bio-rigor

name	bio-rigor
version	1.0.0
description	Enforce scientific rigor whenever you compute or derive a biological result from real data with the Seqeron library — via MCP tools OR the C# API. Use whenever a task involves parsing FASTA/FASTQ/GenBank/GFF/VCF/BED, computing GC%, Tm, distances, k-mers, alignments, variant calls, primers/CRISPR, phylogeny, or popgen metrics, or when asked "is this result reliable/valid?". Enforces tool-only computation (no manual parsing or mental math), a validated operating envelope (LimitationPolicy / LIMITATIONS.md), a provenance chain, cross-checking, unit/coordinate checks, and the alpha / not-for-clinical-use disclaimer.
allowed-tools	Read, Grep, Glob
triggers	["GC content","GC%","melting temperature","Tm","compute","calculate","parse FASTA","parse FASTQ","parse VCF","parse GFF","parse GenBank","parse BED","call variants","variant calling","align","alignment","distance","edit distance","k-mer","kmer","motif","ORF","gene","promoter","primer","probe","CRISPR guide","codon optimization","restriction site","phylogenetic tree","popgen","Fst","taxonomic classification","GC skew","structural variant","is this reliable","is this valid","is this result correct","can I trust","provenance","reproducible","operating envelope","LimitationPolicy","validated"]

bio-rigor — scientific-rigor guardrail for Seqeron

This is a cross-cutting guardrail. It applies whenever you turn real biological input into a computed result using Seqeron — through MCP tools (LLM client) or the C# Seqeron.Genomics API. It does not replace the domain skills; it constrains how they produce answers.

Seqeron is PRE-RELEASE / ALPHA and NOT certified for clinical or diagnostic use (see the disclaimer atop README.md). Enforce the six rules below.

When this applies

• Parsing any bio format (FASTA/FASTQ/GenBank/GFF/VCF/BED/EMBL) into sequences/records.
• Computing any metric: GC%, Tm, composition, complexity, distances, counts, positions, alignment scores, variant calls, primer/CRISPR properties, tree topology, popgen stats.
• Any question of the form "is this result reliable / valid / trustworthy?".

When this does NOT apply

• Pure library development (writing/refactoring/testing Seqeron code) — that is clean-architecture / clean-code / the testing campaigns.
• General discussion of biology with no computation over supplied data.

---

The six rules

1. Tool-only computation

• Parse with a parser, never by eye. Read FASTA/FASTQ/VCF/GFF/GenBank via the parser tool (fasta_parse, fastq_parse, vcf_parse, …) or the C# parser API. Never hand-interpret headers, sequences, coordinates, or quality strings.
• Compute every metric with a call. GC%, Tm, distances, counts, positions, alignment scores — each comes from a tool/Method ID, never from mental math or ad-hoc scratch code.
• No improvised calculation. If no tool/API exists for a step, say so and stop — do not invent a formula, spreadsheet, or one-off script. Use seqeron-discovery to confirm a tool exists.
• This is the discipline the MCP examples repeat verbatim: "Use tools only; no manual parsing or calculations." See docs/mcp/README.md.

2. Envelope awareness

• Before using a guarded algorithm unit, confirm the task is inside its validated operating envelope. The guard is Seqeron.Genomics.Core.LimitationPolicy (3 tiers: Strict < Moderate < Permissive, default Moderate).
• If the task pushes a unit below its MinimumMode, the guarded call throws SeqeronLimitationException. STOP and report the limitation (name it, say why, name the alternative) — do not raise the mode just to force an output.
• The 9 guarded units, their MinimumMode, and the honest scope boundaries live in docs/Validation/LIMITATIONS.md. Details on checking the envelope + the Permissive test bootstrap: reference/envelope.md.

3. Provenance

• Every result carries a provenance block: the tools / Method IDs used, in call order, with the exact parameters. A result without provenance is not reportable.
• Exact block format + a worked example: reference/provenance-format.md.

4. Cross-checking

• Corroborate any critical / decision-relevant result via a second independent tool or an invariant, where one exists. Examples: verify a motif position by re-searching the reverse-complement; confirm a count via an independent counter; check an alignment score against its identity%; validate GC% against explicit A/C/G/T composition.
• If no independent check exists, state that in the provenance rather than implying certainty.

5. Units & coordinates

• Before reporting, verify: coordinate base (0-based vs 1-based — Seqeron tool outputs are 0-based unless a tool doc states otherwise), units (°C for Tm, % for GC, bp for lengths), and alphabet/validity (validate DNA/RNA/protein with the validator tool before deriving anything from it).
• Report units and coordinate base explicitly in the output.

6. Disclaimer

• If a result could be read as clinical, diagnostic, or decision-relevant (e.g. variant pathogenicity, HLA typing, tumour purity, primer/guide for a real assay), surface the alpha / not-for-clinical-use caveat and the "independently validate before relying on it" note from README.md.

---

Dual-mode quick reference

Rule	MCP client	C# API caller
Parse	call *_parse tool	call the parser API, not string ops
Compute	one tool call per metric	one Method ID call per metric
Envelope	read tool doc; stop on guarded-error	check LimitationPolicy; catch SeqeronLimitationException; Permissive bootstrap in tests
Provenance	list tools + args in order	list Method IDs + args in order
Cross-check	second independent tool	second independent method/invariant

Bottom line: parse with a tool, compute with a tool, stay inside the envelope, show your provenance, cross-check what matters, and never let a computed number look more certain — or more clinical — than it is.