pen-stack 3.4.0__tar.gz → 4.0.1__tar.gz

{pen_stack-3.4.0 → pen_stack-4.0.1}/CHANGELOG.md

@@ -3,6 +3,45 @@
 All notable changes to PEN-STACK are documented here. This file follows
 [Keep a Changelog](https://keepachangelog.com/) and the program's phase structure.
 
+## [4.0.1] - 2026-06-09 - data-correctness patch: writer-verification panel verified against Perry 2025
+
+### Fixed
+- **WS-WV frozen panel is now verbatim from the measured Perry 2025 ISCro4 DMS.** The offline-fallback panel
+  in `atlas/writer_verify.py` previously used *illustrative* Z-scores (2.6/2.1/1.7) and invented control
+  variants (G15D/P88R/L120E), and `_CORE_RESIDUES` used illustrative arginines. Replaced with the REAL values
+  from `science.adz0276` Table S3: the top-3 enhancers **N322P (Z 0.754), H50K (0.742), R278M (0.709)**, real
+  near-neutral variants (V21R, S312Q, G286T), the most-deleterious variants (R132E −5.40, R137E −5.12,
+  R195D −4.98), and the documented catalytic residues **D11/E60/D102/D105/S241** ("Residue Groups" sheet). The
+  real-DMS path (on the VM/Drive) was already correct; only the offline fallback constants were illustrative.
+  Added `test_ws_wv.py::test_frozen_panel_matches_real_perry_dms_table_s3` to guard against drift.
+
+## [4.0.0] - 2026-06-09 - v4.0 release: the Oracle Mesh (on top of the foundation models) + writer verification
+
+A major bump: the substrate now *composes* the biomolecular foundation models under one contract and verifies
+the writer enzyme itself. Workstreams WS-{O,WV,ATLAS}, each SHA-locked. No de-novo writer invention — score
+and critique only (the pen-assemble lesson).
+
+### Added
+- **WS-O - the oracle mesh.** `pen_stack/oracles/` with `OracleResult{value, provenance(model+version),
+  native_uncertainty, scope_card, in_scope, extrapolating, output_kind, available, cached}`. Adapters:
+  `genome.py` (AlphaGenome OOD-gated; Evo2 likelihood=claim / generation=candidate; ChromBPNet·Borzoi
+  baseline), `structure.py` (AlphaFold3/Boltz-2/Chai-1/Protenix + `consensus()` that widens the interval on
+  cross-oracle disagreement), `protein_design.py` (RFdiffusion/ProteinMPNN/ESM3 - all candidates), `rna.py`
+  (ViennaRNA - real, hard fold-legality), `energetics.py` (bridge off-target, MC3 gate ≥0.77).
+  `configs/oracles/scope_cards.yaml` (11 models); deterministic version-pinned `oracle_cache/`. Guard:
+  generative candidate `as_claim()` raises. `docs/oracles.md`; `prereg/ws_o.yaml`.
+- **WS-WV - writer verification.** `pen_stack/atlas/writer_verify.py`: DMS- + structure-grounded variant
+  scoring (measured=claimable, unmeasured=not), `blind_recovery` recovers N322P/H50K/R278M above
+  measured-worse controls, and `critique_candidate` (fold/active-site/deliverable/reachable) wired into
+  `verify()` as `Verdict.writer_critique` - always `no_claim=True`. `docs/writer_verification.md`;
+  `prereg/ws_wv.yaml`.
+- **WS-ATLAS - mesh upgrade + delivery oracle.** `wgenome/mesh_features.py` (OOD-gated feature hook + honest
+  blind re-validation reporting parity vs v3.x when oracles are deferred) + a computable
+  `delivery.aav_packaging_margin` soft rule (titre drops near the AAV capsid limit). `prereg/ws_atlas.yaml`.
+
+### Changed
+- Version 3.4.0 -> 4.0.0; `Verdict` gains `writer_critique`; M1 + writer-verification note + M2 updates.
+
 ## [3.4.0] - 2026-06-09 - v3.4 release: the Environment (train/eval surface + bench v0.3 + outcome-calibration)
 
 v3.4 turns the thin Gym interface into a full environment an AI agent can be trained and graded in, ships

{pen_stack-3.4.0 → pen_stack-4.0.1}/CITATION.cff

@@ -1,7 +1,7 @@
 cff-version: 1.2.0
 message: "If you use PEN-STACK, please cite it as below."
 title: "PEN-STACK: open infrastructure for genome writing"
-version: 3.4.0
+version: 4.0.1
 date-released: 2026-06-01
 authors:
   - family-names: "Mahaboob Ali"

{pen_stack-3.4.0 → pen_stack-4.0.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pen-stack
-Version: 3.4.0
+Version: 4.0.1
 Summary: Open infrastructure for genome writing: the Writable Genome atlas, the Writer Atlas, and the Write Planner.
 Author-email: Anees Ahmed Mahaboob Ali <ahmedaneesm@gmail.com>
 License: MIT
@@ -89,8 +89,8 @@ and durably write new DNA, **which enzyme** can write it there, and **how** to d
 [![codecov](https://codecov.io/gh/ahmedanees-m/pen-stack/branch/main/graph/badge.svg)](https://codecov.io/gh/ahmedanees-m/pen-stack)
 [![License: MIT](https://img.shields.io/badge/License-MIT-informational.svg)](LICENSE)
 [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/)
-[![Version](https://img.shields.io/badge/version-3.4.0-blue.svg)](CHANGELOG.md)
-[![Tests](https://img.shields.io/badge/tests-190%20passing-success.svg)](tests/)
+[![Version](https://img.shields.io/badge/version-4.0.1-blue.svg)](CHANGELOG.md)
+[![Tests](https://img.shields.io/badge/tests-208%20passing-success.svg)](tests/)
 [![Lint: ruff](https://img.shields.io/badge/lint-ruff-purple.svg)](https://github.com/astral-sh/ruff)
 [![Runtime: Docker](https://img.shields.io/badge/runtime-docker-2496ED.svg)](docker/)
 [![Validation: pre-registered](https://img.shields.io/badge/validation-pre--registered-critical.svg)](prereg/)
@@ -133,6 +133,25 @@ Two questions gate every genome-writing project, and before PEN-STACK no resourc
 Everything is built on bulk-downloadable public data, runs on a single GPU, and is validated **blind** against
 a pre-registered, honest baseline before release.
 
+## What is new in v4.0 — the Oracle Mesh (sitting on top of the foundation models)
+
+v4.0 makes PEN-STACK the **composition + verification layer over the biomolecular foundation models**. It
+wraps AlphaGenome, Evo2, AlphaFold3, Boltz-2, Chai-1, Protenix, ESM3, RFdiffusion and ProteinMPNN under one
+contract that carries each model's provenance, native uncertainty, and a **scope card** stating what it is
+valid for — then routes their outputs through the rule-grounded verifier and the calibrated trust layer. A
+generated sequence or structure is always a **candidate to be checked, never a claim**. For the writer enzyme
+itself, v4.0 builds **verification, not invention**: proposed/variant writers are scored against measured DMS
+data and predicted structure, recovering known enhanced variants blind and refusing to assert activity for
+anything unsupported.
+
+| Workstream | What it adds | Result |
+|---|---|---|
+| **O — the oracle mesh** | `pen_stack/oracles/` — `OracleResult{value, provenance(model+version), native_uncertainty, scope_card, output_kind}`; adapters for genome / structure / protein-design / RNA / energetics; deterministic version-pinned cache | one contract; **generative output = candidate** (`as_claim()` raises — the pen-assemble lesson in code); AlphaGenome **OOD-gated**; cross-oracle **disagreement widens the interval**; ViennaRNA + energetics real |
+| **WV — writer verification** | `atlas/writer_verify.py` — DMS- + structure-grounded variant scoring; candidate **critique** wired into `verify()` | recovers the known enhancers (**N322P / H50K / R278M**) above measured-worse controls; unmeasured variants flagged, **not claimable**; a generated writer is critiqued (fold/active-site/deliverable/reachable), **never returned as a working pen** |
+| **ATLAS — mesh + delivery oracle** | `wgenome/mesh_features.py` (OOD-gated feature hook + honest blind re-validation) + a computable **AAV packaging-margin** delivery rule | atlas re-validation reports **parity** vs v3.x when oracles are deferred (delta 0.0, never hidden); titre-margin flag fires near the AAV capsid limit; immunogenicity magnitude stays a scope flag |
+
+See `docs/oracles.md`, `docs/writer_verification.md`, and `prereg/ws_{o,wv,atlas}.yaml`.
+
 ## What is new in v3.4 — the Environment (a place to train and grade genome-writing AI)
 
 v3.4 makes PEN-STACK the surface an AI agent can be **trained and graded** in, the counterpart to v3.3's
@@ -377,8 +396,9 @@ pen-stack/
 │   │                                   + v3.2 offtarget_energetics (position x substitution; held-out 0.88, ships)
 │   ├── agent/                        agentic platform: tools / orchestrator / pen_agent / mcp_server / guardrails
 │   │                                   + v3.2 epistemic (3-tier status) / scope (known-unknowns matcher)
+│   ├── oracles/                      v4.0 L1 oracle mesh: OracleResult contract + adapters (genome/structure/protein_design/rna/energetics) over the foundation models; version-pinned cache
 │   ├── rules/                        v3.3 machine-readable rules engine (schema/evaluators/loader/solver) over configs/rules/*.yaml
-│   ├── verify/                       v3.3 verification service: verify(design) -> Verdict (legal+reasons+confidence+scope)
+│   ├── verify/                       v3.3 verification service: verify(design) -> Verdict (legal+reasons+confidence+scope; v4.0 writer_critique)
 │   ├── adapt/                        local recalibration / private-data adaptation behind a gate (v3.1, WS-F)
 │   ├── env/                          v3.4 full Gymnasium environment over router+verifier (genome_writing_env + policies; [env] extra)
 │   ├── monitor/                      PEN-MONITOR living database (Europe PMC)

{pen_stack-3.4.0 → pen_stack-4.0.1}/README.md

@@ -14,8 +14,8 @@ and durably write new DNA, **which enzyme** can write it there, and **how** to d
 [![codecov](https://codecov.io/gh/ahmedanees-m/pen-stack/branch/main/graph/badge.svg)](https://codecov.io/gh/ahmedanees-m/pen-stack)
 [![License: MIT](https://img.shields.io/badge/License-MIT-informational.svg)](LICENSE)
 [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/)
-[![Version](https://img.shields.io/badge/version-3.4.0-blue.svg)](CHANGELOG.md)
-[![Tests](https://img.shields.io/badge/tests-190%20passing-success.svg)](tests/)
+[![Version](https://img.shields.io/badge/version-4.0.1-blue.svg)](CHANGELOG.md)
+[![Tests](https://img.shields.io/badge/tests-208%20passing-success.svg)](tests/)
 [![Lint: ruff](https://img.shields.io/badge/lint-ruff-purple.svg)](https://github.com/astral-sh/ruff)
 [![Runtime: Docker](https://img.shields.io/badge/runtime-docker-2496ED.svg)](docker/)
 [![Validation: pre-registered](https://img.shields.io/badge/validation-pre--registered-critical.svg)](prereg/)
@@ -58,6 +58,25 @@ Two questions gate every genome-writing project, and before PEN-STACK no resourc
 Everything is built on bulk-downloadable public data, runs on a single GPU, and is validated **blind** against
 a pre-registered, honest baseline before release.
 
+## What is new in v4.0 — the Oracle Mesh (sitting on top of the foundation models)
+
+v4.0 makes PEN-STACK the **composition + verification layer over the biomolecular foundation models**. It
+wraps AlphaGenome, Evo2, AlphaFold3, Boltz-2, Chai-1, Protenix, ESM3, RFdiffusion and ProteinMPNN under one
+contract that carries each model's provenance, native uncertainty, and a **scope card** stating what it is
+valid for — then routes their outputs through the rule-grounded verifier and the calibrated trust layer. A
+generated sequence or structure is always a **candidate to be checked, never a claim**. For the writer enzyme
+itself, v4.0 builds **verification, not invention**: proposed/variant writers are scored against measured DMS
+data and predicted structure, recovering known enhanced variants blind and refusing to assert activity for
+anything unsupported.
+
+| Workstream | What it adds | Result |
+|---|---|---|
+| **O — the oracle mesh** | `pen_stack/oracles/` — `OracleResult{value, provenance(model+version), native_uncertainty, scope_card, output_kind}`; adapters for genome / structure / protein-design / RNA / energetics; deterministic version-pinned cache | one contract; **generative output = candidate** (`as_claim()` raises — the pen-assemble lesson in code); AlphaGenome **OOD-gated**; cross-oracle **disagreement widens the interval**; ViennaRNA + energetics real |
+| **WV — writer verification** | `atlas/writer_verify.py` — DMS- + structure-grounded variant scoring; candidate **critique** wired into `verify()` | recovers the known enhancers (**N322P / H50K / R278M**) above measured-worse controls; unmeasured variants flagged, **not claimable**; a generated writer is critiqued (fold/active-site/deliverable/reachable), **never returned as a working pen** |
+| **ATLAS — mesh + delivery oracle** | `wgenome/mesh_features.py` (OOD-gated feature hook + honest blind re-validation) + a computable **AAV packaging-margin** delivery rule | atlas re-validation reports **parity** vs v3.x when oracles are deferred (delta 0.0, never hidden); titre-margin flag fires near the AAV capsid limit; immunogenicity magnitude stays a scope flag |
+
+See `docs/oracles.md`, `docs/writer_verification.md`, and `prereg/ws_{o,wv,atlas}.yaml`.
+
 ## What is new in v3.4 — the Environment (a place to train and grade genome-writing AI)
 
 v3.4 makes PEN-STACK the surface an AI agent can be **trained and graded** in, the counterpart to v3.3's
@@ -302,8 +321,9 @@ pen-stack/
 │   │                                   + v3.2 offtarget_energetics (position x substitution; held-out 0.88, ships)
 │   ├── agent/                        agentic platform: tools / orchestrator / pen_agent / mcp_server / guardrails
 │   │                                   + v3.2 epistemic (3-tier status) / scope (known-unknowns matcher)
+│   ├── oracles/                      v4.0 L1 oracle mesh: OracleResult contract + adapters (genome/structure/protein_design/rna/energetics) over the foundation models; version-pinned cache
 │   ├── rules/                        v3.3 machine-readable rules engine (schema/evaluators/loader/solver) over configs/rules/*.yaml
-│   ├── verify/                       v3.3 verification service: verify(design) -> Verdict (legal+reasons+confidence+scope)
+│   ├── verify/                       v3.3 verification service: verify(design) -> Verdict (legal+reasons+confidence+scope; v4.0 writer_critique)
 │   ├── adapt/                        local recalibration / private-data adaptation behind a gate (v3.1, WS-F)
 │   ├── env/                          v3.4 full Gymnasium environment over router+verifier (genome_writing_env + policies; [env] extra)
 │   ├── monitor/                      PEN-MONITOR living database (Europe PMC)

pen_stack-4.0.1/configs/oracles/scope_cards.yaml

@@ -0,0 +1,114 @@
+# PEN-STACK v4.0 — oracle scope cards (WS-O0). What each wrapped foundation model is VALID for, and what it
+# is NOT — so the substrate can gate and label outputs (the field's evidence that these models do not
+# generalize to unseen loci is made legible here, not hidden). `output_kind`: claim (a checkable prediction),
+# candidate (a generative proposal that must pass writer-verification), baseline (an honest comparator).
+version: "1.0"
+
+oracles:
+  alphagenome:
+    family: genome
+    version: "2025.1"
+    output_kind: claim
+    valid_for: "regulatory-track + variant-effect prediction at IN-DISTRIBUTION loci (trained tracks/tissues)"
+    not_valid_for: "unseen loci / cell types outside training; does NOT generalize to novel regulatory contexts"
+    generalizes_to_unseen_loci: false
+    license: "non-commercial (Google DeepMind terms)"
+
+  evo2:
+    family: genome
+    version: "40b-2025"
+    output_kind: candidate            # generative DNA + likelihood; sequences are proposals, never claims
+    valid_for: "genomic sequence likelihood / zero-shot variant scoring; generative DNA candidates"
+    not_valid_for: "accessibility/expression QTLs; quantitative regulatory tracks; asserting a sequence WORKS"
+    generalizes_to_unseen_loci: false
+    license: "Apache-2.0 (Arc Institute)"
+
+  chrombpnet_borzoi:
+    family: genome
+    version: "borzoi-2024"
+    output_kind: baseline             # kept as an honest comparator to AlphaGenome
+    valid_for: "accessibility / expression baseline tracks (honest comparator)"
+    not_valid_for: "variant effects beyond trained assays"
+    generalizes_to_unseen_loci: false
+    license: "open"
+
+  alphafold3:
+    family: structure
+    version: "3.0-2024"
+    output_kind: claim
+    valid_for: "protein / protein-NA complex structure at confidence (pLDDT/PAE) within trained fold space"
+    not_valid_for: "absolute binding free energies; novel folds far from the PDB; in-vivo behaviour"
+    generalizes_to_unseen_loci: true   # structure prediction is not locus-bound
+    license: "non-commercial weights (DeepMind terms)"
+
+  boltz-2:
+    family: structure
+    version: "2.0-2025"
+    output_kind: claim
+    valid_for: "structure + binding-affinity prediction (open weights); cross-oracle consistency comparator"
+    not_valid_for: "guaranteed affinities; designs outside trained chemical space"
+    generalizes_to_unseen_loci: true
+    license: "MIT"
+
+  chai-1:
+    family: structure
+    version: "1.0-2024"
+    output_kind: claim
+    valid_for: "structure prediction; cross-oracle self-consistency"
+    not_valid_for: "absolute affinities; far-OOD complexes"
+    generalizes_to_unseen_loci: true
+    license: "Apache-2.0"
+
+  protenix:
+    family: structure
+    version: "0.5-2025"
+    output_kind: claim
+    valid_for: "AF3-style structure prediction (open); cross-oracle self-consistency"
+    not_valid_for: "absolute affinities; far-OOD complexes"
+    generalizes_to_unseen_loci: true
+    license: "Apache-2.0"
+
+  esm3:
+    family: protein_design
+    version: "sm-2024"
+    output_kind: candidate
+    valid_for: "protein representation + generative protein design CANDIDATES; variant likelihoods"
+    not_valid_for: "asserting a designed protein FOLDS or is ACTIVE without verification"
+    generalizes_to_unseen_loci: true
+    license: "non-commercial / community"
+
+  rfdiffusion:
+    family: protein_design
+    version: "aa-2024"
+    output_kind: candidate
+    valid_for: "backbone generation CANDIDATES (RFdiffusion / RFdiffusion-AA)"
+    not_valid_for: "asserting function; a backbone is a proposal, not a working enzyme"
+    generalizes_to_unseen_loci: true
+    license: "open (BSD-style)"
+
+  proteinmpnn:
+    family: protein_design
+    version: "ligandmpnn-2024"
+    output_kind: candidate
+    valid_for: "sequence design for a fixed backbone CANDIDATES (ProteinMPNN / LigandMPNN)"
+    not_valid_for: "asserting activity/specificity; must be scored against measured data"
+    generalizes_to_unseen_loci: true
+    license: "MIT"
+
+  viennarna:
+    family: rna
+    version: "2.6"
+    output_kind: claim
+    valid_for: "RNA secondary-structure MFE fold (a HARD legality input for bridge-RNA QC)"
+    not_valid_for: "tertiary structure; in-cell folding kinetics"
+    generalizes_to_unseen_loci: true
+    license: "open"
+
+  bridge_energetics:
+    family: energetics
+    version: "v3.2-mc3"
+    output_kind: claim
+    valid_for: "bridge IS110/ISCro4 off-target relative-risk ranking (beats the 0.77 position-weight baseline)"
+    not_valid_for: "absolute off-target rates; non-bridge writers; a non-recombining background"
+    generalizes_to_unseen_loci: false
+    license: "open (this work)"

{pen_stack-3.4.0 → pen_stack-4.0.1}/configs/rules/delivery.yaml

@@ -29,6 +29,15 @@ rules:
     provenance: { doi: ["10.1089/hum.2017.084"], note: "v3.2 MC2 delivery_constraints scan" }
     test_ref: "tests/unit/test_ws_r.py::test_delivery_controls"
     scope: "labeled heuristic, directional; not a titre predictor"
+  - id: delivery.aav_packaging_margin
+    kind: soft_penalty
+    category: delivery
+    mechanism: "AAV packaging efficiency / titre drops sharply as the cargo approaches the capsid limit (computable from cargo_bp vs vehicle capacity), even while still under capacity (v4.0 delivery-oracle refinement)"
+    evaluator: delivery_aav_packaging
+    param: { margin_frac: 0.9 }
+    provenance: { doi: ["10.1089/hum.2010.245"], note: "AAV genome-size vs packaging-efficiency relationship" }
+    test_ref: "tests/unit/test_ws_atlas.py::test_aav_packaging_margin"
+    scope: "computable efficiency margin, directional; not a titre predictor"
   - id: delivery.immunogenicity_magnitude
     kind: scope_flag
     category: delivery

pen_stack-4.0.1/docs/oracles.md

@@ -0,0 +1,51 @@
+# The oracle mesh (v4.0, WS-O)
+
+PEN-STACK v4.0 sits **on top of** the biomolecular foundation models. `pen_stack.oracles` wraps them under one
+contract so their outputs can be composed, checked by the rule-grounded verifier, and trust-calibrated —
+without losing provenance, native uncertainty, or scope.
+
+## One contract: `OracleResult`
+
+Every adapter returns an `OracleResult`:
+
+```
+OracleResult{oracle, value, provenance{model, version, source, cache_key},
+             native_uncertainty, scope_card, in_scope, extrapolating,
+             output_kind ∈ {claim, candidate, baseline}, available, cached}
+```
+
+Three invariants are encoded in the type:
+
+1. **A generative output is a candidate, never a claim.** `output_kind="candidate"` (Evo2 generation, ESM3,
+   RFdiffusion, ProteinMPNN) → `as_claim()` **raises**. A candidate must pass writer-verification (WS-WV)
+   before any claim. (The pen-assemble lesson — 0 validatable de-novo writers — encoded in code.)
+2. **One contract for every oracle.** Provenance (model + version) and the model's *native* uncertainty are
+   always carried; every call is cache-keyed on `(oracle, model, version, inputs)` and replayable offline.
+3. **Scope is explicit.** Each result carries its scope-card id and an `extrapolating` flag; the field's
+   evidence that these models do not generalize to unseen loci is **labelled**, not hidden.
+
+## Wrapped models (scope cards in `configs/oracles/scope_cards.yaml`)
+
+| Family | Models | Output kind |
+|---|---|---|
+| `genome` | AlphaGenome (OOD-gated), Evo2 (likelihood=claim / generation=candidate), ChromBPNet·Borzoi (baseline) | claim / candidate / baseline |
+| `structure` | AlphaFold3, Boltz-2, Chai-1, Protenix + `consistency()` | claim |
+| `protein_design` | ESM3, RFdiffusion(-AA), ProteinMPNN·LigandMPNN | **candidate** |
+| `rna` | ViennaRNA (real; hard fold-legality input) | claim |
+| `energetics` | bridge off-target (MC3 gate ≥ 0.77) | claim |
+
+## Cross-oracle consistency
+
+`structure.consistency(seq)` runs the available structure predictors and combines them with `consensus()`:
+agreement is a confidence signal, and **disagreement widens the reported interval** (`native_uncertainty`
+grows with the cross-oracle spread) — v4.0 Principle 3.
+
+## Compute / offline policy
+
+Heavy backends (AF3, Evo2, ESM3, …) run on-demand (hosted API / local GPU) and are cached + version-pinned
+under `oracle_cache/` (committed for offline CI). When a backend and a cache entry are both absent, the
+adapter returns a **deferred** result (`available=False`) — it never fabricates a value. ViennaRNA and the
+bridge energetics model are real and run locally / on the VM.
+
+See `docs/writer_verification.md` (scoring/critiquing writers through the mesh), `prereg/ws_o.yaml`, and
+`pen_stack/oracles/`.

pen_stack-4.0.1/docs/writer_verification.md

@@ -0,0 +1,46 @@
+# Writer verification (v4.0, WS-WV) — the honest "better pen"
+
+PEN-STACK does **not** invent writer enzymes de-novo. The earlier `pen-assemble` direction produced **0
+validatable de-novo writers** and could not be checked computationally. v4.0 builds the honest alternative:
+**score and critique** proposed or variant writers against measured data — never assert novel function.
+
+## WV1 — variant scoring (DMS- + structure-grounded)
+
+`pen_stack.atlas.writer_verify.score_variants(variants)` returns, per variant, a **calibrated activity score**
+in [0, 1] with an interval and a scope flag:
+
+- A variant present in the **Perry-2025 ISCro4 deep mutational scan** is scored from its *measured* activity
+  Z-score → `claimable=True`, finite interval.
+- A variant **out of the DMS distribution** → `extrapolating=True`, `claimable=False` — a plausibility screen
+  only, **no activity claim**.
+
+`blind_recovery()` is the deterministic retrospective criterion: ranking the documented panel, the known
+enhancers **N322P / H50K / R278M** land on top, above the measured-worse controls. `real_dms_recovery()`
+reports the same against the full Perry DMS when it is present (on the VM). This is a *catalogue* feature that
+recovers known enhancers — **not** a blind sequence-only predictor, and stated as such.
+
+## WV2 — critique, not invention
+
+`critique_candidate(seq, ...)` takes a **generated** candidate writer (e.g. from `oracles.protein_design`,
+which only ever returns `candidate` outputs) and critiques it:
+
+| Check | Source |
+|---|---|
+| folds? | structure oracle (`oracles.structure.consistency`) — deferred → flagged, never asserted |
+| plausible active site? | retains conserved core residues (heuristic) |
+| deliverable form? | the rule-grounded verifier (delivery rules) |
+| reachable target? | the rule-grounded verifier (reachability rules) |
+
+It returns `pass`/flags + reasons, and **always** `no_claim=True`, `claimable=False`. A generated writer is
+never returned as "a working new pen." The verifier surfaces this as `Verdict.writer_critique` (a scope flag,
+never a confidence): the legality of the write *plan* and the activity *claim* for the candidate are distinct
+axes — a legal plan does not make an unverified candidate claimable.
+
+## Scope & honesty
+
+- Deep DMS exists for **few** enzymes (bridge recombinases); elsewhere WV1 is a labelled plausibility screen.
+- Generative designs are **proposals**; they are scored/critiqued, never asserted active.
+- Structure verification is deferred without an AF3/Boltz/Chai/Protenix backend or a committed cache entry —
+  the candidate is then explicitly *fold-unverified*, not assumed to fold.
+
+See `docs/oracles.md` (the mesh), `prereg/ws_wv.yaml`, and `pen_stack/atlas/writer_verify.py`.

{pen_stack-3.4.0 → pen_stack-4.0.1}/pen_stack/__init__.py

@@ -1,2 +1,2 @@
 """PEN-STACK v3.0 - open infrastructure for genome writing."""
-__version__ = "3.4.0"
+__version__ = "4.0.1"

pen_stack-4.0.1/pen_stack/atlas/writer_verify.py

@@ -0,0 +1,167 @@
+"""Writer-verification branch (v4.0, WS-WV) — the honest "better pen".
+
+We do **not** invent writer enzymes de-novo (pen-assemble produced 0 validatable de-novo writers and could not
+be checked computationally). We **score and critique** proposed/variant writers against measured data:
+
+* **WV1 — variant scoring head.** Combine the MEASURED DMS effect (Perry-2025 ISCro4 deep mutational scan,
+  via the existing `atlas.variant_propose` model) with structural plausibility (the structure oracle) into a
+  *calibrated* activity score + interval + scope flag. On held-out variants it ranks measured-better above
+  measured-worse above a baseline, and **recovers the known enhanced variants blind** (N322P / H50K / R278M).
+  It asserts **no activity** for a variant lacking measured or in-distribution support.
+* **WV2 — critique, not invention.** A generated candidate writer (from `oracles.protein_design`) is
+  *critiqued* — does it fold? plausible active site? deliverable form? reachable target? — returning
+  pass/flag + reasons; it is **never** returned as "a working new pen" (`no_claim=True`, `claimable=False`).
+
+When the Perry DMS is absent (off the VM) a small **frozen documented panel** keeps WV1 exercisable and the
+blind-recovery criterion deterministic — labelled as a retrospective panel, never presented as a blind
+sequence predictor.
+"""
+from __future__ import annotations
+
+import math
+from dataclasses import dataclass
+
+# A FROZEN retrospective panel of REAL values from the Perry 2025 ISCro4 deep mutational scan
+# (science.adz0276 Table S3, sheet "L2FC_Relative_Z-Scores", column Z_Score_wrt_WT) — the three top-ranked
+# enhancers, three near-neutral variants, and the three most-deleterious variants, copied verbatim from the
+# measured table. Used only when the full Perry DMS is absent (offline/CI); never fabricated.
+_FROZEN_DMS_Z = {
+    "N322P": 0.754, "H50K": 0.742, "R278M": 0.709,      # top-3 enhancers (measured Z, ranks 1-3)
+    "V21R": -0.000, "S312Q": -0.001, "G286T": -0.001,   # near-neutral (|Z| ~ 0)
+    "R132E": -5.400, "R137E": -5.115, "R195D": -4.984,  # most-deleterious (measured worst by Z)
+}
+KNOWN_ISCRO4_ENHANCERS = ["N322P", "H50K", "R278M"]
+_WORSE_CONTROLS = ["R132E", "R137E", "R195D"]            # measured-worst variants (Perry Table S3)
+# the catalytic residues a plausible ISCro4-family candidate must retain (Perry Table S3, sheet
+# "Residue Groups", Catalytic_Residues == "Catalytic"): D11, E60, D102, D105, S241 (1-based) -> 0-based below
+_CORE_RESIDUES = {10: "D", 59: "E", 101: "D", 104: "D", 240: "S"}
+
+
+def _sigmoid(x: float) -> float:
+    return 1.0 / (1.0 + math.exp(-x))
+
+
+@dataclass
+class VariantScore:
+    variant: str
+    effect: float | None          # measured/predicted activity effect (higher = better)
+    score: float | None           # calibrated activity score in [0,1]
+    interval: tuple[float, float] | None
+    in_dms: bool                  # backed by measured DMS
+    extrapolating: bool           # out of DMS distribution
+    claimable: bool               # may an activity claim be made? (only with measured/in-dist support)
+    note: str
+
+
+def _dms_lookup():
+    """Return (model_name, {variant: z}) — the real Perry DMS if present, else the frozen panel (labelled)."""
+    try:
+        from pen_stack.atlas.variant_propose import DMSVariantEffectModel
+        m = DMSVariantEffectModel()
+        return m.name, m._z
+    except Exception:  # noqa: BLE001 - Perry tables absent off the VM
+        return "frozen_retrospective_panel", dict(_FROZEN_DMS_Z)
+
+
+def score_variants(variants: list[str], structure_uncertainty: float | None = None) -> list[VariantScore]:
+    """Calibrated activity score per variant. Measured (DMS) variants are claimable with a tight interval;
+    unmeasured variants are flagged extrapolating and are NOT claimable (no activity asserted)."""
+    model_name, z = _dms_lookup()
+    out: list[VariantScore] = []
+    # interval half-width: wider when the structure oracle is uncertain/deferred (no structural support)
+    su = 0.25 if structure_uncertainty is None else float(structure_uncertainty)
+    half = 0.10 + 0.5 * su
+    for v in variants:
+        if v in z:
+            eff = float(z[v])
+            score = _sigmoid(eff)                              # monotone map of the measured Z-score
+            lo, hi = max(0.0, score - half), min(1.0, score + half)
+            out.append(VariantScore(v, eff, round(score, 3), (round(lo, 3), round(hi, 3)),
+                                    in_dms=True, extrapolating=False, claimable=True,
+                                    note=f"measured DMS effect ({model_name})"))
+        else:
+            out.append(VariantScore(v, None, None, None, in_dms=False, extrapolating=True, claimable=False,
+                                    note="OUT of DMS distribution — plausibility screen only, NO activity "
+                                         "claim (v4.0 WS-WV)"))
+    return out
+
+
+def blind_recovery(top_k: int = 5) -> dict:
+    """Deterministic blind-validation over the FROZEN documented panel (the same published enhancers,
+    measured-neutral, and measured-worse controls): the known enhancers must rank on top, above the
+    measured-worse variants. This is a retrospective catalogue criterion, NOT a blind sequence predictor —
+    labelled as such. (The full-Perry-DMS recovery is reported separately by `real_dms_recovery`.)"""
+    scores = {v: _sigmoid(zz) for v, zz in _FROZEN_DMS_Z.items()}
+    ranked = sorted(scores, key=scores.get, reverse=True)
+    top = ranked[:top_k]
+    recovered = {e: (e in top) for e in KNOWN_ISCRO4_ENHANCERS}
+    enh_min = min(scores[e] for e in KNOWN_ISCRO4_ENHANCERS)
+    worse_max = max(scores[w] for w in _WORSE_CONTROLS)
+    return {"available": True, "model": "frozen_retrospective_panel", "n": len(_FROZEN_DMS_Z), "top_k": top_k,
+            "top": top, "recovered": recovered, "all_enhancers_recovered": all(recovered.values()),
+            "enhancers_outrank_worse": bool(enh_min > worse_max),
+            "note": "recovers KNOWN enhancers (N322P/H50K/R278M) above measured-worse controls — a "
+                    "retrospective catalogue criterion, NOT a blind sequence-only predictor."}
+
+
+def real_dms_recovery(top: int = 20) -> dict:
+    """Recovery against the FULL Perry-2025 ISCro4 DMS via the existing validated harness; deferred (and the
+    frozen panel stands in) when the Perry tables are absent off the VM."""
+    try:
+        from pen_stack.atlas.variant_propose import iscro4_dms_recovery
+        rep = iscro4_dms_recovery(top=top)
+        if rep.get("available", True) is not False and "recovered" in rep:
+            return {"available": True, **rep}
+    except Exception:  # noqa: BLE001
+        pass
+    return {"available": False, "note": "Perry 2025 DMS absent (runs on the VM); see blind_recovery (frozen)"}
+
+
+def critique_candidate(candidate_seq: str, writer_family: str = "bridge_IS110",
+                       delivery_vehicle: str | None = None, no_integration: bool = False,
+                       site_seq: str | None = None) -> dict:
+    """Critique a GENERATED candidate writer (WV2) — folds? plausible active site? deliverable? reachable? —
+    returning pass/flag + reasons. NEVER returns 'a working new pen' (no_claim=True, claimable=False)."""
+    flags, reasons = [], []
+
+    # 1. structural plausibility (structure oracle; deferred without a backend -> flagged, not asserted)
+    from pen_stack.oracles import structure
+    st = structure.consistency(candidate_seq)
+    fold_ok = bool(st.available and st.value is not None and float(st.value) >= 0.7)
+    if not st.available:
+        flags.append("fold_unverified")
+        reasons.append("structure oracle deferred (no AF3/Boltz/Chai/Protenix backend or cache) — fold not "
+                       "verified; candidate cannot be claimed to fold")
+
+    # 2. active-site plausibility (heuristic: retains conserved core residues)
+    active_site_ok = all(0 <= i < len(candidate_seq) and candidate_seq[i] == aa
+                         for i, aa in _CORE_RESIDUES.items())
+    if not active_site_ok:
+        flags.append("active_site_implausible")
+        reasons.append("candidate does not retain the conserved core residues expected of the writer family")
+
+    # 3. deliverability + 4. reachability — reuse the rule-grounded verifier where inputs are present
+    deliverable = reachable = None
+    if delivery_vehicle or site_seq:
+        from pen_stack.verify import verify
+        v = verify(dict(write_type="insertion", writer_family=writer_family, site_seq=site_seq,
+                        delivery_vehicle=delivery_vehicle, no_integration=no_integration))
+        named = [x["rule_id"] for x in v.violations]
+        deliverable = not any(r.startswith("delivery.") for r in named)
+        reachable = not any(r.startswith("reachability.") for r in named)
+        if not deliverable:
+            flags.append("not_deliverable")
+            reasons.append("; ".join(x["reason"] for x in v.violations if x["rule_id"].startswith("delivery.")))
+        if not reachable:
+            flags.append("not_reachable")
+            reasons.append("; ".join(x["reason"] for x in v.violations if x["rule_id"].startswith("reachability.")))
+
+    passed = active_site_ok and fold_ok and (deliverable is not False) and (reachable is not False)
+    return {
+        "writer_family": writer_family, "fold_ok": fold_ok, "active_site_ok": active_site_ok,
+        "deliverable": deliverable, "reachable": reachable, "pass": bool(passed), "flags": flags,
+        "reasons": reasons,
+        "no_claim": True, "claimable": False,             # WV2 NEVER asserts a generated writer works
+        "note": "critique only — a generated writer is scored/critiqued against structure + rules, never "
+                "returned as a working new pen (v4.0 Principle 1 + the pen-assemble lesson).",
+    }

pen_stack-4.0.1/pen_stack/oracles/__init__.py

@@ -0,0 +1,65 @@
+"""The L1 oracle mesh (v4.0, WS-O) — one contract over the biomolecular foundation models.
+
+`pen_stack.oracles` wraps AlphaGenome / Evo2 / structure predictors (AF3, Boltz-2, Chai-1, Protenix) /
+protein-design models (ESM3, RFdiffusion, ProteinMPNN) / ViennaRNA / the bridge energetics model under a
+single `OracleResult` contract (value + provenance + native uncertainty + scope card). Heavy backends run
+on-demand (hosted API / local GPU) and are cached + version-pinned; when a backend is absent the adapter
+returns a *deferred* result (available=False) — the core stays runnable offline from cache.
+"""
+from __future__ import annotations
+
+from typing import Any
+
+from pen_stack.oracles.cache import cache_get, cache_key, cache_put, scope_card
+from pen_stack.oracles.schema import OracleResult, Provenance
+
+__all__ = ["OracleResult", "Provenance", "build_result", "consensus", "assert_claimable",
+           "cache_get", "cache_key", "cache_put", "scope_card"]
+
+
+def build_result(oracle: str, model: str, *, value: Any = None, inputs: dict | None = None,
+                 native_uncertainty: float | None = None, available: bool = True, cached: bool = False,
+                 source: str = "adapter", extrapolating: bool = False, in_scope: bool = True,
+                 output_kind: str | None = None, note: str | None = None) -> OracleResult:
+    """Assemble an `OracleResult`, filling version / output_kind / scope from the model's scope card.
+
+    `output_kind` defaults to the scope card's, but may be overridden per call (e.g. an Evo2 *likelihood*
+    score is a claim-scope scalar, while an Evo2 *generated sequence* is a candidate)."""
+    card = scope_card(model) or {}
+    key = cache_key(oracle, model, card.get("version", "0"), inputs or {})
+    return OracleResult(
+        oracle=oracle, value=value,
+        provenance=Provenance(model=model, version=card.get("version", "0"), source=source, cache_key=key),
+        native_uncertainty=native_uncertainty,
+        scope_card=model, in_scope=in_scope, extrapolating=extrapolating,
+        output_kind=output_kind or card.get("output_kind", "claim"),
+        available=available, cached=cached, note=note)
+
+
+def assert_claimable(result: OracleResult) -> OracleResult:
+    """Guard: a generative candidate cannot enter a claim path without writer-verification (Principle 1)."""
+    return result.as_claim()
+
+
+def consensus(results: list[OracleResult], oracle: str = "structure") -> OracleResult:
+    """Cross-oracle self-consistency (v4.0 Principle 3): agreement is a confidence signal, **divergence
+    widens the interval**. Combines redundant numeric oracles (e.g. AF3 / Boltz-2 / Chai-1 / Protenix); the
+    reported native_uncertainty is increased by the spread across the available oracles."""
+    avail = [r for r in results if r.available and isinstance(r.value, (int, float))]
+    if not avail:
+        return build_result(oracle, "consensus", available=False,
+                            note="no available numeric oracle to combine")
+    vals = [float(r.value) for r in avail]
+    mean = sum(vals) / len(vals)
+    spread = (max(vals) - min(vals)) if len(vals) > 1 else 0.0
+    base_unc = max((r.native_uncertainty or 0.0) for r in avail)
+    card = scope_card("boltz-2") or {}
+    return OracleResult(
+        oracle=oracle, value=round(mean, 4),
+        provenance=Provenance(model="consensus", version=card.get("version", "0"), source="adapter",
+                              extra={"members": [r.provenance.model for r in avail], "spread": round(spread, 4)}),
+        # disagreement widens the interval: native uncertainty + half the cross-oracle spread
+        native_uncertainty=round(base_unc + 0.5 * spread, 4),
+        scope_card="boltz-2", in_scope=all(r.in_scope for r in avail),
+        extrapolating=any(r.extrapolating for r in avail), output_kind="claim",
+        available=True, note=f"consensus of {len(avail)} oracles; spread {round(spread, 4)} widens the interval")