/math.rs fused activations = THE substrate homes; zed new_v4 ×5 = external carrier out-of-scope; SplitMix64 ×3 (weight_edit_overlay + discrete_authoring both marker-carrying substrate-side module-locals + rethink trainer example in-doc determinism law) = the documented ~20 family, 4th-consumer trigger unmet; infer mint.rs xorshift64* = seeded partial Fisher–Yates, in-doc "never thread_rng"; Mutex ×18 = resource-handle classes (metal.rs layer-keyed Buffer caches, gemm zero-copy device caches prior-adjudicated, seal-view/quest-server lane statics, refine corpus DATA, zed sessions; sdk authority.rs:482 = the migration-TO-papaya comment itself); Belief structs ×4 = the think-brain pattern WITH citations (ShopOwnerBelief/PlaceBeliefRow in-doc two-brain law, QuestRestockBelief prior-adjudicated, MockBelief test); blake3[..N] = documented KAT families (dapps mintevt/paysig/deposit in-doc formulas) + display seals; coordinate-keyed maps = node-id/span/matrix-index keys, ZERO spatial-position keys (no SpatialIndex bypass); sqrt ×333 sampled across head/mid/tail = geometric/RMS/statistical/closed-form-oracle classes, seal-item KillPos::distance_squared_to = raw wire-DTO squared-reach (correct), seal-render grass/post = render-lane geometry (no MapPos3D in the render crate, BOUNDARY-correct); OsRng = Ed25519 keygen (correct-by-nature) + corpus strings; MonsterThreatSource/HeroThreatLane = prior-adjudicated substrate-side game layer | katgpt-rs (this commit) |

2026-11-13Mode 2 mini-audit — the bandit/selection family (amo idle lane, riir-refine idle loop zed-7d880162; triggered BY the 071-T5 landing: the QueBandit wiring demanded the substrate-first pass it implicitly needed; vocabulary translation over bandit/UCB/Thompson/SafePhased/elo/plackett variants across all 33 contract repos, zero-cargo)CLEAN — ZERO FILINGS; 1 hardening note (pre-window): riir-games-shared delay_bandit/frame_ratio.rs inline UCB1 (ucb1_score/ucb1_best_arm) + the standalone SafePhased mechanism — Plan 112 GOAT'd, feature-gated delay_safe_bandit, materially richer shape (phased-aggression α escalation + soft restarts + per-game delay configs) than the pruners' thin SafePhased strategy variant, but missing only the one-line in-source copy-rationale marker — the 10-30b locomotion_gait sanctioned class, no filing. The family's cross-adjudication is healthy: katgpt-core manifold_bandit's DRY note NAMES every sibling Beta-sampler copy (pruners bandit, fold_bandit, thinking_controller, edge_bandit, partial_scoring_goat) with the Jöhnk's-vs-Gamma-ratio acceptance-rate rationale + the dependency-direction law ("katgpt-core cannot import from katgpt-pruners") + a tracked consolidation follow-up; rating.rs's 4-copy consolidation HOLDS (pruners plackett_luce.rs delegates katgpt_core::rating::elo_from_lambda in-source, Issue 686 cited); CGSP dual_pool = the documented compose partner; riir-router meta_router = the routing-lane consumer shape; the T5 QueBandit itself audits clean (thin composition over the ONLY boundary-reachable bandit substrate — chain's allowlist names katgpt-core, never katgpt-pruners/games-shared). Elo-axis hits = test oracles + the delegating pruners proof lanekatgpt-rs (this commit)
2026-11-05Mode 2 mini-audit — the post-10-31b delta (M3 idle lane, riir-refine idle loop zed-m3-idle3; anchor = the 10-31b carrier 5312b8db5 2026-10-31T20:51Z; the 5-day window — 1691 unique .rs across 20 repos [ai 354, sge 272, remake 243, refine 205, zed 81 external-carrier out-of-scope-for-filings, dapps 75, infer 67, train 63 READ-ONLY, render 61, seal-std 57, chain 42, mmorpg 39, sdk 24, refab 22, katgpt-rs 20, rethink 16, kat 14, ndb 12, reflex 7, instinct 7]; signature greps over the full union via derived path list + body-reads on every hit; zero-cargo, Plan-337 posture — sibling cargo lanes on m3/amo/4090)CLEAN — ZERO FILINGS; 1 hardening note (pre-window): riir-infer-gpu dspark_drafter.rs:129 inline f32 sigmoid — provenance 09-24 8ecbc7f4 (the GGUF/speculative family move; PRE-window, missed by every prior window), a llama.cpp-parity CPU reference port ("the llama.cpp contract" in-doc) = the refab-transformer parity-reference class, but missing that file's in-source "duplicated HERE on purpose" marker — the 1157 delegate menu (exact_sigmoid + bit-pin) is the consistency path if the lane wants it; NOT fresh drift, no filing. Adjudicated in-window classes: elementwise_cubecl.rs sigmoid_f32/softmax_f32/rows_inplace = GPU forward numerics (#[cube] kernels — the 11-04 i1171 tensor-flash wave; CPU substrate fns don't apply to GPU kernels); rethink instinct_encoder_trainer.rs:191 = the trainer-sweep harness class (i662 examples lane); sge lexical.rs:51 = the MODEL CITIZEN — single sanctioned home + the sigmoid-not-softmax rationale in-doc + an in-repo copy-gate TEST asserting no other file re-copies "fn sigmoid(" (the Issue-245 convention — cite as the sanctioned shape); refine traj_store.rs:2109 sigmoid_dot_8 = DELEGATES katgpt_core::sigmoid (the sanctioned class); katgpt-core weight_edit_overlay.rs:1027 module-local SplitMix64 = +1 to the documented ~20 substrate-side module-locals (tolerated class; consolidation trigger stays the 4th CONSUMER-side arrival — player_style.rs proves the types::rng home consumed consumer-side); dflash_config.rs:787 = PRE-window 06-03, prior-adjudicated dflare fusion-init; Mutex/RwLock hits = resource-handle classes (vessel_attestation registry, backstab RootTable e2e closure, infer GPU buffer cache, instinct usage meter; cluster authority.rs hit = the migration-TO-papaya comment itself); the 3 Belief structs (QuestRestockBelief prior-adjudicated + ShopOwnerBelief + PlaceBeliefRow) = the think-brain pattern ITSELF — each doc cites the two-brain law (local, never synced, fog-gated decay); vision_gizmo (u16,u16) = the mesh-handle cache class; blake3 sites = full digests + the blake3_seed(domain‖index LE) documented derivation family; RNG hits = Ed25519 keygen (correct-by-nature) + seeded fixture lanes; zed new_v4 ×5 = external carrier, out of scope for filings. RUN-LOG MAINTENANCE: oldest row dropped, tail held at 15; git log -p recovers itkatgpt-rs (this commit)
2026-10-31bMode 2 mini-audit — the post-10-31-row delta (M3 idle lane, riir-refine idle-loop continuation [context-overflow handoff resuming the queue]; anchor = the 10-31 row's ~18:00Z 10-30 coverage edge; 12 unique .rs across 2 repos [ndb the i641 T1+T2 wave ×5 {evidence_eval battery + embed_contextual + both test lanes + lib wiring}, reflex the --label-perm probe wave ×6 {harness bin + label_perm + runner/label_perm + perm_probe + runner + mod}; boundary-commits verified in-window: riir-ai 1137-T5 15:14Z + seal-remake DOF/overhead-HUD 16:49–16:51Z fall inside the 10-30b/10-31 rows' windows]; signature greps over every delta file + body-reads on every hit + fn-count witnesses on all five zero-hit files; zero-cargo, Plan-337 posture — the i1117 decode leg holds the box)CLEAN — ZERO FILINGS: reflex label_perm = textbook deterministic-seeding DELEGATION (SplitMix64 IMPORTED from the suites.rs substrate home — runner.rs:25's documented MASSIVE seed law; fixed LABEL_PERM_SEED ^ FNV-1a(suite), Sattolo cyclic derangement by construction, BTreeMap throughout, byte-reproducible — the seeded-fixture class, never unseeded RNG); harness.rs:2036 digest[..16] = the display-seal family (a println ack line; the FULL digest is what the read-back refusal above verifies — never a derivation input); perm_probe.rs:1057 sqrt = the L2/RMS statistical class; ndb evidence_eval = pure-f64 battery whose only grep hits are determinism-POSITIVE doc lines ("No HashMap iteration order anywhere" — the Issue-940 law asserted in-source); zero new_v4 / fn-sigmoid / softmax / Mutex-HashMap / RwLock / Belief-struct / coordinate-keyed arrivals in the union; the b321 authoring WIP in riir-refine observed as-found, unaudited per detection-onlykatgpt-rs (this commit)
2026-10-31Mode 2 mini-audit — the post-10-30b delta (M3 idle lane, zed-m3-ownverdict continuation; anchor = the 10-30b audit's ~16:30Z coverage edge; window ≈ 16:30Z → 18:00Z 10-30, 22 unique .rs across 4 repos [riir-train kimi_k3/KD/plan318 trainer wave ×8 — READ-ONLY per the routing restriction, seal-remake overhead-HUD/warp-followup view lanes ×10 + seal-game-editor hud_fx/sealm_sprites ×2 + seal-render bevy_canvas_post(+tests) ×2]; signature greps over every delta file + body-reads on every hit; zero-cargo, Plan-337 posture — the i1117 128K leg holds the box)CLEAN — ZERO FILINGS: riir-train softmax/log-softmax cluster = model-forward TRAINER math (kd.rs KD gradients (softmax(z_S/T) − softmax(z_T/T)) with the ±50 house clamp cited in-doc to distill_shared; kimi_k3_train CE (softmax − one_hot)/L; kimi_k3_lora gated-loss d_logits — the documented trainer-math sanctioned class; lib.rs's shared-math header is the crate substrate itself); seal-view hits = the established handle classes (hp_bars/presence Arc<Mutex<QuestCombatState>> + Mutex<Option<…>> lane statics — papaya N/A, not maps; lib.rs:117 splitmix = the Issue-057 single home VERIFIED holding; damage_popups.rs:977 sqrt = closed-form outline-ring √2 TEST oracle); seal-render + sge delta files zero signature hits; zero new_v4 / fn-sigmoid / Belief-struct / coordinate-keyed / unseeded-RNG arrivals in the unionkatgpt-rs (this commit)
2026-10-30bMode 2 mini-audit — the post-35fa03065 delta (M3 idle lane, riir-refine idle-loop continuation; anchor = the 10-30 carrier 35fa03065 00:16:28Z; 261 unique .rs across 17 repos committed after it [the i1156 qwen38 dflash2/regime-gate wave ×15 infer, riir-gpu relaxed_accept_calibration + hero_routine/quest_combat touches ×38 ai, the seal-render generated-grass + body-shadow waves ×33 + seal-remake gait/review fixes ×35 + sge run_vessel/grass dials ×34, the dapps kat-service contract/routes lanes ×31, refine corpus ×21, reflex runner learnability/synth ×9, ndb walk_rank_poc/bm25 ×9, seal-item mint/loot leaf ×3 seal-std, kat ×6, auth se_custody/se_secret ×6, train cudarc backward mirrors ×6, zed ×6, rethink ×2, deployer ×2, katgpt-rs d2f/flashar/walk_rank ×7]; signature greps over every delta file + body-reads on every hit + pickaxe provenance on both fn-sigmoid sites; zero-cargo, Plan-337 posture — the i1117 128K measurement + the wt-v035 dist release + the i136 seal builds hold the box)1 FILED detection-only (riir-infer .issues/1157): riir-infer-gpu regime_gate.rs:265 fn sigmoid — IN-WINDOW (459984b 08:05+07), katgpt-core a DIRECT dep, doc justifies sigmoid-not-softmax but not the copy itself, no delegation pin — the robust_ab.rs:278 class, site 3; fix deferred (box posture; FIXED same day 23:3x+07 — riir-infer be7d442: delegated to katgpt_core::exact_sigmoid (the bit-stable two-branch form — NOT Cephes sigmoid, whose ±40 clamp a gate must not inherit), pinned bit-for-bit vs the frozen legacy body, issue closed + HISTORY row). 1 HARDENING NOTE, no filing: seal-remake locomotion_gait.rs:72 fn sigmoid (pre-window a00bae11 10-08) — sanctioned feature-isolated local math (katgpt-core NOT a direct seal-view dep, wasm32-first leaf) missing only its one-line copy-rationale marker. REST CLEAN: both 1/(1+exp) clusters = test-oracle/trainer-mirror classes (hero_routine_tests det_trig oracles; riir-train cudarc backward kernels mirroring CUDA-side libm forms where katgpt_core::sigmoid Cephes delegation would be a SEMANTIC change; d2f Zipf power-law fixture; dflash2 walk-confidence softmax test oracle + in-doc "duplicated HERE on purpose" splitmix recomposition); softmax = model-forward numerics (flashar_consensus draft top1/Σexp delegating fast_exp; d2f softmax_argmax_step mechanism name; bevy_canvas_post bilateral falloff weights), corpus rule-STRINGS (kernel_opt tensorfold verbatim quotes), gate-predicate NAMES (softmax_fused_armed); splitmix = walk_rank substrate home + ndb poc fixtures + relaxed_accept_calibration seeded measurement lane (in-doc determinism law) + prior-adjudicated reflex/seal-view sites + seal-item leaf-law local (zero-dep, in-doc); Mutex exactly ONE = sge model_3d_view id-keyed scene-transform cache (app-thread handle class, papaya N/A precedent); RwLock ×6 = refine corpus rule DATA; blake3[..N] = all documented families (shape-id [..12] tamper pins, acct16 [..32], se_custody Debug display-seal, sge bundle content tags, u64-LE stuffing); RNG ×2 = sge run_vessel Ed25519 OsRng keygen (correct-by-nature); zero new_v4 / RwLock / Belief-struct / coordinate-keyed arrivals (QuestRestockBelief + MonsterThreatSource both prior-adjudicated)riir-infer .issues/1157 + katgpt-rs (this commit)
2026-10-30Mode 2 mini-audit — the post-29b delta (amo idle lane, riir-refine idle loop; anchor = the 10-29b carrier 35d30de75 17:49:27Z; 189 unique .rs across 14 repos committed after it [chaind_settle lane ×11 chain + ×20 dapps + ×5 seal-quest-server; the ΩMSR sparse-attention wave ×5 ai + ×5 infer; the reflex harness-runner split ×8 + the 7e3ec2e gate-rho lane; instinct ×11; refine the 1.99 heal wave ×60+; katgpt-core sst_cross_tail + lib re-export; kat ×2; rethink ×9; game-sdk ×2 fixture syncs]; signature greps over every delta file + body-reads on every hit + pickaxe provenance on both fn-sigmoid sites; zero-cargo, Plan-337 posture — the sibling ΩMSR 32K cell holds the box)CLEAN — ZERO FILINGS; adjudicated classes: kernel fn sigmoid katgpt-core lib.rs = THE substrate home itself (fast_sigmoid delegation + exact_sigmoid reference doc — the sst_cross_tail add lane); engine sst_merge_admission test 1/(1+exp) = f64 KL test-oracle (house-rule in-doc); reflex null_probe sigmoid64 + refine rule_embed pub(crate) fn sigmoid = PICKAXE-PROVEN pre-window (10-16 fbdbf10 / 10-08 007435e5, both measured-negative POC lanes, in-doc width/clamp rationale — not fresh drift); softmax = corpus rule-STRING data + bench numpy fixture + the qwen sparse prefill's online-softmax running-max RESCALE in a GPU attention kernel (model-forward numerics, never a decision lane); splitmix = refine corruption.rs SplitMix64 struct (seeded, gate-compliant in-doc) CONSUMED by localization.rs via import (DRY holding) + pav_data local u01 (no-RNG-dep in-doc) + reflex doc-prose only; Mutexes = wallet http.rs scalar/Option handles (round-robin cursor, cert, bearer — rationale in-doc) + do_host LedgerHost handles + bench.rs measured-platform-class note + corpus/proposer fixture strings (papaya N/A, never maps); blake3 [..N] = the documented families (player_id_hash pubkey[..16] copy-gate, explorer display-seal[..8], ledger hex-prefix query[..32], bench-1152 manifest display[..16]); sqrts/exp = L2 norms, Box–Muller, cosine-similarity normalization, Adam vhat.sqrt, geometric classes; zero new_v4 / RNG / RwLock / Belief-struct / coordinate-keyed arrivals anywherekatgpt-rs (this commit)
2026-10-29bMode 2 mini-audit — the post-p008-P4/i1137-T3 delta (amo idle lane, riir-refine idle-loop continuation; anchor = the 10-29T07:49+07 carrier 35d30de75 [00:49:27Z]; 30 unique .rs across 8 repos committed after it: riir-ai the i1137-T3 hero-wave ×8 [hero_routine + quest_combat composition/steer/tick + hero_progress_drive/tactics/threat_lane/threat_memory + i1137_sleep_only_recovery test], riir-infer the plan-623 T3.2/T3.4 verbs_metal/verbs_write examples + multi.rs + qwen38_wgpu_bench + issue1155 exact-slot tests ×2, riir-refine rust_perf corpus entries_52/mod/tests [the bbtree batch-298 wave], riir-rethink origin_gate_gates, riir-viewbridge protocol.rs, seal-game-editor the p008-P4 zstd-press wave ×9 [run_vessel + seal-e2e run_vessel + run_console + feat-map lib + presenter/world + run_session + run_browser + 2 live tests] + this row's own P4 files body-read, seal-render outfit.rs, seal-std seal-anim lib; signature greps over every delta file + body-reads on every hit + fn-count readability witnesses on all 30; zero-cargo, Plan-337 posture)CLEAN — ZERO FILINGS; the wave's centerpiece is textbook substrate consumption: hero_threat_memory imports game_traits::spatial::{belief_confidence, MIN_CONFIDENCE, CONFIDENCE_DECAY_OFFSET} + episodic_sighting::Sighting + game_traits::sigmoid — threat_disc_scale IS the substrate's sigmoid confidence-decay law (belief_confidence(dt, THREAT_DECAY_RATE)) over the Plan-621 sighting vocabulary, and HeroMonsterMemory is a per-monster-id fixed-cap table BUILT ON it, never a parallel GenericSpatialBelief; QuestRestockBelief = the reasoned-refusal shape (doc names SleepTimeAnticipator as wrong-shape substrate, two-brain law cited, local/non-synced EMA — sanctioned); hero_progress_drive splitmix delegates riir_games::swarm::luck (the Issue-1044 designated home); softmax hit = doc-comment prose ("sigmoid, never softmax"); sqrts = geometric over raw [f32;2/3] + the det_trig facing-cone (physical domain, raw — correct); last_known_pos cluster = the think-brain pattern itself, fog-gated through substrate confidence; sge Mutexes = the app-thread handle class (papaya N/A, not maps); sge run_vessel OsRng ×4 = Ed25519 vessel-signing keygen + test throwaway pairs (correct-by-nature); [..N] hits = full-digest blake3 integrity / Ed25519-pubkey-exactly-32 / u32-LE wire framing / test-marker classes; feat-map (u32,u32) HashMap = the prior-row-adjudicated model-keyed mesh cache; zero new_v4 / fn-sigmoid / softmax-impl / parallel-belief-STRUCT / coordinate-keyed / unseeded-RNG / RwLock arrivals in the unionkatgpt-rs (this commit)
2026-10-29Mode 2 mini-audit — the post-970-rollout delta (M3 idle lane, riir-refine idle-loop continuation; anchor = the 10-28T21:36+07 carrier c766e5d88; ~110 unique .rs across 9 repos committed after it: the 970 toolchain-bump+heal wave [riir-gpu KeyBias test-heal ×14 files + lib re-export, riir-auth/kat/infer/refine lint sites + pins], the i188 kat-admin/migrate-normal plane [refine src/tuna_admin + tty_pick + secret_blind + kat admin/genesis/migrate-tally — the two worktree-DELETED files read via git show HEAD:], plan-623 verbs/dist + qwen38-dense + issue1146 [infer-dist ×16 + infer-gpu ×14], refab-bricks wave-2 [conv_decoder/window/root + 3 tests], ndb bm25_facet, deployer origin-vault-v2 [49 KB vault v2 + container net leg + dual-slot roll], remake warp/audio/boxes/assigned_shaders; signature greps over every delta file + body-reads on every hit + fn-count readability witnesses, one deployer re-scan after a transient file-not-found to close the silent-zero hole; zero-cargo, Plan-337 posture)CLEAN — ZERO FILINGS; five adjudicated classes: refine drafter.rs:286 fn sigmoid = the ndb Issue-611 DELEGATION wrapper verbatim (katgpt_core::sigmoid, rationale in-doc — the sanctioned class); infer issue1146 test hits = test-oracle class (f64 gate formula + a PINNED-ORDER f64 log_softmax for the KL measurement — reduction order pinned in-doc per the Issue-940 law; KL is definitionally log-softmax); refab transformer.rs hit = the 10-28-adjudicated torch-parity doc-comment; auth account_key OsRng ×3 = Ed25519 KEY GENERATION (correct-by-nature); riir-gpu dflash_config::softmax_normalize_fusion = dflare_training-gated fusion-weight INIT utility (training-plane spec math, old surface riding the heal wave's lib re-export — not a decision lane); corpus/router hits = rule-name data + doc-comments; zero new_v4 / splitmix / Belief-struct / coordinate-keyed-HashMap / unseeded-RNG arrivals anywherekatgpt-rs (this commit)
2026-10-28Mode 2 mini-audit — the auth-v2-client/ground-marks/body_lod delta (amo idle lane, riir-refine cycle; the gap window this cycle's Protocol-H two-pass ff opened past the prior row's coverage: the SAME ten roots the boundary-guard 305th hunk-verified, fresh .rs surface = riir-kat auth-v2 client tier (kat_keyset_client + kat_identity + kat_genesis_client/protocol + kat_ua TOTP seam + kat_account_client learn_service_root), riir-dapps worker t2.rs factor gate + key add/revoke routes + genesis worker + Issue-149 drain fix, riir-auth se_account.rs + ssh_picker.rs + service_root, seal-render look/lod.rs body tier law + art_cache.rs LruArtCache + clip_floors + ground decal + drape step law, seal-remake body_lod.rs + ClipLift + skills panel, seal-std seal-fx drape ground sheet, sge model_stance + quest-search MATCH atom cap (issue 273) + riir-refab decode-oracle loader fix; signature greps over every delta range + body-reads on every hit; zero-cargo under the 641-T5 pp32k grind, Plan-337 posture)CLEAN — ZERO FILINGS: zero softmax / new_v4 / fn-sigmoid / splitmix / thread_rng / RwLock arrivals across all eight code ranges; the sha256 hits = riir_auth::totp_sha256 consumed (TOTP's RFC 6238-mandated HMAC-SHA family — substrate consumed, never re-implemented; the auth-v2 wire stays blake3/ed25519 throughout); the sge MATCH cap (32 atoms / 4 KiB) = expression-hardening over a SQLite FTS5 builder, not a behavior-lane threshold (no FSM bypass); seal-scene's LruArtCache = the ONE decode-cache primitive both the viewer body lane and the editor scenery caches consume (the BOUNDARY Owns row landed same-commit — single-source-by-primitive holding, not parallel-cash drift)katgpt-rs (this commit)
2026-10-28Mode 2 mini-audit — the auth-v2/MotionBricks/bands delta (amo idle lane; the gap window this cycle's Protocol-H ff opened past the post-102-discharge row's coverage: EIGHT ff'd roots, the fresh .rs surface = riir-refab crates/refab-bricks NEW lane crate ~15 files (nn/vqvae/transformer/rot6d/soma/tpose/predictor/checkpoint/safetensors/npy + 5 test files, the MotionBricks pose-conditioned port), riir-auth key_set.rs (the auth-v2 device key-set state machine) + se_custody.rs (SE-wrapped seed AEAD) + kat-se Swift tool (non-Rust), riir-dapps worker auth-v2 planes (auth_v2_contract.rs + keyset resolve/mining/recovery routes + tests, +1618/−58), riir-kat kat_protocol_keyset.rs/kat_protocol_authz.rs wire vocabulary + sge plan-452/453 skill-bands + seller-shop in-crate .rs + seal-std seal-skill bands twin; signature greps over every delta file + body-reads on every hit; zero-cargo under the 641-T5 pp32k attempt, Plan-337 posture)CLEAN — ZERO FILINGS; ONE ADJUDICATED FALSE POSITIVE: refab transformer.rs exp(x − max)/Σ = an ML forward PORT whose spec is torch's own softmax (doc-comment names the form; the T2 torch-parity gate 1.2e-5 IS its oracle — reproducing the reference numerics is the contract, not a decision-lane violation; the sigmoid-not-softmax law governs workspace decision/scoring lanes, never parity references); auth-v2 vocabulary = pure wire/state-machine surfaces over riir-auth's own verify primitives (the 26h consume-don't-copy form again — no inline crypto, no new_v4, blake3 the hash throughout: mining blake3(lint_key), recovery-id blake3(stamp), shape-id tamper checks recompute blake3(shape)[..12] against the stored pin); zero fn-sigmoid / splitmix / unseeded-RNG / Belief-struct / coordinate-keyed-HashMap arrivalskatgpt-rs (this commit)
2026-10-28Mode 2 mini-audit — the post-102-discharge delta (amo idle-lane continuation; the gap window this cycle's Protocol-H ff opened past the two 07:0xZ rows' coverage: the seal plan-446-step-2 effect-soft-particles wave, 8 fresh .rs across 3 repos [sge feat-map lib.rs +41−2, seal-render bevy_canvas +48−16 / skill_fx trio +169, seal-std seal-fx classify/quads ±8]; signature greps over every delta file + body-reads on every hit; zero-cargo under the 641-T5 soak, Plan-337 posture)CLEAN — ZERO FILINGS; the wave is a CONSOLIDATION: all skill-fx surfaces consume seal_fx::effect::{flight_frame, …} wholesale (the single-source-by-leaf law holding — the canvas scales a returned [f32;16] frame's rotation block frame[..12], never re-derives it); every sqrt = the geometric class over raw [f32;3]/matrix components (feat-map placement distance/scale/length, bevy_canvas vector len, apply column norm); the one Mutex = feat-map TEST-fixture ResolveRecording(Arc<Mutex<Vec<u32>>>) (fixture-handle class); zero new_v4 / fn-sigmoid / softmax / splitmix / OsRng / Belief-struct / blake3-truncation / coordinate-keyed-HashMap arrivalskatgpt-rs (this commit)
2026-10-28Mode 2 mini-audit — ISSUE-102 DISCHARGE (amo idle lane, riir-refine cycle; the 970 lane filed reflex robust_ab.rs:278 as the router.rs copy-gate class site 2, fix deferred to a quiet window; this lane claimed (597dc0e) and discharged it same cycle: delegation to katgpt_core::exact_sigmoid_f64 + the sigmoid_delegation_matches_frozen_legacy_body pin (−200,200 sweep + [−26,14,0,40,200] anchors, bit-identical x≥0 / ≤3 ULP below, far tail < −710 scoped OUT in-test — the one-branch underflow-to-0 vs two-branch-subnormal gap no reachable arg approaches); the 10-27 wave audit's interim "admissible local form" verdict superseded by the filed issue + pin-enforced discharge; lib 487/0, clippy all-targets all-features −D clean, rustfmt clean; reflex HISTORY 102 row)SITE-2 COPY-GATE CLOSED — the class is now at ZERO known instances workspace-side (site 1 router.rs:111 discharged 10-24b sigmoid_delegation_matches_frozen_legacy_body; site 2 robust_ab.rs:278 discharged here; engine.rs + nli_m1 pins predate); no other inline fn sigmoid f64 body outside the documented false-positive classes (test-gate names, delegation-pinning tests, the exact_sigmoid_f64 wrapper itself)katgpt-rs (this commit)
2026-10-28Mode 2 mini-audit — the post-10-27 delta (amo idle-lane continuation [the katgpt-handoff session]; union re-grep over 307 fresh .rs across 19 repos since the 10-27 row's coverage [dapps 66, infer 65, train 28, refine 28, ai 21, remake 18, sge 16, ndb 13, reflex 9, chain 9, katgpt-rs 9, auth 7, kat 5, rethink 4, sdk 4, render 2, instinct 2, seal-std 1; vendor/block ×4 = vendor class]; NEW surface = the reflex harness-runner split [robust_ab/lanes_hosted/lanes_wire/report/run_phases/router_stats_probe/engine_tests], the seal-harness-evolution crate [family/lib/adapter+drive gates — zero signature arrivals], sge capture lanes + treasure-box trio + skill_effect_follow, dapps recovery/migrate day-2 [recovery rows, epoch_slice failsafe row, self_path.rs via git show], infer dist landed surface [stage/planner/handshake/multi/runner — the sibling 623 lane's unstaged WIP observed as-found, unaudited per detection-only]; signature greps + body-reads on every hit; zero-cargo under the league gate2 driver + bench_809, Plan-337 posture)1 FILED → DISCHARGED SAME-CYCLE: reflex robust_ab.rs:278 inline f64 sigmoid 1/(1+exp(−x)) = the router.rs:111 copy-gate class site 2, substrate REACHABLE — THIS lane filed it detection-only (c2017a1, the delegate menu, fix deferred for build contention; ⚠ the discharge row ABOVE reads the filing as "the 970 lane" — the filing commit is this lane's, the attribution likely rides the 190f244 toolchain-bump adjacency); the sibling lane then claimed 597dc0e → fixed b1fc3fe (exact_sigmoid_f64 + frozen-body pin) → closed 5546cea within the hour — class at ZERO, verdict = the row above's. REST CLEAN: train l4_margin_probe fast_sigmoid = thin delegation wrapper; train/dapps/ndb/infer RNG hits = the seeded test/probe fixture class (in-doc "no rand dep"/"pinned"); dapps mining_rewards reward_key = blake3(epoch‖account‖record)[..16] formula in-doc; epoch_slice acct[..8] = the credited-µKAT LE PAYLOAD into the 16-byte failsafe row (row encoding, never a truncation); recovery [..32] nonce / [..16] acct rows = the acct16 wire family; auth totp k[..32] = the HMAC pad (26g holding); sdk chain_seed + refine campaign_digest = the 26g families re-verified; infer msl_field Mutex<HashMap> output_cache = the device-scoped ZeroCopyState resource handle (papaya N/A); zero new_v4 / fn-softmax / belief-struct / coordinate-keyed-HashMap / unseeded-global-RNG arrivals in the unionreflex c2017a1 + katgpt-rs (this commit)
2026-10-27Mode 2 mini-audit — the post-26h delta (amo idle lane, fmt-landing-wait cycle; UTC anchor 26h 19:06Z — methodology note: a naive git log --since parses LOCAL +07 and double-counted the 26h wave's own eae0f05/2a680dc; anchors are UTC; TRUE fresh .rs surface = 3 files in 3 repos: dapps 915b39e + infer 1f99c7e both deny.toml + scripts/deny_gate.sh ONLY (zero .rs, zero dep-graph edges — the b294 T2.4 fan-out's own Research-332 record covers them) + infer c60b34d (i1124-t1.5, the ONE audited .rs, prefill_cuda_full.rs +64/−8) + train caa334c7 margin-probe compile fix (READ-ONLY, routing-restricted); signature greps + body-reads; zero-cargo under the 645/469 arms + i1117 measurement + elephant window-3 capture, Plan-337 posture)CLEAN — ZERO FILINGS: c60b34d = pure kernel-dispatch wiring behind prefill_mmq_v2 cfg + env-gated RIIR_PREFILL_FFN_GLU — failure degrades bit-safely to the incumbent pair path (the 1124-T1 contract in-source), once-only AtomicBool REACH log (the standing reach-evidence pattern), zero state/map/truncation/RNG/sigmoid arrivals; train example = dead-GPU-context removal + doc-comment T0a verdict with re-open trigger (trainer-math sanctioned class per the 26f row)katgpt-rs (this commit)
2026-10-26hMode 2 mini-audit — the post-26g delta (amo idle lane; anchor = the 26g carrier de2b006aa 19:06Z; ONE fresh wave on the audited surface: dapps eae0f05 issue-139 totp gate + its docs close 2a680dc — 32 fresh .rs [src/kat/totp.rs 576-line STORE half, cloudflare totp_routes.rs ROUTE host + totp_contract.rs vectors, ~25 gate-wiring touches across the kat-service contract/route files, migrate_sweep class-Z row families]; ai's two origin commits predate the anchor — already inside 26g's HEAD-state read; sibling WIP observed as-found and untouched [dapps M tests/kat_mining.rs; this repo kv960 M-files + untracked deny.toml/deny_gate.sh = the b294-fanout lane] — the COMMITTED state audited via git show; signature greps + body-reads on every hit; zero-cargo under the 645 base arm + the 469/plan341 margin probes, 3 compute consumers ~600–850% CPU, Plan-337 posture held)CLEAN — ZERO FILINGS; the wave is the textbook consume-don't-copy form: riir_auth::{verify_totp, totp_sha256, TotpVerdict} named in-source as the ONLY verify primitive ("consumed, never copied") — totp_contract's oracle calls totp_sha256, totp.rs carries ZERO math (its sole hmac mention names the primitive in a doc-comment), totp_routes zero crypto hits; the worker adds only what the primitive cannot (persistence: the accepted step IN the action's CommitBatch with early-persist ledger/cross-plane variants in-doc; the pending→confirm enrollment machine; the 5-fail/15-min cooldown where replays-never-count); disarm = zeroing-never-delete (class-Z law in-source); the store UNGATED with the burn/credit precedent cited; signature sweep zero new_v4 / fn-sigmoid / softmax / splitmix / RNG arrivals; [..N] hits all documented families (blake3(pk)[..16] pinned wire contract, [..12] shape-ids, migrate formulas in-doc, one test fixture slice); Mutex hits = the pre-existing worker resource-handle classes; base32_of = the 20-line RFC-4648 encoder with vectors (zero-new-deps posture, codec not substrate); SystemTime::now ×1 new = the worker clock seam. RUN-LOG MAINTENANCE: oldest row (10-19) dropped, tail held at 15; git log -p recovers itkatgpt-rs (this commit)
2026-10-26gMode 2 mini-audit — the post-26f delta (amo idle lane; anchor = the 26f carrier 71ae535a7 16:45Z — NOT the 26e time-anchor, which double-counts the 26f wave; 15 fresh .rs across 6 repos [ai 2 {plan-642 harness_evolution runtime consumer + overlay gates — audited at HEAD state via git show; the worktree carries UNSTAGED deletions of both + dense quest_combat M-files = the in-flight plan-642 lane rework, as-found, unaudited per detection-only; origin/develop ahead of local HEAD}, sge 1 {editor-core run_browser}, refine 3 {blueprint lane}, ndb 3 {moedern falsifier test lanes}, sdk 3 {scenelab arena_generations + goat bench}, auth 6 {totp/rebind/provider_identity/provider_login/signing_session + warm bench}]; signature greps + readability witnesses (grep -c 'fn ') on every zero-hit file — the silent-zero trap; body-reads on every hit; zero-cargo under the 645 softmax arms — 3 compute consumers at ~900–1000% CPU, Plan-337 posture held)CLEAN — ZERO FILINGS: refine campaign_digest = the documented derivation family (formula + truncation in-doc: BLAKE3 over mode‖posture‖sorted relative paths, 16-byte hex, BTreeMap determinism); sdk chain_seed = the documented seed-derivation family (BLAKE3 over fixed-width LE, first 8 bytes as u64, T1.2 generation-chaining law in-doc) + corpus_hash full digest; ai runtime = full-digest AuthoringExport integrity (no truncation), the overlay lane carrying native-only/wasm32 posture + apply-path law + freeze-custody in-source; sge run_browser Arc<Mutex<…>> pair = the app-thread BundleState/opened-URL cell class (papaya N/A, not maps); auth totp SHA-256 = RFC-6238 interop-mandated (blake3-impossible by protocol, rationale in-doc; k[..32] = the HMAC block pad, never a truncation); gates fixture goldens = full-digest pins; zero new_v4 / fn-sigmoid / softmax-impl / belief-struct / coordinate-keyed-HashMap / unseeded-RNG arrivals anywhere in the wave. RUN-LOG MAINTENANCE: tail pruned to the newest 15 (dropped 10-18d + the misordered 10-20a; git log -p recovers both)katgpt-rs (this commit)

References

  • AGENTS.md §"Spatial Cognition (Two-Brain Model)" — the canonical perception rules
  • AGENTS.md §"Manifold Geometry (Stokes Calculus)" — the R296 vocabulary-translation failure pattern (this skill extends it from paper→code to codebase→codebase)
  • AGENTS.md §"Latent vs Raw Space Rules" — domain classification
  • riir-games-shared/src/game_traits/spatial.rs — GenericSpatialBelief<T> substrate
  • ../riir-ai/.docs/05_reasoning/spatial_belief_family.md — the SpatialBelief family census + the ONE decay law belief_confidence (ex-Issue 1051's consolidated record; the issue file was removed 2026-10-29)
  • riir-games-shared/src/game_traits/ — vocabulary translation table source
  • Issue 047 (riir-mmorpg-examples) — the ThreatField canonical failure
  • Issue 490 + Issue 493 (riir-game-sdk) — the orchard/motivation canonical failure
  • boundary-guard skill — sibling (boundary enforcement, post-hoc)
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 467kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 467kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 467kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 193kUpdated Claude CodeCodexVerifiedInstall