dsh-molbio-tools 0.19.0 → 0.20.0

package/benchmark/README.md

@@ -76,11 +76,11 @@ Measured, and printed so the answer can be audited:
 
 ```
 $ touch crispr.mjs && node benchmark/run.mjs --list --changed
-1 of 56 task(s) selected — --tools molbio_grna_design --changed
+1 of 61 task(s) selected — --tools molbio_grna_design --changed
   (from --changed; 2 module(s) affected: crispr.mjs, index.mjs)
 
 $ touch lib.mjs && node benchmark/run.mjs --list --changed
-55 of 56 task(s) selected — --changed (27 module(s) affected: align.mjs, cloning.mjs, …)
+55 of 61 task(s) selected — --changed (27 module(s) affected: align.mjs, cloning.mjs, …)
 ```
 
 `lib.mjs` selecting nearly everything is the **correct** answer, not a failure to
@@ -102,7 +102,7 @@ you changed, so its task runs even if it is full-only.
 | `--tools a,b,c` / `primer_*` | the union | you changed a few |
 | `--changed --tools X` | the INTERSECTION | "of what I changed, test X" |
 | (default) | **core** — 14 tasks | a broad smoke of the catalog |
-| `--tier full` | all 56, 58/58 tools | before a release |
+| `--tier full` | all 61, 60/60 tools | before a release |
 
 The two `--tools` operators are stated because confusing them is the difference
 between testing what you changed and testing **less** than that: alone it unions,
@@ -125,7 +125,7 @@ widens to the full suite, with the reason printed.
 | a module **only** imported by other modules (`svgio.mjs`) | partial — its importers' tasks |
 | docs, changelog, `test/`, `benchmark/`, config | **`none`: nothing to run** |
 
-`none` is deliberately NOT `full`. A documentation commit that triggers 56 model
+`none` is deliberately NOT `full`. A documentation commit that triggers 61 model
 calls is how a `--changed` default gets switched off and the full suite goes back
 to running every time — the exact behaviour this option exists to prevent. It is
 only reachable with an explicit `--changed`, so everything else still fails safe,
@@ -134,7 +134,7 @@ and `test/benchmark-changed.mjs` asserts both directions.
 ### The part that still runs in full, on purpose
 
 A model run with `--changed` or `--tools` first executes the **whole** offline
-gate: all 56 tasks' `where: "tool"` expectations are recomputed against the shipped
+gate: all 61 tasks' `where: "tool"` expectations are recomputed against the shipped
 tools (no model, a few hundred milliseconds). That is what makes partial runs safe
 — the expected values are re-proven for the tasks you skipped, so "skipped because
 unchanged" is a checked claim, not a remembered one.
@@ -168,7 +168,7 @@ asserting whatever the tool happened to emit.
 
 Two guards make "someone forgot" fail loudly instead of quietly:
 
-- `test/benchmark-coverage.mjs` asserts **58/58 registered tools are covered by
+- `test/benchmark-coverage.mjs` asserts **60/60 registered tools are covered by
   at least one task**. Adding a tool without a task is a red build. It also
   asserts the fixtures' premises (the Golden Gate backbone is BsaI-free, the
   mutagenesis template still admits a primer pair), that a `covers` entry names a
@@ -209,7 +209,7 @@ recorded tool text, and it says so instead of silently scoring empty text).
 
 ## The headless profile (`benchmark/profile.mjs`)
 
-The model-facing way to get the 58 tools is the **Molecular Biology Lab** agent
+The model-facing way to get the 60 tools is the **Molecular Biology Lab** agent
 preset. The benchmark cannot use it, and the reason is structural rather than
 configurational — `dsh-headless/lib/index.js` refuses an adoption whose session
 records a preset:
@@ -368,6 +368,25 @@ These are recorded rather than scored, because each one is a question about the
 plugin or the harness, not about the model. The first is a **measured plugin
 defect**.
 
+### `molbio_extract_region` advertised `sequence`, but the resolver only read `vector`
+
+The 0.20.0 full run (61 tasks) failed `region-extract` on **arguments**: the model
+called `molbio_extract_region` with the parameter the schema advertises
+(`sequence`), and the tool answered "provide either vector (a sequence) or
+vector_path". The schema/reality split was the defect: `resolveVector` read only
+`args.vector`, so `sequence` was silently ignored. **Fixed in 0.20.0** by
+accepting every declared inline alias (`vector`/`sequence`/`reference`), with a
+smoke regression that calls the tool through the advertised `sequence` argument.
+
+The same run also showed four **assertion** failures rather than product defects:
+`plasmid-digest` (the model wrote `2,840`, the assertion `2840`),
+`protein-props` (`3,935` vs `3935`), `codon-optimize` (the model said
+`EcoRI-free`, the assertion demanded `no EcoRI`), and `chart-volcano` (the task
+asked for `de.svg`, the assertion pinned `volcano.svg`). `--replay` turns all four
+to PASS; the region-extract live rerun was blocked by the DSH sandbox approval
+for headless subprocesses, so the last measured score is **60/61**, with the
+underlying bug fixed and pinned by smoke.
+
 ### `molbio_design_primers` prints the forward/reverse roles reversed
 
 **Confirmed by direct inspection, not by reading a model's claim.**
@@ -548,7 +567,7 @@ implementation happens to return for one set of arguments.**
   and `args_ok` both true is a *reporting* finding, not a tool bug.
 - **Multimodal hand-off.** `attach_image` is exercised as an argument at most; no
   assertion inspects a decoded PNG (that is `svgpng.mjs` and `contract.mjs`).
-- **Most of the toolset is now covered, but shallowly.** 56 tasks touch all 58
+- **Most of the toolset is now covered, but shallowly.** 61 tasks touch all 60
   tools; most tools appear in exactly one task, and several appearances are a
   single argument check rather than a full workflow. Breadth is asserted by
   `test/benchmark-coverage.mjs`; depth is not, and cannot be.

package/benchmark/sequences.mjs

@@ -165,6 +165,13 @@ export const NAMED_SEQUENCES = {
   cds: FIXTURES.CDS,
   mutagenesis_template: FIXTURES.MUTAGENESIS_TEMPLATE,
   sanger_read: FIXTURES.SANGER_READ,
+  // v0.20.0: the family-geometry fixtures. The Cas12a target is a 5' TTTV PAM
+  // followed by a 23 nt protospacer; the Cas13a target is a 28 nt spacer with a
+  // 3' H PFS. Keeping them here means the instruction and the offline grader
+  // cannot diverge.
+  cas12a_target: 'AAAATTTAACGTACGTACGTACGTACGTACGCCCCTTTCGGGGGGGGGGG',
+  cas13_target: 'AAAAACGTACGTACGTACGTACGTACGTACGTACCCCCCCCCC',
+  silent_cds: 'ATGGCTGAATTCGGTTAAGCTTAA',
   /**
    * The three FASTQ reads the `fastq-qc` task pastes in.
    *

package/benchmark/tasks/02-primers.json

@@ -12,26 +12,54 @@
       "id": "primer-qc",
       "category": "primers",
       "tier": "full",
-      "covers": ["molbio_primer_check", "molbio_primer_tm"],
+      "covers": [
+        "molbio_primer_check",
+        "molbio_primer_tm"
+      ],
       "instruction": "I have these two primers already ordered and I want to know whether they are any good together before I run them. Report the Tm of the forward primer, and flag anything structurally worrying about the pair.\n\nForward: ATTACTCCTGCTCTTCCCATAC\nReverse: CACTACTCCTCTGTACGCAC",
       "timeout_ms": 240000,
-      "expect_tools": ["molbio_primer_check"],
+      "expect_tools": [
+        "molbio_primer_check"
+      ],
       "assertions": [
-        { "where": "tool", "pattern": "primer1: 22 bp, GC 45\\.45%", "note": "the forward primer's own metrics" },
-        { "where": "tool", "pattern": "primer2: 20 bp, GC 55%" },
-        { "where": "tool", "pattern": "self-any 7 \\(≤8\\)", "note": "the reverse primer's self-complementarity, right at the Primer3 limit" },
-        { "where": "tool", "pattern": "pair \\(dimer\\): score 14", "note": "the pair's dimer score" },
-        { "where": "answer", "pattern": "53\\.9|54\\b", "note": "the forward primer's Tm, 53.97 °C" }
+        {
+          "where": "tool",
+          "pattern": "primer1: 22 bp, GC 45\\.45%",
+          "note": "the forward primer's own metrics"
+        },
+        {
+          "where": "tool",
+          "pattern": "primer2: 20 bp, GC 55%"
+        },
+        {
+          "where": "tool",
+          "pattern": "self-any 7 \\(≤8\\)",
+          "note": "the reverse primer's self-complementarity, right at the Primer3 limit"
+        },
+        {
+          "where": "tool",
+          "pattern": "pair \\(dimer\\): score 14",
+          "note": "the pair's dimer score"
+        },
+        {
+          "where": "answer",
+          "pattern": "53\\.9|54\\b",
+          "note": "the forward primer's Tm, 53.97 °C"
+        }
       ]
     },
     {
       "id": "taqman-assay",
       "category": "primers",
       "tier": "full",
-      "covers": ["molbio_design_taqman"],
+      "covers": [
+        "molbio_design_taqman"
+      ],
       "instruction": "Design a TaqMan (5' nuclease) assay on this template. I want primers plus a hydrolysis probe between them, and I want the probe to sit at least a few degrees above the primers in Tm.\n\nTemplate:\n{template:400:7}",
       "timeout_ms": 300000,
-      "expect_tools": ["molbio_design_taqman"],
+      "expect_tools": [
+        "molbio_design_taqman"
+      ],
       "assertions": [
         {
           "where": "tool",
@@ -48,21 +76,39 @@
           "pattern": "vs the hotter primer",
           "note": "the probe-vs-primer Tm margin, which the tool reports per candidate"
         },
-        { "where": "answer", "pattern": "probe|Probe" }
+        {
+          "where": "answer",
+          "pattern": "probe|Probe"
+        }
       ]
     },
     {
       "id": "intron-qpcr",
       "category": "primers",
       "tier": "full",
-      "covers": ["molbio_design_intron_primers"],
+      "covers": [
+        "molbio_design_intron_primers"
+      ],
       "instruction": "I need a qPCR assay that will NOT amplify contaminating genomic DNA. Here is the genomic sequence and the two exon coordinates — design a pair where the forward primer spans the exon-exon junction. Tell me the spliced amplicon size and how far apart the primers sit on the genome.\n\n{seq:intron_genomic}\n\nExons: 1-400 and 701-1100 (1-based).",
       "timeout_ms": 300000,
-      "expect_tools": ["molbio_design_intron_primers"],
+      "expect_tools": [
+        "molbio_design_intron_primers"
+      ],
       "assertions": [
-        { "where": "tool", "pattern": "forward primer spans an exon-exon junction", "note": "the property that makes the assay genomic-DNA-proof" },
-        { "where": "tool", "pattern": "spliced amplicon 320-406 \\(87 bp\\), genomic span 387 bp" },
-        { "where": "tool", "pattern": "junction 12\\+6 bp", "note": "bases contributed on each side of the junction" },
+        {
+          "where": "tool",
+          "pattern": "forward primer spans an exon-exon junction",
+          "note": "the property that makes the assay genomic-DNA-proof"
+        },
+        {
+          "where": "tool",
+          "pattern": "spliced amplicon 320-406 \\(87 bp\\), genomic span 387 bp"
+        },
+        {
+          "where": "tool",
+          "pattern": "junction 12\\+6 bp",
+          "note": "bases contributed on each side of the junction"
+        },
         {
           "where": "answer",
           "pattern": "splic|junction|genomic|intron|87|387",
@@ -74,12 +120,19 @@
       "id": "multiplex-panel",
       "category": "primers",
       "tier": "full",
-      "covers": ["molbio_multiplex_check"],
-      "instruction": "I want to run these two primer pairs in the same multiplex tube. The first pair is on target A (template below), the second on target B. Check them against each other and tell me whether they interfere — dimers, unintended annealing, or bands I will not be able to tell apart.\n\nTarget A template:\n{template:400:7}\n\nPair A: forward ATTACTCCTGCTCTTCCCATAC (Tm 58.4), reverse CACTACTCCTCTGTACGCAC (Tm 58.9)\nPair B: forward GGGTTTCCCAAATTTGGGCCCA, reverse TTTGGGCCCAAATTTGGGAAAC\nTemplate B:\n{template:400:8}",
+      "covers": [
+        "molbio_multiplex_check"
+      ],
+      "instruction": "I want to run these two primer pairs in the same multiplex tube. The first pair is on target A (template below), the second on target B. Check them against each other and tell me whether they interfere — dimers, unintended annealing, or bands I will not be able to tell apart. Also give me the recommended annealing temperature for the panel and a starting primer concentration for each pair.\n\nTarget A template:\n{template:400:7}\n\nPair A: forward ATTACTCCTGCTCTTCCCATAC (Tm 58.4), reverse CACTACTCCTCTGTACGCAC (Tm 58.9)\nPair B: forward GGGTTTCCCAAATTTGGGCCCA, reverse TTTGGGCCCAAATTTGGGAAAC\nTemplate B:\n{template:400:8}",
       "timeout_ms": 300000,
-      "expect_tools": ["molbio_multiplex_check"],
+      "expect_tools": [
+        "molbio_multiplex_check"
+      ],
       "assertions": [
-        { "where": "tool", "pattern": "multiplex panel: 2 target\\(s\\) on 2 template\\(s\\), 4 primer\\(s\\)" },
+        {
+          "where": "tool",
+          "pattern": "multiplex panel: 2 target\\(s\\) on 2 template\\(s\\), 4 primer\\(s\\)"
+        },
         {
           "where": "tool",
           "pattern": "primer-primer interactions above the dimer threshold: [0-9]+",
@@ -94,6 +147,66 @@
           "where": "answer",
           "pattern": "dimer|interfere|conflict|anneal",
           "note": "the panel verdict. Note the measured behaviour this task must tolerate: a model that FIRST checks each pair on its own template often finds — correctly — that a pair as written does not amplify at all, and answers that instead. That is a better answer than 'no dimer conflict', not a worse one."
+        },
+        {
+          "where": "tool",
+          "pattern": "annealing temperature: [0-9.]+ °C \\(range [0-9.]+-[0-9.]+ °C\\)",
+          "note": "the panel-level annealing recommendation (lowest primer Tm minus the offset)"
+        },
+        {
+          "where": "tool",
+          "pattern": "primer concentration plan \\(nM\\):",
+          "note": "the starting primer-concentration plan"
+        },
+        {
+          "where": "answer",
+          "pattern": "anneal|temperature|concentration",
+          "note": "the model reports the new planning outputs, not only the dimer verdict"
+        }
+      ]
+    },
+    {
+      "id": "taqman-mgb",
+      "category": "primers",
+      "tier": "full",
+      "covers": [
+        "molbio_design_taqman"
+      ],
+      "instruction": "Design a TaqMan assay with an MGB probe on this template. Use an MGB probe, write the ordering sheet to mgb-order.csv, and tell me the probe length and Tm window the designer used, the top-ranked probe, and the 5'/3' modifications on its ordering row.\n\nTemplate:\n{template:400:7}",
+      "timeout_ms": 300000,
+      "expect_tools": [
+        "molbio_design_taqman"
+      ],
+      "assertions": [
+        {
+          "where": "tool",
+          "pattern": "Probe chemistry: TaqMan MGB",
+          "note": "the MGB chemistry is selected, not silently the standard one"
+        },
+        {
+          "where": "tool",
+          "pattern": "13-20 nt, Tm 48-60",
+          "note": "the MGB length/Tm window, on the unmodified NN scale"
+        },
+        {
+          "where": "tool",
+          "pattern": "P TTGCTACGCCTGT",
+          "note": "the top-ranked MGB probe"
+        },
+        {
+          "where": "tool",
+          "pattern": "ordering CSV \\([0-9]+ oligo row\\(s\\)\\) saved to",
+          "note": "the ordering sheet was written"
+        },
+        {
+          "where": "tool",
+          "pattern": "MGB probes: the reported probe Tm is the unmodified DNA NN estimate",
+          "note": "the tool refuses to fabricate a modified Tm"
+        },
+        {
+          "where": "answer",
+          "pattern": "MGB|unmodified|modif",
+          "note": "the model explains the MGB Tm caveat or the modifications"
         }
       ]
     }

package/benchmark/tasks/03-restriction.json

@@ -12,67 +12,144 @@
       "id": "plasmid-digest",
       "category": "restriction",
       "tier": "full",
-      "covers": ["molbio_restriction_sites"],
+      "covers": [
+        "molbio_restriction_sites"
+      ],
       "instruction": "I have pUC118.dna in my working directory. Show me the restriction map for EcoRI, HindIII and PvuII on the circular plasmid: where each enzyme cuts, and what fragments each one produces. Then tell me which of the three cuts the plasmid more than once.",
       "timeout_ms": 240000,
-      "expect_tools": ["molbio_restriction_sites"],
-      "fixtures": { "pUC118.dna": "test/fixtures/pUC118.dna" },
+      "expect_tools": [
+        "molbio_restriction_sites"
+      ],
+      "fixtures": {
+        "pUC118.dna": "test/fixtures/pUC118.dna"
+      },
       "assertions": [
-        { "where": "tool", "pattern": "EcoRI \\(G\\^AATTC\\):[\\s\\S]*?cut at 927", "note": "EcoRI's single cut on pUC118" },
-        { "where": "tool", "pattern": "HindIII \\(A\\^AGCTT\\):[\\s\\S]*?cut at 876" },
-        { "where": "tool", "pattern": "PvuII \\(CAG\\^CTG\\):[\\s\\S]*?cut at 785" },
-        { "where": "tool", "pattern": "PvuII[\\s\\S]*?2 fragment\\(s\\): 2840, 322 bp", "note": "PvuII is the double cutter" },
-        { "where": "answer", "pattern": "PvuII" },
-        { "where": "answer", "pattern": "2840" },
-        { "where": "answer", "pattern": "322" }
+        {
+          "where": "tool",
+          "pattern": "EcoRI \\(G\\^AATTC\\):[\\s\\S]*?cut at 927",
+          "note": "EcoRI's single cut on pUC118"
+        },
+        {
+          "where": "tool",
+          "pattern": "HindIII \\(A\\^AGCTT\\):[\\s\\S]*?cut at 876"
+        },
+        {
+          "where": "tool",
+          "pattern": "PvuII \\(CAG\\^CTG\\):[\\s\\S]*?cut at 785"
+        },
+        {
+          "where": "tool",
+          "pattern": "PvuII[\\s\\S]*?2 fragment\\(s\\): 2840, 322 bp",
+          "note": "PvuII is the double cutter"
+        },
+        {
+          "where": "answer",
+          "pattern": "PvuII"
+        },
+        {
+          "where": "answer",
+          "pattern": "2[,.]?840",
+          "note": "the larger PvuII fragment; accepts the thousands separator a human writes"
+        },
+        {
+          "where": "answer",
+          "pattern": "322"
+        }
       ]
     },
     {
       "id": "double-digest-buffer",
       "category": "restriction",
       "tier": "core",
-      "covers": ["molbio_double_digest"],
+      "covers": [
+        "molbio_double_digest"
+      ],
       "instruction": "I have a 512 bp linear construct in the workspace as a sequence I will paste, but simpler: use the sequence below. Cut it with EcoRI and HindIII together. Do these two enzymes share a reaction buffer, and what fragment sizes should I see on the gel?\n\n{seq:construct}",
       "timeout_ms": 240000,
-      "expect_tools": ["molbio_double_digest"],
+      "expect_tools": [
+        "molbio_double_digest"
+      ],
       "assertions": [
-        { "where": "tool", "pattern": "shared buffer\\(s\\):[^\\n]*CutSmart", "note": "a shared buffer, so a single-tube digest is fine" },
+        {
+          "where": "tool",
+          "pattern": "shared buffer\\(s\\):[^\\n]*CutSmart",
+          "note": "a shared buffer, so a single-tube digest is fine"
+        },
         {
           "where": "tool",
           "pattern": "→ 206, 205, 101 bp",
           "note": "THREE fragments, not two: the two site insertions add 6 bp each, so the two sites are 206 bp apart and the digest yields three pieces. This was probed, not reasoned about — the first draft of the fixture documented '200 + 300 bp' and was wrong."
         },
-        { "where": "answer", "pattern": "206" },
-        { "where": "answer", "pattern": "205" },
-        { "where": "answer", "pattern": "101" }
+        {
+          "where": "answer",
+          "pattern": "206"
+        },
+        {
+          "where": "answer",
+          "pattern": "205"
+        },
+        {
+          "where": "answer",
+          "pattern": "101"
+        }
       ]
     },
     {
       "id": "pcr-simulation",
       "category": "pcr",
       "tier": "full",
-      "covers": ["molbio_pcr_simulate"],
+      "covers": [
+        "molbio_pcr_simulate"
+      ],
       "instruction": "Check whether this primer pair amplifies the template I am pasting, and tell me the product size and coordinates. Use the forward primer as given and the reverse primer as given — both are written 5' to 3' as I would order them.\n\nTemplate (512 bp):\n{seq:construct}\n\nForward primer: {seq:pcr_forward}\nReverse primer: {seq:pcr_reverse}",
       "timeout_ms": 240000,
-      "expect_tools": ["molbio_pcr_simulate"],
+      "expect_tools": [
+        "molbio_pcr_simulate"
+      ],
       "assertions": [
-        { "where": "tool", "pattern": "1 product\\(s\\) of 201 bp" },
-        { "where": "tool", "pattern": "201 bp: 101-120 → 282-301", "note": "the amplicon's coordinates on the top strand" },
-        { "where": "tool", "pattern": "specific", "note": "the tool's verdict word for a single clean product" },
-        { "where": "answer", "pattern": "201" },
-        { "where": "answer", "pattern": "101" },
-        { "where": "answer", "pattern": "301" }
+        {
+          "where": "tool",
+          "pattern": "1 product\\(s\\) of 201 bp"
+        },
+        {
+          "where": "tool",
+          "pattern": "201 bp: 101-120 → 282-301",
+          "note": "the amplicon's coordinates on the top strand"
+        },
+        {
+          "where": "tool",
+          "pattern": "specific",
+          "note": "the tool's verdict word for a single clean product"
+        },
+        {
+          "where": "answer",
+          "pattern": "201"
+        },
+        {
+          "where": "answer",
+          "pattern": "101"
+        },
+        {
+          "where": "answer",
+          "pattern": "301"
+        }
       ]
     },
     {
       "id": "gel-preview",
       "category": "visualisation",
       "tier": "full",
-      "covers": ["molbio_virtual_gel"],
+      "covers": [
+        "molbio_virtual_gel"
+      ],
       "instruction": "I am running an EcoRI+HindIII double digest of pUC118.dna (full length 3162 bp, in your working directory) on a 1 kb ladder gel. Draw me the expected band pattern, and tell me how many bands I should see and whether the small fragment will be visible.",
       "timeout_ms": 240000,
-      "expect_tools": ["molbio_virtual_gel"],
-      "fixtures": { "pUC118.dna": "test/fixtures/pUC118.dna" },
+      "expect_tools": [
+        "molbio_virtual_gel"
+      ],
+      "fixtures": {
+        "pUC118.dna": "test/fixtures/pUC118.dna"
+      },
       "assertions": [
         {
           "where": "tool",
@@ -84,13 +161,20 @@
           "pattern": "3111|3\\.1 kb|3\\.1kb|3162 ?- ?51",
           "note": "the backbone fragment. Accepted as the exact size or its conventional `~3.1 kb` shorthand: both are correct, and demanding the digits graded how the model chose to round."
         },
-        { "where": "answer", "pattern": "\\b51\\b", "note": "the small fragment's exact size — pinned, because it explains the whole answer" },
+        {
+          "where": "answer",
+          "pattern": "\\b51\\b",
+          "note": "the small fragment's exact size — pinned, because it explains the whole answer"
+        },
         {
           "where": "answer",
           "pattern": "1 band|one band|2 band|two band|1 fragment|2 fragment|single .{0,12}band",
           "note": "the band count. The model is EXPECTED to distinguish the two theoretical fragments from the single band a 1 kb-ladder gel actually resolves, so both counts are accepted."
         },
-        { "where": "answer", "pattern": "not visible|too small|will not|won't|below|faint|run off|dye front|below detection" }
+        {
+          "where": "answer",
+          "pattern": "not visible|too small|will not|won't|below|faint|run off|dye front|below detection"
+        }
       ]
     }
   ]