pen-stack 7.2.1__tar.gz → 7.2.3__tar.gz

{pen_stack-7.2.1 → pen_stack-7.2.3}/CHANGELOG.md

@@ -3,6 +3,38 @@
 All notable changes to PEN-STACK are documented here. This file follows
 [Keep a Changelog](https://keepachangelog.com/).
 
+## [7.2.3] - 2026-07-02 - Off-Target: professional wording (self-contained, no external-tool comparisons) + a plain-language explainer
+
+### Changed
+- **Removed all external-tool comparisons** ("like CRISPOR / CHOPCHOP", "CRISPOR-class", "CRISPOR-comparable") from
+  every off-target surface — the Off-Target page, nav, code docstrings, REST docstring, the data card, `positioning.md`,
+  the prereg (re-SHA-locked), paper-4 prereg, and the v7.2.0 CHANGELOG entry. The capability is now described on its
+  own terms: a genome-wide off-target *finder* — given a guide/target it enumerates and ranks the off-target sites
+  itself. No functional change.
+- **Off-Target page now explains off-target effects in plain language**: a new top card covers *what an off-target
+  effect is*, *what you enter* (per writer class), and *what you get back* (ranked genome-wide sites with
+  coordinates + a risk band + the confirming assay), plus a "how it works" line (scan GRCh38 within the mismatch
+  tolerance → score → rank; heavy scan on the VM, replayed or honest-abstain). The per-metric ScoreGuide was
+  reworded to match.
+
+## [7.2.2] - 2026-07-02 - Integrase: precise capability framing (the negative is about a method, not the biology)
+
+A framing/presentation refinement of the v7.2.1 sealed-negative result — no science changes, the number is
+unchanged and the negative stands.
+
+### Changed
+- **Precise claim, everywhere:** what the sealed PhiC31 benchmark showed is that a specific *method* (attP
+  sequence-similarity scanning) is insufficient — **NOT** that pseudosites are unpredictable. The signal is real
+  (Chalberg's palindromic consensus; IntQuery's success on 410,776 cryptic attB), just not capturable by simple
+  identity. So the honest, narrow claim is "sequence similarity is insufficient," and this negative is the
+  empirical justification for a learned/DMS model (the indicated next step). Sharpened in `offtarget_integrase.py`,
+  `phic31_recall_metrics.json`, and the data card so nothing reads as "unpredictable."
+- **App presentation:** `nominate_integrase` now returns a `capability` disclosure (badge + plain microcopy). The
+  Off-Target page shows a badge reading **"mechanism-based · not predictive by sequence alone"** (not a bare red
+  flag) plus one line — *"φC31 pseudosite activity is sequence-guided but not predictable from similarity alone
+  (sealed benchmark, negative); confirm empirically"* — separating the grounded facts (verified att + documented
+  pseudo-attP with GenBank accessions) from the honest capability limit. A trust-building disclosure, not an alarm.
+
 ## [7.2.1] - 2026-07-02 - PhiC31 integrase: close the O-WS3/O-G2 data gap with verified data + a sealed recall benchmark (honest NEGATIVE)
 
 Closes the one genuine open item from v7.2.0 — the PhiC31 integrase data gap — with independently-verified,
@@ -35,10 +67,10 @@ open, citable sequences (no fabrication) and a **sealed recall benchmark whose r
 
 ## [7.2.0] - 2026-07-01 - PEN-OFFTGT v2: Stage E becomes a genome-wide, per-mechanism off-target FINDER (all 5 writer classes)
 
-Stage E was a candidate *scorer* — it ranked off-target sites you supplied. A real off-target tool (CRISPOR /
-CHOPCHOP) takes a guide/target and *finds* the genome-wide off-target set itself. PEN-OFFTGT v2 closes that
-enumeration gap AND applies the **correct off-target mechanism for each writer class**, each carrying a truthful
-per-mechanism validation status (O-WS0–O-WS9 of the plan).
+Stage E was a candidate *scorer* — it ranked off-target sites you supplied. An off-target *finder* instead takes
+a guide/target and enumerates the genome-wide off-target sites itself. PEN-OFFTGT v2 closes that enumeration gap
+AND applies the **correct off-target mechanism for each writer class**, each carrying a truthful per-mechanism
+validation status (O-WS0–O-WS9 of the plan).
 
 ### Per-mechanism paths (O-WS3–O-WS6)
 - **Serine integrase** (`offtarget_integrase.py`, O-WS3): a genome-wide **pseudo-attP** scan — a fixed-sequence

{pen_stack-7.2.1 → pen_stack-7.2.3}/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: 7.2.1
+version: 7.2.3
 date-released: 2026-07-02
 authors:
   - family-names: "Mahaboob Ali"

{pen_stack-7.2.1 → pen_stack-7.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pen-stack
-Version: 7.2.1
+Version: 7.2.3
 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

{pen_stack-7.2.1 → pen_stack-7.2.3}/benchmarks/offtarget/integrase/SHA256SUMS

@@ -1,4 +1,4 @@
 701efce23441ba6ec3865060aa3a749d7165d013628f3ed8010737861b306557  README.md
 888dd4ba69bbe1eccd323797adc89b242b1c8ec6208a5f9f41df839d641d10e6  harness.py
 fe23827ff34204bc2cd1f903fcd3e32b120c4521787342e6590824249573a4d4  phic31_pseudo.fasta
-2a0f22d700f0db0bafd1c98b8b8893f14e86ae9574ff789042bd3ad1c7f6a3c0  phic31_recall_metrics.json
+8ac9ee21ddab7ddc9af4d68f6eef872c30069f6dbc6ae8de89f62c981d1868cd  phic31_recall_metrics.json

{pen_stack-7.2.1 → pen_stack-7.2.3}/docs/cards/offtarget_data.md

@@ -103,7 +103,7 @@ disclosure** where it does not. No mechanism is ever presented as validated wher
 | Writer class | Mechanism | Enumeration | Status | Ground truth / benchmark |
 |---|---|---|---|---|
 | **Nuclease** (SpCas9/SaCas9/Cas12a) | mismatch-tolerant cleavage at protospacer+PAM | Cas-OFFinder genome scan | ✅ **validated** | 4-assay Off-Target-Bench (above) + **O-G1**: enumeration recovers 100% of EMX1's documented GUIDE-seq off-targets ≤5 mm |
-| **Serine integrase** (Bxb1, PhiC31) | recombination at genomic pseudo-attP | fixed-sequence att-window scan | 🔵 **mechanism-based, sealed-NEGATIVE benchmark** (v7.2.1) | att verified (Bxb1: FlyBase FBto0000359 / Ghosh 2003 `10.1016/S1097-2765(03)00444-1`; PhiC31: **PDB 9U2T/9U2S** + Groth 2000 `10.1073/pnas.090527097`); the PhiC31 human pseudo-attP set is verified (Thyagarajan 2001, **GenBank AF333429/30/31**, `10.1128/MCB.21.12.3926-3934.2001`; chr8/16/15). The v7.2.1 **SEALED recall benchmark** (`benchmarks/offtarget/integrase/`) is **NEGATIVE**: att-sequence-similarity does NOT recover the documented pseudo-attP above random background (all three at 14 mm / 30 bp = the background median). The genome-wide similarity RANKING is therefore **unvalidated** — φC31 recognition needs palindromic architecture (Chalberg 2006, `10.1016/j.jmb.2005.11.098`) / a learned model (IntQuery), not raw identity. Documented pseudosites surfaced as verified known loci; the negative is reported verbatim (a stronger, more honest O-G2 than an untested "semi-validated"). |
+| **Serine integrase** (Bxb1, PhiC31) | recombination at genomic pseudo-attP | fixed-sequence att-window scan | 🔵 **mechanism-based, sealed-NEGATIVE benchmark** (v7.2.1) | att verified (Bxb1: FlyBase FBto0000359 / Ghosh 2003 `10.1016/S1097-2765(03)00444-1`; PhiC31: **PDB 9U2T/9U2S** + Groth 2000 `10.1073/pnas.090527097`); the PhiC31 human pseudo-attP set is verified (Thyagarajan 2001, **GenBank AF333429/30/31**, `10.1128/MCB.21.12.3926-3934.2001`; chr8/16/15). The v7.2.1 **SEALED recall benchmark** (`benchmarks/offtarget/integrase/`) is **NEGATIVE**: att-sequence-similarity does NOT recover the documented pseudo-attP above random background (all three at 14 mm / 30 bp = the background median). The genome-wide similarity RANKING is therefore **not predictive by sequence alone**. **Precise claim:** what failed is a specific *method* (attP sequence-similarity), NOT the existence of predictable pseudosites — the signal is real (Chalberg 2006's palindromic consensus `10.1016/j.jmb.2005.11.098`; IntQuery's 410,776 cryptic attB), just not capturable by simple identity. "Sequence similarity is insufficient," never "unpredictable." This negative is the empirical justification for a learned/DMS model (the indicated next step). Documented pseudosites surfaced as verified known loci; the negative is reported verbatim (a stronger, more honest O-G2 than an untested "semi-validated"). |
 | **Bridge** (IS110/IS621) | recombination at bridge-RNA target-loop matches | core-seeded genome scan (pysam) | 🔵 **mechanism-based, unvalidated** | **NO genome-wide unbiased CELLULAR off-target assay exists** (technology ~2024). The mismatch-tolerance RANKER is validated on the measured Perry-2025 in-vitro DMS specificity (held-out ranking AUROC 0.88), but genomic recovery is unvalidated. |
 | **CAST** (ShCAST V-K / VchCAST I-F) | guide-directed integration + guide-independent untargeted transposition | spacer scan + per-system untargeted background | 🔵 **mechanism-based, unvalidated** | untargeted-transposition rates documented per system (`data/curated/cast_systems.yaml`): ShCAST high/AT-biased (Strecker 2019 `10.1126/science.aax9181`; Science 2024 `10.1126/science.adj8543`), VchCAST >95–99% on-target (Klompe 2019 `10.1038/s41586-019-1323-z`; Vo 2021 `10.1038/s41587-020-00745-y`). No genome-wide cellular assay for human-cell CAST. |
 | **PASTE / (ee)PASSIGE** | Cas9-nickase off-target + integrase pseudo-attP | compose(nuclease, integrase) | composite (✅ nickase + 🟡 integrase) | inherits the nuclease benchmark for the nickase and the integrase status for the installed att; recommends BOTH a nuclease assay AND an integrase assay |
@@ -113,5 +113,5 @@ data-gap disclosure (integrase-PhiC31, bridge, CAST) — never a fabricated metr
 **hard-locked** to 🔵 unvalidated in code.
 
 **Honest limits (v7.2):** enumeration recall depends on the mismatch tolerance (≤5 mm nuclease, ≤8 mm integrase
-att window); very divergent off-targets can be missed (a limitation shared with CRISPOR). DNA/RNA bulges are not
+att window); very divergent off-targets can be missed (a limitation inherent to any mismatch-bounded genome search). DNA/RNA bulges are not
 enumerated in v2.0 (substitutions only). The engine nominates and ranks; it does not clear a design.

{pen_stack-7.2.1 → pen_stack-7.2.3}/docs/positioning.md

@@ -2,8 +2,8 @@
 
 ## The gap
 
-Genome **editing** (Cas9, base, and prime editors) is mature and well-benchmarked: CRISPOR, GUIDE-seq,
-DeepCRISPR, off-target predictors, and many guide-design tools all answer *how to change a base in place*.
+Genome **editing** (Cas9, base, and prime editors) is mature and well-benchmarked: established guide-design
+tools, off-target predictors, and unbiased assays (GUIDE-seq) all answer *how to change a base in place*.
 Genome **writing** - installing new information: inserting genes, flipping or excising kilobases, placing
 programmable landing pads with serine integrases and RNA-guided bridge recombinases - is the harder, more
 clinically transformative modality, and it has **no canonical reference layer and no benchmark**. Each lab
@@ -23,7 +23,7 @@ benchmark for the writing side. It is deliberately complementary, not competitiv
 | Question | how to change a base in place | where to write, what writer, how to design, what risk |
 | Unit | a guide / edit at a target | a site x writer x cargo x delivery plan |
 | Tasks | on-target efficiency, off-target cleavage | site selection, writer selection, within-locus, off-target, intent, no-fabrication |
-| Off-target | Cas9 cleavage (CRISPOR-class) | bridge-recombinase insertion (the unoccupied gap) |
+| Off-target | Cas9 cleavage (mature sequence-similarity predictors) | bridge-recombinase insertion (the unoccupied gap) |
 | Agents | rarely scored | a grounded LLM-agent leaderboard with a no-fabrication hard gate |
 
 ## Design commitments (why it is trustworthy)

{pen_stack-7.2.1 → pen_stack-7.2.3}/pen_stack/__init__.py

@@ -1,2 +1,2 @@
 """PEN-STACK v3.0 - open infrastructure for genome writing."""
-__version__ = "7.2.1"
+__version__ = "7.2.3"

{pen_stack-7.2.1 → pen_stack-7.2.3}/pen_stack/server/api.py

@@ -379,8 +379,8 @@ def offtarget_endpoint(req: dict):
 
     v7.2 (finder): for a nuclease guide with NO ``candidate_sites``, enumerates the genome-wide off-target set
     over GRCh38 (Cas-OFFinder, replayed from the committed cache) and ranks it by the real CRISOT-Score +
-    mismatch-calibrated risk + chromatin annotation - the CRISPOR-like default. Supplying ``candidate_sites``
-    keeps the v6.10 score-my-candidates path. Abstains (never fabricates) for a novel guide with no VM scan."""
+    mismatch-calibrated risk + chromatin annotation - the default. Supplying ``candidate_sites`` keeps the v6.10
+    score-my-candidates path. Abstains (never fabricates) for a novel guide with no VM scan."""
     from pen_stack.wgenome.offtarget_predict import nominate_offtargets
     return nominate_offtargets(
         req.get("writer_family", ""), guide=req.get("guide"), candidate_sites=req.get("candidate_sites"),

{pen_stack-7.2.1 → pen_stack-7.2.3}/pen_stack/wgenome/offtarget_enumerate.py

@@ -1,9 +1,9 @@
-"""Genome-wide off-target ENUMERATION — the CRISPOR/CHOPCHOP-like search step (PEN-OFFTGT v2, O-WS1).
+"""Genome-wide off-target ENUMERATION — the search step of the off-target finder (PEN-OFFTGT v2, O-WS1).
 
 Stage E used to only SCORE candidate sites a caller supplied. This module adds the missing half: given a guide +
 enzyme, ENUMERATE every genomic site within the mismatch tolerance across GRCh38 — returning coordinates, strand,
-matched sequence, and mismatch count — so the downstream scorer ranks a genome-wide candidate set (what a real
-off-target tool does), not a hand-typed one.
+matched sequence, and mismatch count — so the downstream scorer ranks a genome-wide candidate set, not a
+hand-typed one (the caller supplies a guide and gets the off-target sites back).
 
 **Where it runs.** A full GRCh38 scan is heavy, so enumeration executes ONLY where Cas-OFFinder + the genome are
 present — the VM, via the `casoffinder:tools` Docker image (Bae, Park & Kim, Bioinformatics 2014,

{pen_stack-7.2.1 → pen_stack-7.2.3}/pen_stack/wgenome/offtarget_integrase.py

@@ -24,6 +24,22 @@ _ALIAS = {"bxb1": "Bxb1", "serine_integrase": "Bxb1", "pe_integrase": "Bxb1",
           "phic31": "PhiC31", "phic31_integrase": "PhiC31"}
 _CONFIRM = "Cryptic-seq / HIDE-seq (Tome Biosciences, 2024 preprint)"
 
+# Capability disclosure (v7.2.2): be PRECISE about what the sealed benchmark showed. What failed is a specific
+# METHOD (attP sequence-similarity scanning), NOT the existence of predictable pseudosites — the signal is real
+# (Chalberg's ~30-bp palindromic consensus; IntQuery's success on 410,776 cryptic attB), just not capturable by
+# simple similarity. The honest claim is narrow: "sequence similarity is insufficient", never "unpredictable".
+_CAPABILITY = {
+    "badge": "mechanism-based · not predictive by sequence alone",
+    "microcopy": ("phiC31 pseudosite activity is sequence-guided but NOT predictable from similarity alone "
+                  "(sealed benchmark, negative); confirm empirically (Cryptic-seq / HIDE-seq)."),
+    "what_failed": ("attP sequence-SIMILARITY scanning (a specific method) — NOT the existence of predictable "
+                    "pseudosites."),
+    "signal_is_real": ("the pseudosite signal IS real — Chalberg's ~30-bp palindromic consensus and IntQuery's "
+                       "success on 410,776 cryptic attB show it — just not capturable by simple sequence identity."),
+    "indicated_next_step": ("a learned / DMS model (as IntQuery uses); this sealed negative is the empirical "
+                            "justification for it, not evidence that pseudosites are unpredictable."),
+}
+
 
 @lru_cache(maxsize=1)
 def integrase_att() -> dict:
@@ -69,15 +85,17 @@ def nominate_integrase(integrase: str = "Bxb1", top: int = 20) -> dict:
                 "att_doi": rec["doi"], "pdb": rec.get("pdb"),
                 "documented_pseudo_attP": rec.get("pseudo_attP_positive_set"),
                 "documented_source": rec.get("positive_set_source"), "documented_doi": rec.get("positive_set_doi"),
-                "sealed_recall_benchmark": bench, "similarity_ranking_validated": False,
+                "sealed_recall_benchmark": bench, "similarity_ranking_validated": False, "capability": _CAPABILITY,
                 "specificity_note": rec.get("specificity_note"),
                 "method": ("PhiC31 documented human pseudo-attP (verified GenBank AF333429/30/31, Thyagarajan 2001) "
-                           "are surfaced as known off-target loci; the att-similarity genome scan is mechanism-"
-                           "based and its SEALED recall benchmark is NEGATIVE (does not recover them above "
-                           "background) -> the genome-wide ranking is unvalidated (reported verbatim)."),
+                           "are surfaced as verified known off-target loci. Separately, the att-SIMILARITY genome "
+                           "scan is a mechanism-based method whose sealed recall benchmark is NEGATIVE: sequence "
+                           "similarity is INSUFFICIENT to rank these sites (the signal is real but needs a learned "
+                           "model, not raw identity). Reported verbatim."),
                 "confirm_assay": _CONFIRM,
-                "honesty": "verified documented pseudosites + an honest NEGATIVE recall benchmark; NOT a validated "
-                           "predictor and NOT a clearance.",
+                "honesty": "verified documented pseudosites (grounded facts) PLUS an explicit capability disclosure: "
+                           "sequence-similarity scanning does not reliably rank them; NOT a clearance. Confirm "
+                           "empirically.",
                 "nomination_is_not_clearance": True}
 
     # Bxb1 (and any integrase with a committed genome-scan cache): genome-wide similarity candidates
@@ -100,12 +118,12 @@ def nominate_integrase(integrase: str = "Bxb1", top: int = 20) -> dict:
             "scan_window": window, "source": enum["source"], "att_doi": rec["doi"],
             "n_sites_genome_wide": len(noms), "nominations": noms[:top],
             "specificity_note": rec.get("specificity_note"),
-            "sealed_recall_benchmark": bench, "similarity_ranking_validated": False,
+            "sealed_recall_benchmark": bench, "similarity_ranking_validated": False, "capability": _CAPABILITY,
             "method": ("fixed-sequence Cas-OFFinder scan of the attP core window over GRCh38, scored by att-arm "
-                       "similarity. The similarity RANKING is mechanism-based and UNVALIDATED: the sealed PhiC31 "
-                       "recall benchmark is NEGATIVE (att-similarity does not recover documented pseudo-attP "
-                       "above background)."),
+                       "similarity. The similarity RANKING is mechanism-based and NOT predictive by sequence alone: "
+                       "the sealed PhiC31 recall benchmark is NEGATIVE (sequence similarity is insufficient to "
+                       "rank pseudo-attP; the signal is real but needs a learned model, per IntQuery)."),
             "confirm_assay": _CONFIRM,
-            "honesty": "genome-wide pseudo-attP candidates by att similarity; the ranking is unvalidated "
-                       "(sealed-negative benchmark) and NOT a clearance.",
+            "honesty": "genome-wide pseudo-attP candidates by att similarity; sequence similarity is insufficient "
+                       "to rank them (sealed-negative benchmark) -> mechanism-based, confirm empirically. NOT a clearance.",
             "nomination_is_not_clearance": True}

{pen_stack-7.2.1 → pen_stack-7.2.3}/pen_stack/wgenome/offtarget_nuclease.py

@@ -1,7 +1,7 @@
 """Nuclease off-target FINDER (PEN-OFFTGT v2, O-WS2): enumerate -> CRISOT -> risk -> chromatin.
 
 Chains the genome-wide enumeration (O-WS1) into the existing, validated nuclease scorer, so a GUIDE returns a
-genome-wide ranked off-target set (what CRISPOR/CHOPCHOP do) rather than a hand-supplied one. The scoring is
+genome-wide ranked off-target set rather than a hand-supplied one. The scoring is
 UNCHANGED from v6.10 — the real CRISOT-Score, the mismatch-calibrated risk band, and the chromatin annotation
 (validated, not a re-ranker). v2 only adds the enumeration front end. Status: **validated** (CRISOT beats
 homology on four unbiased assays; enumeration reproduces the documented off-target set, gate O-G1).

{pen_stack-7.2.1 → pen_stack-7.2.3}/pen_stack/wgenome/offtarget_predict.py

@@ -222,8 +222,8 @@ def nominate_offtargets(writer_family: str, guide: str | None = None, candidate_
     ``cell_type`` enable the REAL Stage B chromatin-accessibility modifier for nucleases.
 
     v7.2 (O-WS2): for a nuclease guide WITHOUT supplied ``candidate_sites``, this runs the genome-wide FINDER
-    (enumerate GRCh38 -> CRISOT -> risk -> chromatin), the CRISPOR-like default. Supplying ``candidate_sites``
-    keeps the v6.10 score-my-candidates path (backward compatible)."""
+    (enumerate GRCh38 -> CRISOT -> risk -> chromatin) by default. Supplying ``candidate_sites`` keeps the v6.10
+    score-my-candidates path (backward compatible)."""
     from pen_stack.wgenome.offtarget_assay import recommend_assay
     fam = (writer_family or "").lower()
     assay_rec = recommend_assay(writer_family)

{pen_stack-7.2.1 → pen_stack-7.2.3}/pen_stack.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pen-stack
-Version: 7.2.1
+Version: 7.2.3
 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

pen_stack-7.2.3/prereg/SHA256_LOCK_ws_offtarget2.json

@@ -0,0 +1,9 @@
+{
+  "cycle": "v7.2",
+  "workstream": "PEN-OFFTGT-v2",
+  "prepared": "2026-07-01",
+  "amended": "2026-07-02 (v7.2.1 O-G2 sealed result; v7.2.2 professional wording pass)",
+  "sha256": {
+    "prereg/ws_offtarget2.yaml": "208a000c11cf697c73e1e92f785fed5b72cd469d5f7c54c81975c1489e37e054"
+  }
+}

{pen_stack-7.2.1 → pen_stack-7.2.3}/prereg/paper4.yaml

@@ -1,5 +1,5 @@
 # PEN-STACK v3.0 - Paper 4 (Bridge-recombinase off-target engine) pre-registered success criteria.
-# SHA-locked. Paper 4 is the beachhead: "a CRISPOR for bridge recombinases."
+# Paper 4 is the beachhead: a genome-wide off-target screening tool for RNA-guided bridge recombinases.
 paper: 4
 title: "A screening tool to nominate genome-wide off-target locations for RNA-guided bridge recombinases"
 prepared: "2026-06-02"

{pen_stack-7.2.1 → pen_stack-7.2.3}/prereg/ws_offtarget2.yaml

@@ -2,8 +2,8 @@
 # Locks the mechanism taxonomy, per-mechanism enumeration parameters, the status-label rules, and the acceptance
 # gates BEFORE building, per the standing "pre-register + SHA-lock" discipline. Lands as v7.2.0 (Stage E).
 #
-# The load-bearing change: Stage E becomes a genome-wide FINDER (enumerates candidates itself, like CRISPOR/
-# CHOPCHOP) instead of only a SCORER of supplied candidates. Enumeration is heavy and runs on the VM over GRCh38;
+# The load-bearing change: Stage E becomes a genome-wide FINDER (given a guide, it enumerates the candidate sites
+# itself) instead of only a SCORER of supplied candidates. Enumeration is heavy and runs on the VM over GRCh38;
 # the live app replays a committed coordinate cache or abstains (the enumerated coordinates are public-genome facts).
 
 version: "1.0"
@@ -16,7 +16,7 @@ invariants:
   - nomination_is_not_clearance   # every candidate ships with the confirming empirical assay
   - no_fabrication                # never invent a site or a validated status; abstain / flag where no data
   - per_mechanism_honesty         # each writer class returns its own mechanism, algorithm, and status label
-  - genome_wide_by_default        # the engine enumerates candidates itself (the CRISPOR-like property)
+  - genome_wide_by_default        # the engine enumerates candidates itself (a guide in, the off-target sites out)
 
 # The enumerated off-target coordinates are FACTS derivable from the public GRCh38 assembly with a public tool
 # (Cas-OFFinder). Unlike the CRISOT weights (CC-BY-NC, VM-only), the coordinates carry no license restriction and
@@ -37,7 +37,7 @@ mechanisms:
     off_target_mechanism: mismatch/bulge-tolerant cleavage at protospacer + PAM
     pam: {SpCas9: NGG, SaCas9: NNGRRT, AsCas12a: TTTV, LbCas12a: TTTV}
     protospacer_len: {SpCas9: 20, SaCas9: 21, AsCas12a: 23, LbCas12a: 23}
-    max_mismatches: 5               # enumeration tolerance (recall/-precision trade; CRISPOR-comparable)
+    max_mismatches: 5               # enumeration tolerance (a recall / precision trade-off)
     max_dna_bulge: 0                # v2.0 ships substitutions only; bulges are a follow-up
     max_rna_bulge: 0
     scorer: crisot                  # existing real CRISOT-Score (VM, CC-BY-NC, derived scores cached)
@@ -78,7 +78,7 @@ mechanisms:
 acceptance_gates:
   O-G1:
     workstream: O-WS1
-    claim: enumeration reproduces a documented Cas9 off-target set (behaves like CRISPOR's search step)
+    claim: enumeration reproduces a documented Cas9 off-target set (the finder's genome-wide search step)
     metric: recall of the EMX1 documented/validated GUIDE-seq off-targets among the enumerated candidates
     floor: 0.90                     # >=90% of the documented validated off-targets are enumerated at k<=5
     guide: EMX1 (GAGTCCGAGCAGAAGAAGAA, NGG)

{pen_stack-7.2.1 → pen_stack-7.2.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "pen-stack"
-version = "7.2.1"
+version = "7.2.3"
 description = "Open infrastructure for genome writing: the Writable Genome atlas, the Writer Atlas, and the Write Planner."
 readme = "README.md"
 requires-python = ">=3.11"

pen_stack-7.2.1/prereg/SHA256_LOCK_ws_offtarget2.json

@@ -1,9 +0,0 @@
-{
-  "cycle": "v7.2",
-  "workstream": "PEN-OFFTGT-v2",
-  "prepared": "2026-07-01",
-  "amended": "2026-07-02 (v7.2.1: O-G2 sealed PhiC31 recall benchmark, result NEGATIVE)",
-  "sha256": {
-    "prereg/ws_offtarget2.yaml": "6cf66e9cd972e638366e585072464287c7bfacca9b9575ac91a6a9317a7e22df"
-  }
-}