Fold Abide as productized soft-rule lint; kev Hub/NOTA delta by 24601 · Pull Request #2 · 24601/Augustus

24601 · 2026-09-18T19:07:25Z

Why this PR exists

#1 merged as v0.3.0. This branch is the next hourly System One watch fold. Docs-only: skill, mental-model, and architecture notes — not a hit list.

Archer Hume 27B drop is still WATCH. Omni perception→decision does not wait for it.

Non-negotiable: exact work stays in code or policy; the model owns narrow judgment; never launder a soft Noul as a proof. No wrapper, no SDK/plugin how-to, no copied ports/key-file paths. No invented metrics. TypeSafe Jev is the exemplar, not the monopoly. Causes below are hunches, not promoted Contract.

This update — 16:30 UTC / ~10:30 Boise 2026-09-19 (`research/notes.md` §86)

Docs-only dedicated productization of uehaj/jev-semgrep (light-noted §61; not a re-card). Grep by meaning via Jev Noul. Commit f1e2847. Skip Archer. No wrapper. No invented metrics. Do not copy npm / npx / .env / marketplace. Not a gate (rh-guard skip).

Material how-to-apply:

Proposition ≠ embedding — README theirs: the line and the question are read together (cross-encoder shape). Who did what, negation, and “asked for” vs “already done” change the answer. An embedding of the line is fixed before it ever sees the query.
Contrast-set pedagogy — all six lines are “about a refund”; only the customer-asking lines pass. Cosine between angry customer and angry support agent is close to 1.
Boolean composition of soft judgments — AND/OR/NOT are boolean ops on thresholded bits in code, not negative-query tricks. Do not multiply parallel p (independence caveat; ≠ jev-combinators digital-design metaphor). Calibrated ~0.5 vs top-k cosine.
No index — whole corpus every query; vector index wins for repeated large fixed corpora. Cross-lingual (EN meanings safer near threshold).
Semgrep.dev name collision — rename if both installed. ≠ SAST. ≠ jevgrep (file/chunk packed search).

Name locks: ≠ semgrep.dev ≠ jevgrep ≠ jev-sift ≠ jevex ≠ jev-combinators. Stars ephemeral (0 → SIGNAL ★42 → 51 this pass). 0.94/0.98 is LLM-as-judge on 10 cases × 51-line corpus theirs, not Harbor. HEAD 21120e9; README SHA 923e6a5. LICENSE MIT / GitHub SPDX NOASSERTION.

Reviewer entry points (this fold)

research/notes.md §86 — meaning-grep HIGH (hunches labeled); quote README
references/mixed-architecture.md — fail table: proposition ≠ embedding; boolean after threshold; Semgrep.dev / not a gate
references/faq.md — “Is jev-semgrep Semgrep.dev? Are AND/OR/NOT embedding tricks? Is it a gate?”
references/applied-mappings.md §4 / mappings.md §4 — dedicated delta over the §61 paragraph
references/validation.md contrast-set row; composition-algebra.md item 26

Sections changed (this fold)

Skill: SKILL.md; mixed-architecture.md; faq.md; mental-models.md; agent-self-assessment.md; applied-mappings.md; mappings.md; validation.md; toolbox-mapping.md; composition-algebra.md; question-design.md; methods-catalog.md
Research: notes.md §86; sources.json (464 sources, 461 unique URLs, retrieved 2026-09-19T16:30Z); refresh-log.md; archive/findings.md batch 0923 fold: novel HIGH + REVISIT densify #69
Index: CHANGELOG.md; README.md; docs/ecosystem.md
Unchanged: evaluate_decisions.py

Verification

evaluate_decisions.py --self-test ok
sources.json 464 / 461 unique (three pre-existing duplicate URLs left untouched)
git diff --check clean

Prior on this branch — 16:25 UTC / ~10:25 Boise 2026-09-19 (`research/notes.md` §85)

Docs-only external pedagogy / how-to-apply, not a new species and not a TypeSafe how-to. @akshay_pachaar “Jev Clearly Explained” / article. LLM hammer; schema-safe ≠ correct; 200×/400× TypeSafe ceiling theirs. Commit e6e9e42.

Prior on this branch — 16:20 UTC / ~10:20 Boise 2026-09-19 (`research/notes.md` §83–§84)

Docs-only dual-signal fold. reddpy/AgentGhost wrap-as-execution ALLOW/ASK/DENY (MIT 2★; ASK throws; fail-closed; rh-guard owns the gate cousin). @studio_yebisu JP genre atlas — apps by hole, not a bake-off; stars research-time. Commit a18d6b4.

Prior on this branch — 16:10 UTC / ~10:10 Boise 2026-09-19 (`research/notes.md` §82)

Docs-only productized ASR observe→score-among-candidates→code-acts. moritzkremb/jev-voice-browser (MIT; 103★). Spoken confirm ≠ auth. 27/27 fixtures theirs ≠ Harbor. Commit 38ae518.

Prior on this branch — 16:05 UTC / ~10:05 Boise 2026-09-19 (`research/notes.md` §81)

awlevin/typesafe-computer-use (MIT; 427★). Never screenshot-to-frontier for the decision. 155× theirs one screenshot ≠ Harbor. Commit a5c9d33.

Prior on this branch — 15:50 UTC / ~09:50 Boise 2026-09-19 (`research/notes.md` §80)

khordoo/jev-reflex-autonomy-lab delta of §46. S1 never stalls; S2 never grants. Commit 4b8298c.

Prior on this branch — earlier folds

§79 hourly 0842 (74c0c18); §78 JevBench v1.2 (40a5f12); §77 @airesearch12 census (db654b5); §76 NandhaKishorM/laya (bcf66f1); §75 githubnext/localjev (daa70b7); §74 choxos/jev-reviewer (03fddc6); §73 classifier-dev (5f44bf4); §72–§50 as previously documented on this branch.

Verification

Docs-only; no app UI to browser-test
evaluate_decisions.py --self-test ok
sources.json parses (464 sources, retrieved 2026-09-19T16:30Z; 461 unique URLs — three pre-existing duplicate URLs left untouched)
git diff --check clean
Latest commit f1e2847 on existing PR Fold Abide as productized soft-rule lint; kev Hub/NOTA delta #2 branch cursor/augustus-store-envelope-00b4; no duplicate PR

…ctice Archer drop still Watch — no architecture rewrite. Novel vs the 11:59 digest: sqlite-jev in-engine vs jevql CLI; bitrate-advisor and the JOB planner as soft judgment inside a hard envelope; jev-routing host adapter; jev-claw classify-then-policy. Already-folded HIGH get frames only (memory gate, receipts economics, encoder vs decoder replica). Docs-only; no Jev wrapper. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Place jaredpalmer/kev beside Laya / Nimble / Archer Watch: Qwen2.5-0.5B LoRA + pointer, Apache-2.0, POST /v1/systemone drop-in. Public gold, not a Jev teacher. Contrast vs TypeAR, encoder DeBERTa, proprietary Jev. Cite README isolation/ECE/acc/permute/IIA/forgery; laptop-local development/eval, not a knowledge substitute. Docs-only; no serve how-to. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…n LoRA Cite Hub/GitHub receipts only. Archer 27B remains Watch. Skill cards carry mental models (omni decide without waiting, ECE/NLL/Brier bake-off, decision-token training, allowlist-as-proof, forced Choice, S1 keeps control), not a hit list or serve how-to. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Cite README + replay receipt only. Mental models: soft rules vs linter, edit vs turn observation window, banded fail-open, rubric as artifact. jev-pref stays the contract; rh-guard stays eval-integrity. Not a hit list, not multimodal, no hook how-to. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Do not rewrite §45. Hub jaredpalmer/kev-0.5b; PEFT FEATURE_EXTRACTION. Choice other must appear as a wrong alternative too — wellposed request lint is not enough. No invented NOTA rates. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only fold of novel categorization/scoring placements into PR #2: extractive quotes + offline redecide, pointer-not-generator, contract- compatible /v1/systemone (stub until hf), structured observe→decide→act, dataframe accessors (jevframe beside jevpandas), route≠memory, advisory sidecar, structure induction, collab-arm measurement, AST∩semantic lint. Archer remains Watch. No wrappers or copied how-tos. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Independent when-it-holds atlas (extractable-from-state), DMB frozen protocol vs constrained LLMs, jevals-data recompute-from-logs, and Kahneman S1/S2 cascade. Brief: ARC combinatorial negative, packed open-LLM System One, tiny SAN local drop-in, kev star delta. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Place m-newhauser/gliner25-compaction as architecture notes: pointer compaction vs summarizers; fail-closed keep_full under a mutation envelope; same job as pi/fast-jev-compaction with a GLiNER backend; shadowMode default true. Not Jev, not multimodal. Docs-only. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only architecture notes for latch (cluster-then-policy PASS/BLOCK), wakegate (fail-open VOI resume), GLiNER S1 indexer (10-50x unfilled), clear-head claim/evidence Stop, Harbor on/off routing, and cousins. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only: sahibzada-allahyar/gliner2-ultrafast proves observe→score-among-candidates→code-acts is backend-agnostic (Jev Ultrafast ↔ local GLiNER2). Hybrid local decide + remote fill; DONE ≠ verified success. Numbered §52 after the 16:48 Boise watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only: tamaratran/jev-pruner Noul-prunes command output after a hard ≤10k/JSON-diff envelope, before the main LLM sees it. Same extractive family as fast-jev-compaction / gliner25-compaction; different job. Fail-safe keep original; archive for recovery. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only: trycua/cua libs/cua-s1 option-attention among observed elements (fill/check/click/skip). Plan ≠ execute; dry-run default; source-only. Not TypeSafe Jev — parallel S1 naming in CUA. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…eval Fold jevify (uncalibrated local likelihoods), decision-native RAG, carryforward, hunch, explore-typesafe-ai, and jev-baselines-eval (honest negative + cascade sign-flip) into PR #2 skill cards. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only: jev-sift classify-first (retrieve-wide → decide → evidence-set on agent I/O) and jevable.com class patterns (not a 342-title dump). No invented metrics; Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only: draft stack #2951–#2955 as harness productization of observe→score-among-candidates→code-acts. Extract judge + pick-and-copy uses their 37/75 ~0.5s vs 4.37s card; pick is a fast path, not a replacement. No invented metrics; Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…tness Docs-only skill/architecture notes for ask-jev-ai, jevgrep, egma-ai/jev-reviewer, jev-oxlint, jev-adaptive-thinking, and Jev-RAG (plus MED pointers). Harbor/jevals practice, not a hit list. TypeSafe Jev remains exemplar, not monopoly. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…ation arena Interlock as ring-0 sensor≠policy (secrets never in agent; type-safe ≠ correct). Typed control plane around DSPy (drafts after route+action). Native-probability jev-arena plus sonar/vickrey/bracket fan-out. Stockfish owns truth / Jev owns judgment. Human-confirmed port-cleanup gate UX. Brief toolbelt; skip parody/farming. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…er, ORDER BY ranking, GLiFormer backend Docs-only: Domain-jev-maker calibration/VOI, jav-email-cascade Harbor compare arms, jev-orderby-bench ranking family, jeff wire-compat economics, sysone loopback gateway. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…g-grep, closed-vote CU, Jev vs PCD Harbor Fold seven HIGH System One signals into PR #2 as how-to-apply / mental-model / architecture / toolbelt + Harbor/jevals notes. Skip empty jev-compactor and laya-jolt. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…ion outline, eval instrument-not-score Weave omp-greenlight, skill-broker (outline only), and dinostomp into how-to-apply / mental models / architecture / Harbor-jevals notes. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…e ≠ verdict, attention/VOI never-block Docs-only. Hunches labeled. Archer still Watch. slo-router measured p95 77.93→490.38 ms same routes; construct-auto-classifier Jev 0 dangerous/975; jev-lens never blocks and never green unless sure. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…ranking ≠ calibration Harbor/jevals practice: treat question packs as measured artifacts (measurement owns endorsement). Canonical SENSOR slogan: Jev supplies evidence, code owns authority. Ranking ≠ calibration — never hard-threshold raw decision-model p as a frequency. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…-self-train hides Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…rboard, OOD/AUC≠ECE, turnstile Harbor/jevals practice: compose System One as control-plane primitives (Then/Gate/Vote/Cascade/Weighted, not chat turns); receipts not leaderboard; skill-library VOI with hook fail-open; AUC ≠ ECE and sign of miscalibration by type; attach thinking budget on frontier-100; turnstile joins actiongate (evidence ≠ authority). Docs-only. Hunches labeled. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…edger, sureness, JevBench v1.1, CI typed gate Harbor/jevals practice: compose formal/semi-formal methods with Jev-class judgment (escalate instead of hard-gate); coverage ledger answers whether the world may change; Choice confidence = max_prob (generous); calibration reported not scored. Docs-only. Hunches labeled. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

… tools≠use, open-Jev class, HA-Jev Docs-only fold of novel HIGH plus material deltas into PR #2. Do not re-fold 02:38 items except sibling contrast. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…routing bench, zeroshot displacement, typed handoff Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…r, sentence-as-rule lint, open replica substrates Docs-only. Measurement crystallizing (calibration+cost first-class); pick≠fill; ast-grep × sentence lint; VOI hunk admission; class substrates under one contract. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…, jev-gpt never-generates, cookbook/jevfeed, openJev-verdict claim-audit Harbor-shaped Jev vs schema-guided LLM-as-judge (no quality headline yet); mixed-architecture control-plane productization; generation as a tree of Choices; applied recipes plus a personal-history feed; competing NAR claims as calibration/claim-verification, not endorsement. Empty IPECTER compaction-proxy skip. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…xir SDK, commitjev/jevex VOI, mailordinal, laya-multilingual Docs-only. Mental models (decision theory, calibration, VOI, signal detection, anti-soundness-theater), not a thin Jev skill. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…shold, silent FALLBACK Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…Noul, human check as product Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

… structured-read, Harbor bake-off Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…e-p, vs-Jev honesty Docs-only. Hub Laya already watched; this hour is PyPI + Router + published vs-Jev caveats. Skip Archer. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Quote the Benchmark Heaven tweet as a named census, not a rank. GLiNER2 and routers are a class-boundary; incompleteness vs our watch is lag. Watch the board URL; do not paste live scores here. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Measurement / Harbor-adjacent practice, not a hit list. Quote the live board; mark numbers theirs. Calibration now on the rank; weighting is a product design; option-order 72→21; instruction models in the class table; latency ×2 and est. costs named. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

24601

Interim adversarial review (parent — freeze in progress)

Fold agent is draining then freezing for merge. Checklist to clear before APPROVE:

No invented metrics — every number marked theirs or quoted from README/board.
Name collisions — always disambiguate choxos/jev-reviewer (systematic review) vs egma-ai/jev-reviewer (code attention); uehaj/jev-semgrep ≠ Semgrep.dev; githubnext/localjev wire-compat ≠ OpenJev logit-read.
Archer — Qwen3.8-27B / Chutes / LocalJev DiffusionGemma are not Archer Hume’s drop (HF archerhume still empty).
Schema-safe ≠ correct — soft Noul / confidence gates must not read as hard safety without calibration/Harbor caveat.
Class discipline — hosted TypeSafe product APIs and prompted LocalJev are not shared-prefix multimodal species.

Will re-review after freeze + dedicated adversarial agent returns.

…undness-theater PR gate Named HIGHs were already §73–§78. Extract how-to-apply (wire≠logit, FALLBACK, packaging honesty, pointer-not-generator, census≠score) instead of re-carding. Skip leftover thin. Archer still Watch. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

24601

Adversarial review — WATCH (draft; freeze first)

Fail-closed against HEAD 74c0c181cd4d0ee9efbd8208c59eecf69a6545cb (skill / notes §78–§79). Docs-only. Independent of Jev-omni merge order.

Check	Verdict	Falsifier / proof
Invented metrics	PASS	Board digits marked theirs; v1.1 87.6 (cal off) vs v1.2 75.3 (geo-mean I/C/S/K, cal ON) fail-tabled as not comparable. Live board matched #1 Jev 75.3 / #2 SemIf 74.6 / OpenJev 67.6 / Luna I 96.8 rank #7 / option-order 72%→21%. ×2 latency named assumption.
Class-table inflation	PASS	Packaging ≠ species (NandhaKishorM/laya); githubnext/localjev kept out of NAR clones; instruction models on the board called class-boundary.
Name collisions	WATCH	choxos ≠ egma-ai held; uehaj/jev-semgrep ≠ Semgrep.dev held (§61 / skip table); githubnext/localjev wire ≠ logit held. Unnamed third: `getvoicify/jev-reviewer` (0★ TS) not listed. Census likes/views marked ephemeral (live X already past SIGNAL 417/9).
Archer Hume false positive	PASS	HF `archerhume`/`4rcherhume` 404 this watch; Qwen3.8 27B Chutes TEE ≠ Archer; drop still WATCH.
Soft Noul as hard safety	PASS	Fail table + §79: hard-gate Noul as PR/quality = soundness theater (`totally-tim/jev-gate`, `claude-jev-warden`). 0.85 / Noul ≥ 0.5 stay theirs / product copy, not Harbor τ.
Silent-fallback	PASS	classifier-dev granite FALLBACK F1 0.546 vs advertised ~0.800 named; recipe says mark FALLBACK.
Duplicate / contradictory	WATCH	§73–§78 already folded; §79 is apply-the-five not a re-card — good. Draft still accruing hourly folds; freeze before APPROVE.

Before APPROVE

Freeze hourly intake (this is still a draft).
Optionally name getvoicify/jev-reviewer as a third collision.
Do not promote census engagement or ×2 latency to Harbor ground truth.

No code pushed. Independent of Jev-omni #2/#3. COMMENT, not APPROVE.

24601 · 2026-09-19T15:46:54Z

+| Collapse instruction models into NAR clones because they share the table | **Fail closed** (class-boundary) | Luna/Gemini/DeepSeek/Qwen3.8 are JSON-schema instruction models. Needle 3 is function-calling (C none → 0). OpenJev on board = razorback16 ≠ IamBusy `/v1/decide` |
+| Treat the geometric mean as a natural law | **Fail closed** (weights are a choice) | Limits *theirs*. Balanced no-cal puts SemIf #1; Emphasis Cost puts system-one-open #1 / Jev #5 |
+| Re-card localjev / classifier.dev / Laya / choxos / census / v1.2 because they reappear on the hourly | **Fail closed** (already folded) | Apply the five as a recipe (`notes.md` §79). Do not dump the hit list again |
+| Hard-gate a Noul as a PR merge / quality seal | **Fail closed** (soundness theater) | Soft Noul attends or escalates; an exact envelope proves the irreversible act. [totally-tim/jev-gate](https://github.com/totally-tim/jev-gate) (0★; MIT; Action/CLI/OpenCode) and [connectedGraph/claude-jev-warden](https://github.com/connectedGraph/claude-jev-warden) (1★; MIT; “Art Director Warden”) are this hour’s skip with that risk. **≠** [vinilana/jev-gateway](https://github.com/vinilana/jev-gateway), [MongLong0214/jev-gate](https://github.com/MongLong0214/jev-gate) (model routing), [SargeDev/jev-gate-student-b](https://huggingface.co/SargeDev/jev-gate-student-b). Cousin: ci-gatekeeper (cheap typed pre-review, operator-owned). Attention filter never blocks ([rashedInt32/jev-lens](https://github.com/rashedInt32/jev-lens)) |


PASS (soft Noul ≠ hard merge). Fail table correctly labels hard-gating a Noul as PR/quality seal as soundness theater, with name collisions (jev-gate vs jev-gateway vs Hub student-b). Calibration caveat is present. Do not later “productize” this as a merge gate without an exact envelope.

24601 · 2026-09-19T15:46:56Z

+| Collapse GLiNER2 or routers into NAR / TypeSafe clones | **Fail closed** (class-boundary) | GLiNER2 locates/categorizes; Succinct 14M / jev-model-router / Director / Loki route. Same job family ≠ replica. Needle 3 is function-calling (already §67) |
+| Treat an incomplete openjev list as our watch being wrong | **Fail closed** (census lag) | Laya, githubnext/localjev, kev, TypeAR, openvons, chakuho, jevinf, grande, laya-jolt, blackwood, classifier-dev missing. Lesson, not a dunk |
+| Quote tweet likes/views as quality | **Fail closed** (ephemeral) | SIGNAL ~417/9/3; this pass 564/15/5. Do not copy Stripe |
+| Mix v1.1 Main 87.6 with v1.2 Score 75.3 as a drop | **Fail closed** (not comparable) | Different tiers and scoring *theirs*. Cal now ON the composite. Keep v1.1 as historical (`notes.md` §67, §78) |


PASS (v1.1 87.6 vs v1.2 75.3 not comparable). Fail row holds the configuration difference (cal off Main vs geo-mean I/C/S/K 25% each, cal ON). ×2 latency / est. costs called assumptions. Qwen3.8 27B ≠ Archer. OpenJev-on-board = razorback16 ≠ IamBusy /v1/decide. Matches live board at review time.

24601 · 2026-09-19T15:46:58Z

+   Formal/semi-formal: deterministic policy is the hard
+   gate; Jev is soft evidence.
+
+3. **[`Adilmp/does-jev-confidence-mean-anything`](https://github.com/Adilmp/does-jev-co


WATCH (name collisions mostly held; one unnamed third). choxos ≠ egma-ai is explicit. uehaj/jev-semgrep ≠ Semgrep.dev is in the skip table. githubnext/localjev ≠ kunchenguid/local-jev is explicit. Unnamed: getvoicify/jev-reviewer is a third jev-reviewer (0★ TS) not listed here — add on freeze if the hourly still scans GitHub name collisions. Census likes/views are marked ephemeral (live X metrics already moved past SIGNAL 417/9). Draft: freeze before APPROVE; no invented metrics found. Archer HF archerhume/4rcherhume still 404.

…ion telemetry, Local ≠ localjev Docs-only §80 of PR #2. Delta of the §46 one-liner, not a new species. Quote README: S2 is advisory and does not fly; Jev does not pause. Purple confidence marks consumed guidance; Local controller is rule-based, not githubnext/localjev; 20% starting gate is still soft. Skip Archer. No npm/.dev.vars recipes. No invented metrics. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…ve actions, 155× one-screenshot Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…eech VOI, spoken confirm ≠ auth Docs-only §82 of PR #2. Expand the §39 tweet into productized ASR perception: Web Speech → one 9–11-question Jev request → Playwright. Closed-set may act on a partial; free-text waits. Pointer spans. Spoken confirm is convenience, not authorization. Overlay numbers need no second model. 27/27 fixtures are theirs, not a Harbor taskset. Skip Archer. No npm/.env recipes. No invented metrics. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only: reddpy/AgentGhost ALLOW/ASK/DENY wrap is execution (rules first, ASK throws, fail-closed); JP tweet is an application genre atlas, not a bake-off. Skip Archer. rh-guard owns the gate cousin. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Docs-only: LLM hammer, schema-safe ≠ correct, questions-as-code, shadow rollout. Mark 200×/400× as TypeSafe ceiling claims, not Harbor. Skip Archer. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

…g, Semgrep.dev collision Dedicated §86 over the §61 light note: boolean composition of thresholded Nouls, refund contrast-set, name collision, not a gate. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

) * Fold hourly 1047 HIGH + deferred 0945 into Augustus how-to-apply. Docs-only off main (PR #2 merged). Mental models / architecture / Harbor-jevals for 13 Boise 1047 signals plus Laya vision, Cerebellum-2B, and laya-grounded. Formal methods compose with scoring; Noul is SENSOR; hard-gating test/PR/HA write is soundness theater. Skip Archer. Quote theirs; no wrappers or invented metrics. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Land mapping-index cards the 1047 review required. Cerebellum in applied-mappings §5; hourly FM envelopes out of the fill-in template; compaction species locks pi-jev-compact and fast-jev-compaction-pi; VOI/Leveson/search cards now match SKILL.md. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Fold queued SIGNALs + jevsubrouter into Augustus mental models. Docs-only on PR #3. Jeff-1 LoRA replica vs logan-markewich/jeff, stanley empty≠approve, findme beam ≠ JevFind, jevsubrouter prices workers. Soft Noul ≠ hard safety. Quote theirs; no wrappers. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> * Close §88 review Importants: mapping-index landings, JevFind cousin, verbatim quote. Retarget Jeff-1 unique fragment into the when-to-use table; land stanley-code notChecked in the formal-methods envelope; cite item 30 for findme ≠ JevFind; quote jevsubrouter README verbatim. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com> --------- Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Folds in the plan-v3 errata (D-a..D-e), every open finding from the Fable 5.1 xhigh and Astra max v3 reviews, the M0 tabputer-1 receipt deviations, and the decision-changing items from the two trainer sweeps and the PAW/RAP, CLM and Jev-Omni cards. Main changes: - Empirical Bernstein intervals for bounded losses carry the family error claim; normal-theory intervals become descriptive only (Astra v3 #2). - Acquisition custody: quarantine, augctl-owned splitting, label-free confirmation inputs, and B16 as an adversary test (Astra v3 #1). - Three named download windows, with the design lock before any experiment data (Fable v3 P1-1, Astra v3 #3). - The ladder becomes artifact forms: exact program, synthesized program, compiled function (PAW), model rungs, generalist, under one gate. - RAP as required prior art: exogenous rule checks are detection, not gating. - The measured host profile: uids 48201/48202, --memory-swap=16g, systemd-run --user, explicit GPU budgets, a real MemAvailable watchdog, and vLLM batch nondeterminism handling. All numbers are recomputed by research/080/calc/calc_v4.py. Amp-Thread-ID: https://ampcode.com/threads/T-01a0d176-6c28-75e9-a727-f948d3747917 Co-authored-by: Basit Mustafa <basit.mustafa@gmail.com>

cursoragent and others added 2 commits September 18, 2026 19:06

cursor Bot changed the title ~~Fold 12:58 Boise watch: store-index fork, hard envelopes, harness practice~~ Fold 12:58 Boise watch + kev (runnable Archer reconstruction) Sep 18, 2026

cursor Bot changed the title ~~Fold 12:58 Boise watch + kev (runnable Archer reconstruction)~~ Fold 14:03 Boise watch: open multimodal RLCD, bake-off, decision-token LoRA Sep 18, 2026

cursor Bot changed the title ~~Fold 14:03 Boise watch: open multimodal RLCD, bake-off, decision-token LoRA~~ Fold Abide as productized soft-rule preference lint Sep 18, 2026

cursor Bot changed the title ~~Fold Abide as productized soft-rule preference lint~~ Fold Abide as productized soft-rule lint; kev Hub/NOTA delta Sep 18, 2026

cursoragent and others added 21 commits September 18, 2026 21:04

Fold 00:39 Boise remainder: hot-click CU, verbatim compact+gate, anti…

223967d

…-self-train hides Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

cursoragent and others added 10 commits September 19, 2026 10:56

Fold 04:39 Boise watch: digital-design combinators, VOI cache, skill-…

0265a9d

…routing bench, zeroshot displacement, typed handoff Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Fold classifier-dev: productized System One HTTP, escalate-under-thre…

5f44bf4

…shold, silent FALLBACK Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Fold choxos/jev-reviewer: systematic-review pointer, two-pass Choice+…

03fddc6

…Noul, human check as product Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Fold githubnext/localjev: wire-compat ≠ logit-equiv, prompted JSON vs…

daa70b7

… structured-read, Harbor bake-off Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Fold NandhaKishorM/laya: packaging ≠ new species, Router script-befor…

bcf66f1

…e-p, vs-Jev honesty Docs-only. Hub Laya already watched; this hour is PyPI + Router + published vs-Jev caveats. Skip Archer. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

24601 commented Sep 19, 2026

View reviewed changes

cursoragent and others added 6 commits September 19, 2026 15:50

Fold awlevin/typesafe-computer-use: OCR+AX observe→score→act, exclusi…

a5c9d33

…ve actions, 155× one-screenshot Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

Fold Akshay Pachaar “Jev Clearly Explained” as pedagogy

e6e9e42

Docs-only: LLM hammer, schema-safe ≠ correct, questions-as-code, shadow rollout. Mark 200×/400× as TypeSafe ceiling claims, not Harbor. Skip Archer. Co-authored-by: Basit Mustafa <24601@users.noreply.github.com>

24601 marked this pull request as ready for review September 19, 2026 16:32

24601 merged commit f916b4f into main Sep 19, 2026

24601 deleted the cursor/augustus-store-envelope-00b4 branch September 19, 2026 16:35

24601 mentioned this pull request Sep 19, 2026

Fold hourly 1047 HIGH + queued SIGNALs + deferred 0945 into Augustus #3

Merged

24601 mentioned this pull request Sep 20, 2026

Fold hourly 1248 HIGH #46

Merged

Navigation Menu

Repository navigation

Fold Abide as productized soft-rule lint; kev Hub/NOTA delta - #2

Fold Abide as productized soft-rule lint; kev Hub/NOTA delta#2
24601 merged 43 commits into
mainfrom
cursor/augustus-store-envelope-00b4

24601 commented Sep 18, 2026 •

edited by cursor Bot

Loading

Uh oh!

24601 left a comment

Uh oh!

24601 left a comment

Uh oh!

24601 Sep 19, 2026

Uh oh!

24601 Sep 19, 2026

Uh oh!

24601 Sep 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

24601 commented Sep 18, 2026 • edited by cursor Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why this PR exists

This update — 16:30 UTC / ~10:30 Boise 2026-09-19 (research/notes.md §86)

Reviewer entry points (this fold)

Sections changed (this fold)

Verification

Prior on this branch — 16:25 UTC / ~10:25 Boise 2026-09-19 (research/notes.md §85)

Prior on this branch — 16:20 UTC / ~10:20 Boise 2026-09-19 (research/notes.md §83–§84)

Prior on this branch — 16:10 UTC / ~10:10 Boise 2026-09-19 (research/notes.md §82)

Prior on this branch — 16:05 UTC / ~10:05 Boise 2026-09-19 (research/notes.md §81)

Prior on this branch — 15:50 UTC / ~09:50 Boise 2026-09-19 (research/notes.md §80)

Prior on this branch — earlier folds

Verification

Uh oh!

24601 left a comment

Choose a reason for hiding this comment

Interim adversarial review (parent — freeze in progress)

Uh oh!

24601 left a comment

Choose a reason for hiding this comment

Adversarial review — WATCH (draft; freeze first)

Before APPROVE

Uh oh!

24601 Sep 19, 2026

Choose a reason for hiding this comment

Uh oh!

24601 Sep 19, 2026

Choose a reason for hiding this comment

Uh oh!

24601 Sep 19, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

24601 commented Sep 18, 2026 •

edited by cursor Bot

Loading

This update — 16:30 UTC / ~10:30 Boise 2026-09-19 (`research/notes.md` §86)

Prior on this branch — 16:25 UTC / ~10:25 Boise 2026-09-19 (`research/notes.md` §85)

Prior on this branch — 16:20 UTC / ~10:20 Boise 2026-09-19 (`research/notes.md` §83–§84)

Prior on this branch — 16:10 UTC / ~10:10 Boise 2026-09-19 (`research/notes.md` §82)

Prior on this branch — 16:05 UTC / ~10:05 Boise 2026-09-19 (`research/notes.md` §81)

Prior on this branch — 15:50 UTC / ~09:50 Boise 2026-09-19 (`research/notes.md` §80)