Add phylogenetic export support for TNT, PhyG, and IQ-TREE

Extends the phylogenetic pipeline with exporters for TNT, PhyG, and IQ-TREE that generate executable scripts, cost matrices, and recoded alignments. Adds internal helpers for state indexing, floating-point matrix scaling, and Floyd-Warshall metric closure to satisfy external tool constraints. Adjusts visibility modifiers for sibling iterators and entropy annex structs, and updates design documentation for evolutionary distance metrics.

DevDocMD/theory/evolutionary_distances.md

@@ -2182,6 +2182,92 @@ Covered by `iqtree::tests::iqtree_min_freq_folds_rare_states_into_missing`
 from the written `_iqtree_states.csv` and `A`/`C` still present). Full
 workspace `cargo test` green.
 
+## `--distance` unification: SNP corrections as first-class metrics (discussion, 2026-08-28)
+
+**Decided, not yet implemented.** `--metric` (renamed `--distance` — several of
+its existing values, e.g. Bray-Curtis, aren't metrics in the strict sense,
+`--metric` was a misnomer) gains a family of `snp-*` values computed from the
+central-position SNP pipeline, routed internally to the sibling-annex
+machinery (`PairwiseTally`, `obikphylo::siblings::algorithms::pairwise`)
+instead of `cache.distance(...)`'s existing per-layer traversal — a different
+code path behind the same CLI surface, not just another branch of one
+formula function.
+
+**Why unify at the CLI level despite the implementation split**: phylogenetically
+a SNP-corrected distance is a distance like any other — NJ/UPGMA are agnostic
+to how the matrix was produced, so exposing it as a special-cased subcommand
+instead of a `--distance` value would misrepresent its role. The
+implementation divergence (sibling-annex-based vs. plain index scan) is real
+but belongs at the routing layer, invisible to the CLI's own vocabulary.
+
+**`--subsample` becomes optional for `snp-*` distances** (it stays mandatory
+for `--sankoff`/`--pseudo-alignment`, unrelated commands): absent means
+exhaustive, achieved for free by reusing `sample_index`'s existing
+proportional-per-layer-quota mechanism with `n` set to the index-wide total
+non-monomorphic-minorant count (already available from the sibling-annex
+stats) — every layer's quota then equals its own full count, giving Bernoulli
+`p = 1` everywhere, i.e. every eligible family is drawn. No second,
+exhaustive-only driver needed. Present means sampled, exactly as `--sankoff`
+already behaves.
+
+**One shared tally, many derived formulas.** `PairwiseTally`'s `subst[4][4]`
+per-pair substitution counts (plus marginal base frequencies derived from it)
+are the sufficient statistic for every closed-form correction below — each
+is a small pure function `PairwiseTally -> Array2<f64>`, at the same level as
+the already-implemented `raw_snp_distance`/`base_pair_tally`/
+`cardinality_tally`. No new full scan per formula, whether the tally itself
+was built exhaustively or from a subsample.
+
+**`--raw-snp-counts` stays a separate, unrelated flag** — same underlying
+tally, but a diagnostic (`n_snp`/`n_shared`/`n_eligible` per genome pair, one
+row per pair) rather than a distance value, and its long-table shape doesn't
+fold into a single N×N matrix the way a distance does. No change to its
+existing CSV format.
+
+### `snp-*` distance catalog
+
+All closed-form (method-of-moments / direct formula), none requiring
+per-pair or per-tree maximum-likelihood fitting — that excludes HKY85's
+*tree*-ML usage but not its *pairwise* estimator, which is closed-form like
+F84/TN93 and is included below. `snp-` prefix on every CLI value.
+
+| value | corrects for | inputs beyond raw counts |
+|---|---|---|
+| `snp-raw` | nothing (uncorrected p-distance) | — |
+| `snp-jc` (Jukes-Cantor, JC69) | multiple substitutions per site | — |
+| `snp-k2p` (Kimura 2-parameter, K80) | + transition/transversion rate bias | ts/tv split |
+| `snp-k81` (Kimura 3-parameter, K3ST) | + splits transversions into 2 categories | ts/tv split, by category |
+| `snp-f81` (Felsenstein 81) | + unequal base frequencies (no ts/tv split) | empirical base freqs |
+| `snp-tajima-nei` (Tajima-Nei 1984) | same goal as F81 (equal-input model), different formula, better small-sample behavior | empirical base freqs |
+| `snp-t92` (Tamura 3-parameter) | K2P + GC-content bias | ts/tv split, GC content |
+| `snp-f84` (Felsenstein 84) | full empirical base freqs + single ts/tv rate | empirical base freqs, ts/tv split |
+| `snp-hky85` (Hasegawa-Kishino-Yano, pairwise estimator) | same inputs as F84, different formula | empirical base freqs, ts/tv split |
+| `snp-tn93` (Tamura-Nei) | full empirical base freqs + separate purine/pyrimidine transition rates + transversion rate | empirical base freqs, purine-ts/pyrimidine-ts/tv split |
+| `snp-logdet` (LogDet / paralinear) | no shared-model or stationarity assumption at all — general divergence-matrix determinant | full empirical 4×4 divergence matrix (already `subst[4][4]`) |
+| `snp-tv` (transversions-only p-distance) | diagnostic/deep-divergence variant — drops transitions entirely (they saturate first) | tv-only counts |
+
+**`+Γ` rate-heterogeneity modifier, applicable to `snp-jc`, `snp-k2p`,
+`snp-k81`, `snp-t92`, `snp-f84`, `snp-hky85`, `snp-tn93`** (not `snp-raw`,
+nothing to correct; not `snp-logdet`, no standard gamma formulation) — same
+formula as the base correction, weighted by a shape parameter `α` supplied
+by the user (`--gamma-shape <alpha>`), not estimated by ML. A modifier on
+existing values, not a separate enum arm per distance.
+
+### Output format: PHYLIP-relaxed by default for the distance matrix
+
+**Decided, not yet implemented.** The primary distance-matrix output
+(`_dist.csv` today) gains multiple formats: **PHYLIP-relaxed becomes the
+default** (widely read by external NJ tools — PHYLIP `neighbor`, FastME,
+T-REX, SplitsTree — relaxed rather than strict to avoid the 10-character
+label truncation, since genome labels here routinely exceed it), a `--csv`
+flag opts back into the current CSV format, PHYLIP-strict is a possible
+future addition (not now). This changes the *default* output of every
+existing `--distance` value (jaccard, hamming, bray-curtis, ...), not just
+the new `snp-*` ones — accepted explicitly (pre-release, single developer
+user, no external consumers to break). Scoped to the distance matrix only:
+`--shared-kmers` and `--raw-snp-counts` are counts, not distances, and keep
+their existing CSV-only format.
+
 ## References
 
 The Mash mutation-rate model this discussion contrasts with:

src/obikmer2/src/cmd/phylo/args.rs

@@ -40,11 +40,12 @@ impl From<MetricArg> for DistanceMetric {
 /// annex diagnostics (`--sibling-stats`, `--sibling-hist`), entropy
 /// reporting (`--shannon`), SNP pseudo-alignment sampling
 /// (`--pseudo-alignment`, `--subsample`, `--free-loss`, `--no-ambiguity`,
-/// `--entropy`/`--entropy-sd`) and Sankoff cost-matrix calibration
+/// `--entropy`/`--entropy-sd`), Sankoff cost-matrix calibration
-/// (`--sankoff`, `--sankoff-ratio-ceiling`) — everything else sibling-annex-based
+/// (`--sankoff`, `--sankoff-ratio-ceiling`) and its TNT/PhyG/IQ-TREE exports
-/// (`--tnt`/`--phyg`/`--iqtree`, raw SNP distance, family overlap, ...)
+/// (`--tnt`, `--phyg`, `--iqtree`/`--iqtree-min-freq`,
-/// stays in `obikmer` until the rest of `obikphylo::siblings` is
+/// `--sankoff-cost-scale`) — everything else sibling-annex-based (raw SNP
-/// reconnected (see the project memory on this).
+/// distance, family overlap, ...) stays in `obikmer` until the rest of
+/// `obikphylo::siblings` is reconnected (see the project memory on this).
 #[derive(Args)]
 pub struct PhyloArgs {
     /// Index directory
@@ -142,6 +143,68 @@ pub struct PhyloArgs {
     #[arg(long, default_value = "0.5")]
     pub sankoff_ratio_ceiling: f64,
 
+    /// Also write <prefix>_sankoff.tnt, a ready-to-run TNT script (`proc
+    /// <file>;`) for the same matrix/alignment `--sankoff` computes —
+    /// recoded to TNT's default xread alphabet (0-9A-F only; TNT rejects
+    /// the wider IUPAC set `--sankoff`'s own output uses unless `nstates
+    /// dna` is set, which imposes TNT's own incompatible DNA encoding
+    /// instead) with integer-scaled costs (TNT's smatrix/cost commands
+    /// reject decimals). Implies `--sankoff`.
+    #[arg(long)]
+    pub tnt: bool,
+
+    /// Also write <prefix>_sankoff.tcm and <prefix>_sankoff.pg, a
+    /// custom-alphabet cost matrix and a ready-to-run PhyG script (`read`/
+    /// `search`/`report`) for the same matrix/alignment `--sankoff`
+    /// computes. Reuses `--sankoff`'s own `_sankoff.fasta` directly — PhyG's
+    /// `tcm:` alphabet is read from the matrix file itself, so the IUPAC+`0`
+    /// alphabet needs no recoding here, unlike `--tnt`. Implies `--sankoff`.
+    #[arg(long)]
+    pub phyg: bool,
+
+    /// Also write <prefix>_iqtree.model and <prefix>_iqtree.fasta, a
+    /// custom-model file and a matching recoded alignment for genuine
+    /// maximum-likelihood inference with IQ-TREE (`iqtree3 -s ...
+    /// --seqtype MORPH -m ...+ASC`) — real branch lengths, unlike
+    /// `--tnt`/`--phyg`'s parsimony step counts. The model is the
+    /// reversible `Q(i,j) = R(i,j)·π_j` construction: `R` (exchangeability,
+    /// symmetric) recovered from the same calibrated cost matrix
+    /// `--sankoff` computes, `π` the real empirical state frequencies
+    /// counted from the alignment. Only the states that actually occur in
+    /// this alignment are kept, compactly renumbered (IQ-TREE infers its
+    /// state count from the alignment itself, and a gap in the numbering
+    /// would silently misalign the model file). Implies `--sankoff`.
+    #[arg(long)]
+    pub iqtree: bool,
+
+    /// Under `--iqtree --free-loss`, also recode to `?` (the same
+    /// missing-data treatment as `-`) any state whose empirical frequency
+    /// in the alignment falls below this threshold — not just genuinely
+    /// absent calls. States encoding 3 or 4 simultaneously-observed central
+    /// bases (IUPAC `V`/`H`/`K`.../`N` for 3, `N` for 4) are rare by
+    /// construction and often land in exactly this low-frequency range —
+    /// more likely assembly/detection noise than a genuine, widely-preserved
+    /// multi-way polymorphism, the same "sampling failure, not true signal"
+    /// reasoning `--free-loss` already applies to absence. No effect
+    /// without `--free-loss` (there is no missing-data symbol to recode to
+    /// otherwise). `<prefix>_iqtree_states.csv` reports the frequency
+    /// actually used to decide.
+    #[arg(long, default_value = "0.001")]
+    pub iqtree_min_freq: f64,
+
+    /// Scale factor applied before rounding real-valued costs to the
+    /// integers both `--tnt`'s smatrix/cost commands and `--phyg`'s `tcm:`
+    /// matrix require. Keep this small: the total tree score is this scale
+    /// times the sum of per-character costs across every character, and
+    /// there are hints in TNT's own manual that at least some of its
+    /// internal accumulators are 32-bit — a large scale risks a silent
+    /// integer overflow (undetectable, not just a crash) far more costly
+    /// than the resolution a bigger factor would buy. Shared between `--tnt`
+    /// and `--phyg` rather than split into two flags: both scale the same
+    /// calibrated matrix for the same reason (integer-only cost commands).
+    #[arg(long, default_value = "100")]
+    pub sankoff_cost_scale: f64,
+
     /// Distance metric to compute
     #[arg(long, value_enum, default_value = "jaccard")]
     pub metric: MetricArg,

src/obikmer2/src/cmd/phylo/iqtree.rs

@@ -0,0 +1,502 @@
+use std::io::{BufWriter, Write};
+use std::path::PathBuf;
+
+use obifastwrite::{JsonVal, write_record};
+use obikphylo::siblings::SnpAlignment;
+use tracing::info;
+
+use super::sankoff::{STATE_SYMBOL, state_index_table};
+
+// ── Sankoff-calibrated data → IQ-TREE custom ML model + recoded alignment ──
+//
+// Not itself a Sankoff computation — IQ-TREE does maximum likelihood, not
+// parsimony. Only the *source data* is shared with `--tnt`/`--phyg` (the
+// calibrated 16-state cost matrix, the pseudo-alignment); the operation
+// performed on it here is different, hence no "sankoff" in these names,
+// unlike `tnt::write_sankoff_tnt`/`phyg::write_sankoff_phyg`.
+//
+// The mechanism: pass a **file path** directly as `-m`, containing (as
+// whitespace/newline-separated numbers) the lower-triangular exchangeability
+// matrix `R` (`k(k-1)/2` values, PAML row-major order) immediately followed
+// by the `k` state frequencies `π` on the same stream —
+// `ModelMarkov::readRates`/`readStateFreq` read them in that exact order, no
+// header, no separator required.
+//
+// `R` is recovered from the calibrated Sankoff cost matrix via
+// `R(a,b) = exp(-cost(a,b))` (the cost is `-ln(rate)`), symmetric by
+// construction (the underlying tally never captured direction). `π` is the
+// real, empirical, non-uniform marginal frequency of each state across the
+// whole alignment. IQ-TREE reconstructs the (generally asymmetric) rate
+// matrix internally as `Q(i,j) = R(i,j)·π_j` — reversible for *any* `π`, not
+// just uniform, because `R` is symmetric.
+//
+// IQ-TREE infers its state count from the highest-ordinal symbol actually
+// present in the alignment, not from a declared count. So states that never
+// occur anywhere in this particular alignment are dropped, and the
+// survivors are renumbered compactly (`0..k-1`, order preserved) rather than
+// leaving gaps that would silently misalign every value IQ-TREE reads. Both
+// the model and the alignment must agree on this same renumbering, so it's
+// computed once (`CompactAlphabet`) and shared between them.
+
+const IQTREE_STATE_SYMBOL: [char; 16] = [
+    '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'A', 'B', 'C', 'D', 'E', 'F',
+];
+
+struct CompactAlphabet {
+    /// Canonical (0..16) state index -> compact index, for states that occur.
+    old_to_compact: [Option<u8>; 16],
+    /// Compact index -> canonical state index, order-preserving.
+    compact_to_old: Vec<u8>,
+    /// Empirical frequency of each compact-indexed state (sums to 1).
+    freq: Vec<f64>,
+}
+
+impl CompactAlphabet {
+    fn k(&self) -> usize {
+        self.compact_to_old.len()
+    }
+}
+
+/// Under `--free-loss`, non-detection (`-`) becomes IQ-TREE's own missing
+/// symbol (`?`) — ignored when IQ-TREE checks a site's constancy for
+/// `+ASC`. A family kept as "variable" by the sampling (`family_size() >=
+/// 2`, a whole-annex property, oblivious to any one column's actual calls)
+/// can still turn constant *among the genomes that actually have data* once
+/// the non-detected ones are excluded from that check — the same failure
+/// mode `--exclude-genome` already had to account for, just triggered by
+/// hiding cells instead of dropping whole rows. Same remedy: rescan columns
+/// treating `-` as ignored, drop any where the remaining calls agree on a
+/// single state. Parsimony (`--tnt`/`--phyg`) has no no-invariant-site
+/// requirement, so this only runs on IQ-TREE's own copy of the alignment,
+/// never mutating the one the caller also hands to those two exports.
+fn drop_ascertainment_noninformative(alignment: &SnpAlignment) -> SnpAlignment {
+    let n_sites = alignment.sequences.first().map(|s| s.len()).unwrap_or(0);
+    let keep: Vec<bool> = (0..n_sites)
+        .map(|site| {
+            let mut first: Option<u8> = None;
+            for seq in &alignment.sequences {
+                let b = seq[site];
+                if b == b'-' {
+                    continue;
+                }
+                match first {
+                    None => first = Some(b),
+                    Some(f) if f != b => return true,
+                    _ => {}
+                }
+            }
+            false // all calls missing, or all calls agree — non-informative
+        })
+        .collect();
+
+    let sequences = alignment
+        .sequences
+        .iter()
+        .map(|seq| {
+            seq.iter()
+                .zip(keep.iter())
+                .filter(|&(_, &k)| k)
+                .map(|(&b, _)| b)
+                .collect()
+        })
+        .collect();
+    SnpAlignment { sequences, genome_indices: alignment.genome_indices.clone() }
+}
+
+/// Recode every occurrence of a byte in `symbols` to `-` — the same
+/// "absent" byte `drop_ascertainment_noninformative`/`compact_alphabet`
+/// already treat specially under `--free-loss` (recoded to `?` further
+/// downstream). Used by `--iqtree-min-freq` to fold rare, likely-noisy
+/// states into the missing-data treatment before a second
+/// `compact_alphabet` pass, without duplicating that treatment's logic.
+fn recode_symbols_as_absent(alignment: &SnpAlignment, symbols: &[u8]) -> SnpAlignment {
+    let sequences = alignment
+        .sequences
+        .iter()
+        .map(|seq| {
+            seq.iter()
+                .map(|&b| if symbols.contains(&b) { b'-' } else { b })
+                .collect()
+        })
+        .collect();
+    SnpAlignment { sequences, genome_indices: alignment.genome_indices.clone() }
+}
+
+fn compact_alphabet(alignment: &SnpAlignment, free_loss: bool) -> CompactAlphabet {
+    let iupac_to_state = state_index_table();
+
+    let mut occurs = [false; 16];
+    let mut counts = [0u64; 16];
+    for seq in &alignment.sequences {
+        for &b in seq {
+            if free_loss && b == b'-' {
+                // `?`: IQ-TREE's own missing-data symbol for `--seqtype
+                // MORPH`, marginalised by Felsenstein pruning — not a
+                // numbered state, so excluded from `occurs`/`counts` and
+                // from the compact alphabet built below.
+                continue;
+            }
+            let b = if b == b'-' { b'0' } else { b };
+            let state = iupac_to_state[b as usize] as usize;
+            occurs[state] = true;
+            counts[state] += 1;
+        }
+    }
+
+    let mut old_to_compact: [Option<u8>; 16] = [None; 16];
+    let mut compact_to_old: Vec<u8> = Vec::new();
+    for old in 0..16 {
+        if occurs[old] {
+            old_to_compact[old] = Some(compact_to_old.len() as u8);
+            compact_to_old.push(old as u8);
+        }
+    }
+
+    let total: u64 = compact_to_old.iter().map(|&old| counts[old as usize]).sum();
+    let freq: Vec<f64> = compact_to_old
+        .iter()
+        .map(|&old| counts[old as usize] as f64 / total as f64)
+        .collect();
+
+    CompactAlphabet {
+        old_to_compact,
+        compact_to_old,
+        freq,
+    }
+}
+
+/// Write `<prefix>_iqtree_states.csv`: the mapping from IQ-TREE's own
+/// compact state symbols (`0..9A-F`, what actually appears in
+/// `_iqtree.fasta`/`_iqtree.model`) back to the canonical 16-state
+/// alphabet (`STATE_SYMBOL` — the same one `_sankoff_matrix.csv` is
+/// indexed by), plus each state's empirical frequency at full precision
+/// (`_iqtree.model`'s own frequency line is truncated to 6 decimals).
+/// Without this file, a compact index in `_iqtree.model`'s `R`/`π` output
+/// (e.g. "state 0 has zero exchangeability with everything else") can't be
+/// traced back to which real state that is.
+fn write_iqtree_states_csv(alphabet: &CompactAlphabet, output: &Option<PathBuf>) -> String {
+    let path = output
+        .as_ref()
+        .map(|p| format!("{}_iqtree_states.csv", p.display()))
+        .unwrap_or_else(|| "iqtree_states.csv".into());
+    let mut f = BufWriter::new(std::fs::File::create(&path).unwrap_or_else(|e| {
+        eprintln!("error creating {path}: {e}");
+        std::process::exit(1);
+    }));
+    writeln!(f, "iqtree_symbol,canonical_symbol,frequency").unwrap();
+    for (compact, &old) in alphabet.compact_to_old.iter().enumerate() {
+        writeln!(
+            f,
+            "{},{},{}",
+            IQTREE_STATE_SYMBOL[compact], STATE_SYMBOL[old as usize], alphabet.freq[compact]
+        )
+        .unwrap();
+    }
+    path
+}
+
+/// Write the `R` (exchangeability) + `π` (frequencies) model file IQ-TREE's
+/// `-m <file>+ASC` reads. Returns the path, so the caller can print a
+/// single combined "how to run this" message once the alignment is also
+/// written. The one bit of real computation this whole adapter does:
+/// `R(a,b) = exp(-cost(a,b))`, recovering the exchangeability rate a
+/// calibrated Sankoff parsimony cost implies for a continuous-time model —
+/// a one-line inversion of the cost matrix's own `-ln(rate)` construction,
+/// not a new estimate.
+fn write_iqtree_model(
+    matrix: &[[f64; 16]; 16],
+    alphabet: &CompactAlphabet,
+    output: &Option<PathBuf>,
+) -> String {
+    let rate = |old_i: u8, old_j: u8| (-matrix[old_i as usize][old_j as usize]).exp();
+
+    let model_path = output
+        .as_ref()
+        .map(|p| format!("{}_iqtree.model", p.display()))
+        .unwrap_or_else(|| "iqtree.model".into());
+    let mut f = BufWriter::new(std::fs::File::create(&model_path).unwrap_or_else(|e| {
+        eprintln!("error creating {model_path}: {e}");
+        std::process::exit(1);
+    }));
+    for i in 1..alphabet.k() {
+        let row: Vec<String> = (0..i)
+            .map(|j| {
+                format!(
+                    "{:.6}",
+                    rate(alphabet.compact_to_old[i], alphabet.compact_to_old[j])
+                )
+            })
+            .collect();
+        writeln!(f, "{}", row.join(" ")).unwrap();
+    }
+    writeln!(
+        f,
+        "{}",
+        alphabet
+            .freq
+            .iter()
+            .map(|p| format!("{p:.6}"))
+            .collect::<Vec<_>>()
+            .join(" ")
+    )
+    .unwrap();
+    info!(
+        "IQ-TREE model file → {model_path} ({} of 16 states present in the alignment)",
+        alphabet.k()
+    );
+    model_path
+}
+
+/// Write the pseudo-alignment recoded to the same compact `0..k-1` alphabet
+/// as `write_iqtree_model`'s matrix — not `--sankoff`'s own IUPAC alphabet,
+/// since IQ-TREE needs the symbol ordinal itself to match the surviving
+/// state count (see this module's own doc comment on state inference).
+fn write_iqtree_alignment(
+    alignment: &SnpAlignment,
+    labels: &[String],
+    alphabet: &CompactAlphabet,
+    output: &Option<PathBuf>,
+    free_loss: bool,
+) -> (String, usize) {
+    let iupac_to_state = state_index_table();
+
+    let fasta_path = output
+        .as_ref()
+        .map(|p| format!("{}_iqtree.fasta", p.display()))
+        .unwrap_or_else(|| "iqtree.fasta".into());
+    let mut f = BufWriter::new(std::fs::File::create(&fasta_path).unwrap_or_else(|e| {
+        eprintln!("error creating {fasta_path}: {e}");
+        std::process::exit(1);
+    }));
+    let n_sites = alignment.sequences.first().map(|s| s.len()).unwrap_or(0);
+    for (&g, seq) in alignment.genome_indices.iter().zip(alignment.sequences.iter()) {
+        let recoded: Vec<u8> = seq
+            .iter()
+            .map(|&b| {
+                if free_loss && b == b'-' {
+                    return b'?';
+                }
+                let b = if b == b'-' { b'0' } else { b };
+                let old = iupac_to_state[b as usize] as usize;
+                let compact = alphabet.old_to_compact[old]
+                    .expect("state occurs in the alignment, so it must have a compact index");
+                IQTREE_STATE_SYMBOL[compact as usize] as u8
+            })
+            .collect();
+        write_record(
+            &recoded,
+            &labels[g],
+            &[("n_sites", JsonVal::Num(n_sites as u64))],
+            &mut f,
+        )
+        .unwrap_or_else(|e| {
+            eprintln!("error writing {fasta_path}: {e}");
+            std::process::exit(1);
+        });
+    }
+    (fasta_path, n_sites)
+}
+
+pub(super) fn write_iqtree(
+    matrix: &[[f64; 16]; 16],
+    alignment: &SnpAlignment,
+    labels: &[String],
+    output: &Option<PathBuf>,
+    free_loss: bool,
+    min_freq: f64,
+) {
+    let filtered;
+    let alignment = if free_loss {
+        let before = alignment.sequences.first().map(|s| s.len()).unwrap_or(0);
+        filtered = drop_ascertainment_noninformative(alignment);
+        let after = filtered.sequences.first().map(|s| s.len()).unwrap_or(0);
+        if after != before {
+            info!(
+                "--free-loss: {before} → {after} sites (dropped columns non-informative once `-` \
+                 is treated as missing — required for +ASC)"
+            );
+        }
+        &filtered
+    } else {
+        alignment
+    };
+
+    let mut alphabet = compact_alphabet(alignment, free_loss);
+
+    // `--iqtree-min-freq`: fold rare (likely-noisy) states into the same
+    // missing-data treatment `-` already gets under `--free-loss`, then
+    // recompute the alphabet on the further-filtered alignment.
+    let refiltered;
+    let alignment = if free_loss {
+        let low_freq_symbols: Vec<u8> = alphabet
+            .compact_to_old
+            .iter()
+            .zip(alphabet.freq.iter())
+            .filter(|&(_, &f)| f < min_freq)
+            .map(|(&old, _)| STATE_SYMBOL[old as usize] as u8)
+            .collect();
+        if low_freq_symbols.is_empty() {
+            alignment
+        } else {
+            let recoded = recode_symbols_as_absent(alignment, &low_freq_symbols);
+            let before = recoded.sequences.first().map(|s| s.len()).unwrap_or(0);
+            refiltered = drop_ascertainment_noninformative(&recoded);
+            let after = refiltered.sequences.first().map(|s| s.len()).unwrap_or(0);
+            info!(
+                "--iqtree-min-freq {min_freq}: {} rare state(s) ({}) recoded as missing, {before} → {after} sites",
+                low_freq_symbols.len(),
+                low_freq_symbols
+                    .iter()
+                    .map(|&b| b as char)
+                    .collect::<String>(),
+            );
+            alphabet = compact_alphabet(&refiltered, free_loss);
+            &refiltered
+        }
+    } else {
+        alignment
+    };
+
+    let states_path = write_iqtree_states_csv(&alphabet, output);
+    let model_path = write_iqtree_model(matrix, &alphabet, output);
+    let (fasta_path, n_sites) =
+        write_iqtree_alignment(alignment, labels, &alphabet, output, free_loss);
+
+    let prefix_name = output
+        .as_ref()
+        .and_then(|p| p.file_name())
+        .map(|n| format!("{}_iqtree", n.to_string_lossy()))
+        .unwrap_or_else(|| "iqtree".into());
+    info!(
+        "IQ-TREE alignment → {fasta_path} ({n_sites} sites, {} states)\n\
+         IQ-TREE state mapping → {states_path}\n\
+         Run with:\n  \
+         iqtree3 -s {fasta_path} --seqtype MORPH -m {model_path}+ASC --prefix {prefix_name} -T AUTO\n\
+         \n\
+         options -alrt 1000 -B 1000 can be added to evaluate robustness of the tree",
+        alphabet.k()
+    );
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn alignment(sequences: Vec<Vec<u8>>) -> SnpAlignment {
+        let genome_indices = (0..sequences.len()).collect();
+        SnpAlignment { sequences, genome_indices }
+    }
+
+    #[test]
+    fn free_loss_excludes_absent_state_and_freq_sums_to_one() {
+        // 3 genomes, 2 sites. Site 0: g1='A', g2='C', g3='-' (absent).
+        // Site 1: g1='-', g2='-', g3='G'. Under free_loss, every '-' must
+        // be excluded from the frequency count entirely (not folded into
+        // state 0).
+        let alignment = alignment(vec![vec![b'A', b'-'], vec![b'C', b'-'], vec![b'-', b'G']]);
+
+        let alphabet = compact_alphabet(&alignment, true);
+
+        assert!(
+            !alphabet.compact_to_old.contains(&0),
+            "state 0 (absent) must not appear in the compact alphabet under --free-loss, got {:?}",
+            alphabet.compact_to_old
+        );
+        let sum: f64 = alphabet.freq.iter().sum();
+        assert!(
+            (sum - 1.0).abs() < 1e-9,
+            "frequencies must sum to 1, got {sum} ({:?})",
+            alphabet.freq
+        );
+        assert_eq!(alphabet.k(), 3, "A, C, G — 3 real states, `-` excluded");
+    }
+
+    #[test]
+    fn without_free_loss_absent_state_is_counted_normally() {
+        let alignment = alignment(vec![vec![b'A', b'-'], vec![b'C', b'-'], vec![b'-', b'G']]);
+
+        let alphabet = compact_alphabet(&alignment, false);
+
+        assert!(
+            alphabet.compact_to_old.contains(&0),
+            "state 0 (absent, recoded from '-') must be counted when --free-loss is off"
+        );
+        let sum: f64 = alphabet.freq.iter().sum();
+        assert!(
+            (sum - 1.0).abs() < 1e-9,
+            "frequencies must sum to 1, got {sum} ({:?})",
+            alphabet.freq
+        );
+    }
+
+    #[test]
+    fn states_csv_maps_compact_symbols_back_to_canonical_ones() {
+        // 'A' (state 1) and 'G' (state 4) occur, '-' (state 0) excluded by
+        // --free-loss — compact index 0 -> 'A', compact index 1 -> 'G'.
+        let alignment = alignment(vec![vec![b'A', b'-'], vec![b'-', b'G']]);
+        let alphabet = compact_alphabet(&alignment, true);
+        let output = Some(
+            std::env::temp_dir().join(format!("obikmer2_test_iqtree_states_{}", std::process::id())),
+        );
+
+        let path = write_iqtree_states_csv(&alphabet, &output);
+        let csv = std::fs::read_to_string(&path).unwrap();
+        std::fs::remove_file(&path).ok();
+        let mut lines = csv.lines();
+        assert_eq!(
+            lines.next(),
+            Some("iqtree_symbol,canonical_symbol,frequency")
+        );
+        assert_eq!(lines.next(), Some("0,A,0.5"));
+        assert_eq!(lines.next(), Some("1,G,0.5"));
+        assert!(lines.next().is_none());
+    }
+
+    #[test]
+    fn iqtree_min_freq_folds_rare_states_into_missing() {
+        // 20 common A/C sites (60 calls total across 3 genomes) plus one
+        // site where genome 0 carries the rare ambiguity state `M` and
+        // genome 1 carries `A` (kept informative by the first
+        // ascertainment filter: two distinct non-`-` calls) — `M` ends up
+        // at 1/62, well below the 0.05 threshold used here.
+        let mut sequences: Vec<Vec<u8>> = vec![Vec::new(); 3];
+        for i in 0..20 {
+            let (a, b, c) = if i % 2 == 0 {
+                (b'A', b'C', b'A')
+            } else {
+                (b'C', b'A', b'C')
+            };
+            sequences[0].push(a);
+            sequences[1].push(b);
+            sequences[2].push(c);
+        }
+        sequences[0].push(b'M');
+        sequences[1].push(b'A');
+        sequences[2].push(b'-');
+        let alignment = alignment(sequences);
+        let labels = vec!["g1".to_string(), "g2".to_string(), "g3".to_string()];
+        let matrix = [[0.0f64; 16]; 16];
+        let prefix = std::env::temp_dir().join(format!(
+            "obikmer2_test_iqtree_minfreq_{}",
+            std::process::id()
+        ));
+        let output = Some(prefix.clone());
+
+        write_iqtree(&matrix, &alignment, &labels, &output, true, 0.05);
+
+        let states_path = format!("{}_iqtree_states.csv", prefix.display());
+        let csv = std::fs::read_to_string(&states_path).unwrap();
+        assert!(
+            !csv.contains(",M,"),
+            "M (freq ~1/62) must be folded into missing under --iqtree-min-freq 0.05, got:\n{csv}"
+        );
+        assert!(
+            csv.contains(",A,") && csv.contains(",C,"),
+            "A/C must survive (well above threshold), got:\n{csv}"
+        );
+
+        for suffix in ["_iqtree_states.csv", "_iqtree.model", "_iqtree.fasta"] {
+            std::fs::remove_file(format!("{}{suffix}", prefix.display())).ok();
+        }
+    }
+}

src/obikmer2/src/cmd/phylo/mod.rs

@@ -1,5 +1,8 @@
 mod args;
+mod iqtree;
+mod phyg;
 mod sankoff;
+mod tnt;
 
 use std::io::{self, BufWriter, Write};
 use std::sync::Arc;
@@ -14,7 +17,10 @@ use obikphylo::{Metrics, neighbor_joining, upgma};
 use obisys::{Reporter, Stage};
 use tracing::info;
 
+use iqtree::write_iqtree;
+use phyg::write_sankoff_phyg;
 use sankoff::{write_sankoff_alignment_fasta, write_sankoff_matrix_csv, write_sankoff_params};
+use tnt::write_sankoff_tnt;
 
 pub use args::PhyloArgs;
 
@@ -201,8 +207,8 @@ pub fn run(args: PhyloArgs) {
         info!("pseudo-alignment ({n_sites} site(s), {} genome(s)) → {path}", alignment.genome_indices.len());
     }
 
-    // ── Sankoff cost-matrix calibration (`--sankoff`) ───────────────────────────
+    // ── Sankoff cost-matrix calibration (`--sankoff`, `--tnt`, `--phyg`, `--iqtree`) ──
-    if args.sankoff {
+    if args.sankoff || args.tnt || args.phyg || args.iqtree {
         let Some(subsample_n) = args.subsample else {
             eprintln!("error: --sankoff requires --subsample <N>");
             std::process::exit(1);
@@ -239,6 +245,30 @@ pub fn run(args: PhyloArgs) {
             &args.output,
         );
         write_sankoff_alignment_fasta(&bundle.alignment, &labels, &args.output, args.free_loss);
+
+        if args.tnt {
+            write_sankoff_tnt(
+                &matrix,
+                &bundle.alignment,
+                &labels,
+                &args.output,
+                args.sankoff_cost_scale,
+                args.free_loss,
+            );
+        }
+        if args.phyg {
+            write_sankoff_phyg(&matrix, &args.output, args.sankoff_cost_scale);
+        }
+        if args.iqtree {
+            write_iqtree(
+                &matrix,
+                &bundle.alignment,
+                &labels,
+                &args.output,
+                args.free_loss,
+                args.iqtree_min_freq,
+            );
+        }
     }
 
     info!("computing {:?} distances for {} genome(s)", args.metric, n);

src/obikmer2/src/cmd/phylo/phyg.rs

@@ -0,0 +1,80 @@
+use std::io::{BufWriter, Write};
+use std::path::PathBuf;
+
+use tracing::info;
+
+use super::sankoff::{STATE_SYMBOL, scaled_metric_matrix};
+
+// ── Sankoff cost matrix → PhyG custom-alphabet TCM + ready-to-run script ────
+//
+// PhyG's `tcm:STRING` format needs no alphabet recoding, unlike `--tnt`:
+// its parser reads the alphabet straight from the tcm file's own first
+// line, so `--sankoff`'s own `_sankoff.fasta` (already IUPAC+`0`) is reused
+// as-is via `prefasta:`. PhyG auto-adds its own indel/gap state as an
+// (n+1)-th row/column of the tcm — inert here since the alignment already
+// encodes absence as an ordinary state (`0`), never as `-` (see
+// `sankoff::write_sankoff_alignment_fasta`'s own comment on why). The gap
+// row/column below reuses `matrix[i][0]`/`matrix[0][j]` (cost to/from `∅`)
+// as the closest principled value for a state that, in practice, is never
+// actually triggered.
+
+pub(super) fn write_sankoff_phyg(matrix: &[[f64; 16]; 16], output: &Option<PathBuf>, cost_scale: f64) {
+    let scaled_matrix = scaled_metric_matrix(matrix, cost_scale);
+
+    let basename = |suffix: &str| -> String {
+        output.as_ref()
+            .and_then(|p| p.file_name())
+            .map(|n| format!("{}{suffix}", n.to_string_lossy()))
+            .unwrap_or_else(|| format!("sankoff{suffix}"))
+    };
+    let full_path = |suffix: &str| -> String {
+        output.as_ref()
+            .map(|p| format!("{}{suffix}", p.display()))
+            .unwrap_or_else(|| format!("sankoff{suffix}"))
+    };
+
+    let tcm_path = full_path("_sankoff.tcm");
+    let mut f = BufWriter::new(std::fs::File::create(&tcm_path).unwrap_or_else(|e| {
+        eprintln!("error creating {tcm_path}: {e}");
+        std::process::exit(1);
+    }));
+    let alphabet_line = STATE_SYMBOL.iter().map(|c| c.to_string()).collect::<Vec<_>>().join(" ");
+    writeln!(f, "{alphabet_line}").unwrap();
+    for i in 0..16 {
+        let mut row: Vec<i64> = (0..16).map(|j| scaled_matrix[i][j]).collect();
+        row.push(scaled_matrix[i][0]); // gap column: same cost as to/from ∅
+        writeln!(f, "{}", row.iter().map(|v| v.to_string()).collect::<Vec<_>>().join(" ")).unwrap();
+    }
+    let mut gap_row: Vec<i64> = (0..16).map(|j| scaled_matrix[0][j]).collect();
+    gap_row.push(0);
+    writeln!(f, "{}", gap_row.iter().map(|v| v.to_string()).collect::<Vec<_>>().join(" ")).unwrap();
+    info!("PhyG TCM → {tcm_path}");
+
+    let pg_path = full_path("_sankoff.pg");
+    let mut f = BufWriter::new(std::fs::File::create(&pg_path).unwrap_or_else(|e| {
+        eprintln!("error creating {pg_path}: {e}");
+        std::process::exit(1);
+    }));
+    let fasta_name = basename("_sankoff.fasta");
+    let tcm_name = basename("_sankoff.tcm");
+    let tre_name = basename("_sankoff.tre");
+    writeln!(f, "read(prefasta:\"{fasta_name}\", tcm:\"{tcm_name}\")").unwrap();
+    writeln!(f, "search(seconds:300, instances:4)").unwrap();
+    writeln!(f, "report(\"{tre_name}\", graphs, newick, overwrite)").unwrap();
+
+    let pg_dir = std::path::Path::new(&pg_path).parent()
+        .filter(|d| !d.as_os_str().is_empty())
+        .map(|d| d.display().to_string())
+        .unwrap_or_else(|| ".".into());
+    let pg_name = std::path::Path::new(&pg_path).file_name()
+        .map(|n| n.to_string_lossy().into_owned())
+        .unwrap_or_else(|| pg_path.clone());
+    info!(
+        "PhyG script → {pg_path} (costs scaled x{cost_scale:.0}, runs a default 300s/4-instance \
+         search and writes trees to {tre_name})\n\
+         Run it with:\n  \
+         cd {pg_dir} && phyg {pg_name}\n\
+         (`phyg` must run from that directory — `read()`/`report()` in the script use relative \
+         file names)"
+    );
+}

src/obikmer2/src/cmd/phylo/sankoff.rs

@@ -66,6 +66,17 @@ pub(super) const STATE_SYMBOL: [char; 16] = [
     '0', 'A', 'C', 'M', 'G', 'R', 'S', 'V', 'T', 'W', 'Y', 'H', 'K', 'D', 'B', 'N',
 ];
 
+/// `STATE_SYMBOL` byte -> state index (0..16), for adapters that need to
+/// translate an alignment written in this alphabet into their own. Shared
+/// rather than rebuilt per adapter (`tnt`, `iqtree`).
+pub(super) fn state_index_table() -> [u8; 128] {
+    let mut table = [0u8; 128];
+    for (state, &sym) in STATE_SYMBOL.iter().enumerate() {
+        table[sym as usize] = state as u8;
+    }
+    table
+}
+
 pub(super) fn write_sankoff_matrix_csv(matrix: &[[f64; 16]; 16], output: &Option<PathBuf>) {
     let path = output.as_ref()
         .map(|p| format!("{}_sankoff_matrix.csv", p.display()))
@@ -166,3 +177,39 @@ pub(super) fn write_sankoff_params(
     });
     info!("Sankoff calibration parameters → {path}");
 }
+
+/// Scale `matrix` by `cost_scale` and round to integers (TNT's smatrix/cost
+/// and PhyG's `tcm:` commands both reject decimals), then take the *metric
+/// closure* of the result (Floyd-Warshall over the 16 states again, on the
+/// now-integer values).
+///
+/// `pairwise_cost_matrix`'s row-normalise-then-`-ln` construction gives no
+/// guarantee of being a metric (unlike a cost graph closed by shortest path
+/// by construction) — so this closure isn't only needed to correct
+/// integer-rounding artifacts (two real costs of `1.734` each round to
+/// `173`, summing to `346`, while their own real sum `3.468` rounds to
+/// `347` — TNT then reports "triangle inequality violated ... Fixed" and
+/// silently substitutes its own corrected value), it may also be the only
+/// thing making the *real-valued* matrix a metric in the first place.
+/// Re-closing after rounding makes both corrections explicit and
+/// reproducible here instead, rather than left implicit and
+/// tool-version-dependent.
+pub(super) fn scaled_metric_matrix(matrix: &[[f64; 16]; 16], cost_scale: f64) -> [[i64; 16]; 16] {
+    let mut m = [[0i64; 16]; 16];
+    for i in 0..16 {
+        for j in 0..16 {
+            m[i][j] = (matrix[i][j] * cost_scale).round() as i64;
+        }
+    }
+    for k in 0..16 {
+        for i in 0..16 {
+            for j in 0..16 {
+                let via = m[i][k] + m[k][j];
+                if via < m[i][j] {
+                    m[i][j] = via;
+                }
+            }
+        }
+    }
+    m
+}

src/obikmer2/src/cmd/phylo/tnt.rs

@@ -0,0 +1,184 @@
+use std::io::{BufWriter, Write};
+use std::path::PathBuf;
+
+use obikphylo::siblings::SnpAlignment;
+use tracing::info;
+
+use super::sankoff::{scaled_metric_matrix, state_index_table};
+
+// ── Sankoff cost matrix + alignment → ready-to-run TNT script ──────────────
+//
+// TNT's *default* xread reader only accepts its own 0-9A-F alphabet (see
+// its manual: "up to 16 states are allowed by xread, using symbols 0-9 ...
+// and A-F") — the wider IUPAC set `STATE_SYMBOL` uses is rejected as an
+// "alien symbol" unless `nstates dna` is set, which imposes TNT's own fixed
+// DNA encoding instead, incompatible with a custom smatrix. And TNT's
+// `smatrix`/`cost` commands reject decimal costs ("found symbol . when
+// reading transformation costs"). So: recode to TNT's alphabet and
+// integer-scale the costs here, at this adapter's boundary, rather than
+// degrading the project's own canonical (IUPAC, real-valued) output.
+
+const TNT_STATE_SYMBOL: [char; 16] = [
+    '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'A', 'B', 'C', 'D', 'E', 'F',
+];
+
+pub(super) fn write_sankoff_tnt(
+    matrix: &[[f64; 16]; 16],
+    alignment: &SnpAlignment,
+    labels: &[String],
+    output: &Option<PathBuf>,
+    cost_scale: f64,
+    free_loss: bool,
+) {
+    let path = output
+        .as_ref()
+        .map(|p| format!("{}_sankoff.tnt", p.display()))
+        .unwrap_or_else(|| "sankoff.tnt".into());
+    let mut f = BufWriter::new(std::fs::File::create(&path).unwrap_or_else(|e| {
+        eprintln!("error creating {path}: {e}");
+        std::process::exit(1);
+    }));
+
+    // IUPAC-ish symbol -> bitmask, to translate the alignment (which uses
+    // `STATE_SYMBOL`, `-` already normalised to `0` by `sankoff_bundle`
+    // callers) into TNT's alphabet without re-deriving state indices.
+    let iupac_to_state = state_index_table();
+
+    let n_sites = alignment.sequences.first().map(|s| s.len()).unwrap_or(0);
+    let kept_labels: Vec<&String> = alignment.genome_indices.iter().map(|&g| &labels[g]).collect();
+    writeln!(f, "xread").unwrap();
+    writeln!(f, "mxram 16000;").unwrap();
+    writeln!(f, "taxname =;").unwrap();
+    writeln!(f, "taxname +50;").unwrap();
+
+    writeln!(
+        f,
+        "'obikmer central-position SNP families, calibrated Sankoff 16-state encoding'"
+    )
+    .unwrap();
+    writeln!(f, "{n_sites} {}", kept_labels.len()).unwrap();
+    for (label, seq) in kept_labels.iter().zip(alignment.sequences.iter()) {
+        write!(f, "{label} ").unwrap();
+        for &b in seq {
+            if free_loss && b == b'-' {
+                // `?`: TNT's own missing-data symbol, read directly, not
+                // routed through `TNT_STATE_SYMBOL` (there is no state for
+                // it) — see `write_sankoff_alignment_fasta`'s doc comment.
+                write!(f, "?").unwrap();
+                continue;
+            }
+            let b = if b == b'-' { b'0' } else { b };
+            let state = iupac_to_state[b as usize];
+            write!(f, "{}", TNT_STATE_SYMBOL[state as usize]).unwrap();
+        }
+        writeln!(f).unwrap();
+    }
+    writeln!(f, ";\n").unwrap();
+
+    let scaled_matrix = scaled_metric_matrix(matrix, cost_scale);
+
+    writeln!(f, "smatrix =0 (family16)").unwrap();
+    for i in 0..16 {
+        for j in (i + 1)..16 {
+            writeln!(
+                f,
+                "{}/{} {}",
+                TNT_STATE_SYMBOL[i], TNT_STATE_SYMBOL[j], scaled_matrix[i][j]
+            )
+            .unwrap();
+        }
+    }
+    writeln!(f, ";\n").unwrap();
+
+    writeln!(f, "ccode ( 0.{} ;", n_sites - 1).unwrap();
+    writeln!(f, "smatrix +0 0.{} ;", n_sites - 1).unwrap();
+    writeln!(f).unwrap();
+
+    // Basename only (not the full `path`/`output` prefix): TNT's natural
+    // workflow is to `cd` into the output directory before `proc`-ing the
+    // script, and an absolute path here would break if that directory is
+    // later moved or copied elsewhere.
+    let tre_name = output
+        .as_ref()
+        .and_then(|p| p.file_name())
+        .map(|n| format!("{}_sankoff.tre", n.to_string_lossy()))
+        .unwrap_or_else(|| "sankoff.tre".into());
+
+    // TNT's plain command parser has no comment syntax of its own — `/* */`
+    // and `[ ]` are only recognised inside the (separately-enabled) macro
+    // scripting language, and fail with "No command!" here otherwise
+    // (verified against this file with the local TNT binary). `quote` is
+    // the closest working equivalent: it prints free text and does not
+    // otherwise affect parsing, so it doubles as an explanation of the
+    // defaults below when the script is run. `;` ends a `quote` block like
+    // any other TNT command, so the text itself must avoid semicolons.
+    writeln!(f, "quote").unwrap();
+    writeln!(
+        f,
+        "Default search below (edit or delete this block to run your own strategy):"
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "  hold N     : size of TNT's tree buffer (how many equally-parsimonious"
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "               trees it keeps in memory at once), 20 is a small, fast"
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "               default, raise it if mult reports it had to drop trees."
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "  mult       : traditional search (random addition sequences followed by"
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "               TBR branch-swapping, TNT's own default replication count),"
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "               a reasonable first-pass strategy on this data's memory"
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "               footprint, xmult's ratchet/drift/tree-fusion buffers ran"
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "               this out of RAM at TNT's default mxram on this dataset."
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "  export - F : write the trees held in the buffer to file F, in"
+    )
+    .unwrap();
+    writeln!(
+        f,
+        "               TNT/Hennig86 format ('-' means trees, as opposed to data)."
+    )
+    .unwrap();
+    writeln!(f, ";").unwrap();
+    writeln!(f, "hold 20;").unwrap();
+    writeln!(f, "mult;").unwrap();
+    writeln!(f, "export - {tre_name};").unwrap();
+
+    info!(
+        "TNT script → {path} (costs scaled x{cost_scale:.0}, runs a default `hold 20; mult;` \
+         search and writes trees to {tre_name} in TNT's working directory — edit the trailing \
+         comment block in the script to change this)\n\
+         Run it with:\n  \
+         printf 'proc {path};\\nquit;\\n' | tnt\n\
+         (or start `tnt` interactively and type `proc {path};`)"
+    );
+}

src/obikphylo/src/siblings/algorithms/mod.rs

@@ -27,20 +27,21 @@ mod subsample;
 
 use obikidxcache::index_cache::IndexCache;
 
-pub use cardcomp::{cardinality_transition_probs, composition_transition_probs, pairwise_cost_matrix};
 pub use alignment::SnpAlignment;
+pub use cardcomp::{
+    cardinality_transition_probs, composition_transition_probs, pairwise_cost_matrix,
+};
 pub use pairwise::{BasePairTally, CardinalityTally, RawSnpDistanceOutput};
 pub use sankoff::SankoffBundle;
 pub use stats::SiblingAnnexStats;
-pub use subsample::EntropyBias;
+pub use subsample::{EntropyBias, SurvivingFamily};
 
 pub(crate) use alignment::snp_pseudo_alignment;
 pub(crate) use annex::build_layer_sibling_annex;
-pub(crate) use entropy::{ensure_layer_entropy_annex, family_entropy, family_entropy_4, iter_full_entropy};
+pub(crate) use entropy::{ensure_layer_entropy_annex, family_entropy, family_entropy_4};
 pub(crate) use family_scan::{Selection, scan_layer_families};
 pub(crate) use sankoff::sankoff_bundle;
 pub(crate) use stats::{sibling_annex_stats, sibling_family_size_histogram};
-pub(crate) use subsample::sample_index;
 
 /// Whether every layer number in `cache` fits in a `FamilyMask` field
 /// (`< 7`) — a single fact about the whole index, decided once here so
@@ -51,5 +52,10 @@ pub(crate) use subsample::sample_index;
 /// interprets that field (`SiblingExt::build_sibling_annex`,
 /// `SiblingBuilder::ensure_entropy_annexes`), so they can never disagree.
 pub(crate) fn is_fast_mode(cache: &IndexCache) -> bool {
-    cache.partitions().filter_map(|p| cache.n_layer(p)).max().unwrap_or(0) <= 7
+    cache
+        .partitions()
+        .filter_map(|p| cache.n_layer(p))
+        .max()
+        .unwrap_or(0)
+        <= 7
 }

src/obikphylo/src/siblings/algorithms/subsample.rs

@@ -71,7 +71,7 @@ const MAX_TOPUP_ROUNDS: usize = 5;
 /// outlive that callback, but a `SurvivingFamily` is meant to be handed to
 /// [`sample_index`]'s caller as part of a whole layer's batch, well after
 /// that callback returns.
-pub(crate) struct SurvivingFamily {
+pub struct SurvivingFamily {
     pub family_idx: usize,
     pub mask: FamilyMask,
     pub genome_mask: Vec<u8>,
@@ -162,7 +162,10 @@ pub(crate) fn sample_index(
         }
     }
 
-    let total_eligible: u64 = layers.iter().map(|(_, _, e, _)| e.view().count_ones()).sum();
+    let total_eligible: u64 = layers
+        .iter()
+        .map(|(_, _, e, _)| e.view().count_ones())
+        .sum();
     if total_eligible == 0 {
         return Ok(0);
     }
@@ -272,7 +275,9 @@ fn sample_layer(
             Some(_) => Box::new(circular_entropy_values(&layer_dir, round_start)?),
             None => Box::new(std::iter::repeat(0.0f32)), // never read: kernel forced to 1.0 below
         };
-        let candidates = (round_start..n_minorants).chain(0..round_start).zip(entropies);
+        let candidates = (round_start..n_minorants)
+            .chain(0..round_start)
+            .zip(entropies);
 
         let mut accepted: HashSet<usize> = HashSet::new();
         for (family_idx, entropy) in candidates {

src/obikphylo/src/siblings/entropy_annex.rs

@@ -35,41 +35,47 @@ const SENTINEL: f32 = -1.0;
 
 pub(crate) const ENTROPY_ANNEX_FILE_NAME: &str = "entropy.pent";
 
-pub(crate) struct EntropyAnnex {
+pub struct EntropyAnnex {
     mmap: Mmap,
     n: usize,
 }
 
 impl EntropyAnnex {
-    pub(crate) fn open(path: &Path) -> io::Result<Self> {
+    pub fn open(path: &Path) -> io::Result<Self> {
         let mmap = unsafe { Mmap::map(&File::open(path)?)? };
         if mmap.len() < HEADER_SIZE {
-            return Err(io::Error::new(io::ErrorKind::InvalidData, "PENT file too short"));
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                "PENT file too short",
+            ));
         }
         if mmap[0..4] != MAGIC {
             return Err(io::Error::new(io::ErrorKind::InvalidData, "bad PENT magic"));
         }
         let n = u64::from_le_bytes(mmap[8..16].try_into().unwrap()) as usize;
         if mmap.len() < HEADER_SIZE + n * 4 {
-            return Err(io::Error::new(io::ErrorKind::InvalidData, "PENT file truncated"));
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                "PENT file truncated",
+            ));
         }
         Ok(Self { mmap, n })
     }
 
-    pub(crate) fn len(&self) -> usize {
+    pub fn len(&self) -> usize {
         self.n
     }
 
     /// `None` for the sentinel (not yet computed — shouldn't happen against
     /// a fully-built annex).
-    pub(crate) fn get(&self, idx: usize) -> Option<f32> {
+    pub fn get(&self, idx: usize) -> Option<f32> {
         let off = HEADER_SIZE + idx * 4;
         let v = f32::from_le_bytes(self.mmap[off..off + 4].try_into().unwrap());
         (v >= 0.0).then_some(v)
     }
 }
 
-pub(crate) struct EntropyAnnexBuilder {
+pub struct EntropyAnnexBuilder {
     mmap: MmapMut,
 }
 
@@ -79,7 +85,10 @@ impl EntropyAnnexBuilder {
     pub(crate) fn new(n: usize, path: &Path) -> io::Result<Self> {
         let file_size = HEADER_SIZE + n * 4;
         let file = OpenOptions::new()
-            .read(true).write(true).create(true).truncate(true)
+            .read(true)
+            .write(true)
+            .create(true)
+            .truncate(true)
             .open(path)?;
         file.set_len(file_size as u64)?;
         let mut mmap = unsafe { MmapMut::map_mut(&file)? };

src/obikphylo/src/siblings/iter.rs

@@ -21,8 +21,10 @@
 //! | batch      | [`SiblingBatchIter`]  | [`MinorantBatchIter`]  |
 //!
 //! No separate "enumerate" variant (unlike `KmerIter`/`enumerate_kmers`):
-//! [`SiblingEntry`] already carries `order` for free, since pairing with
+//! pairing with the annex is a plain positional `zip` internal to
-//! the annex requires it anyway.
+//! [`SiblingIter`] (a running counter indexing straight into the annex's
+//! mmap) — nothing a consumer needs to see, so [`SiblingEntry`] doesn't
+//! carry it.
 
 use std::sync::Arc;
 
@@ -31,19 +33,17 @@ use obikseq::CanonicalKmer;
 
 use super::{FamilyMask, SiblingAnnex};
 
-/// One layer entry: a k-mer's position in the layer's iteration order (the
+/// One layer entry: a k-mer and its family mask, paired positionally with
-/// same index the sibling annex is keyed on — not an MPHF slot), the k-mer
+/// the sibling annex (see [`SiblingIter`]'s own docs for how).
-/// itself, and its family mask.
 #[derive(Debug, Clone, Copy)]
-pub(crate) struct SiblingEntry {
+pub struct SiblingEntry {
-    pub order: usize,
     pub kmer: CanonicalKmer,
     pub mask: FamilyMask,
 }
 
-/// Streams `(order, kmer, mask)` triples for one layer, in iteration order.
+/// Streams `(kmer, mask)` pairs for one layer, in iteration order.
 /// Produced by [`SiblingLayerExt::iter_siblings`].
-pub(crate) struct SiblingIter {
+pub struct SiblingIter {
     kmers: KmerIter,
     annex: Arc<SiblingAnnex>,
     order: usize,
@@ -61,7 +61,7 @@ impl Iterator for SiblingIter {
             // docs) — shouldn't happen against a fully-built annex, but
             // skip rather than misalign the two streams if it does.
             if let Some(mask) = self.annex.get(order) {
-                return Some(SiblingEntry { order, kmer, mask });
+                return Some(SiblingEntry { kmer, mask });
             }
         }
     }
@@ -69,7 +69,7 @@ impl Iterator for SiblingIter {
 
 /// Batches of [`SiblingIter`]'s entries, `batch_size` at a time — the last
 /// batch may be shorter. Produced by [`SiblingLayerExt::iter_siblings_batch`].
-pub(crate) struct SiblingBatchIter {
+pub struct SiblingBatchIter {
     inner: SiblingIter,
     batch_size: usize,
 }
@@ -85,7 +85,7 @@ impl Iterator for SiblingBatchIter {
 /// Like [`SiblingIter`], filtered to the minorant of each family — the
 /// common case, since a family is tallied once, at its minorant. Produced
 /// by [`SiblingLayerExt::iter_minorants`].
-pub(crate) struct MinorantIter {
+pub struct MinorantIter {
     inner: SiblingIter,
 }
 
@@ -100,7 +100,7 @@ impl Iterator for MinorantIter {
 /// Batches of [`MinorantIter`]'s entries, `batch_size` at a time — the last
 /// batch may be shorter. Produced by
 /// [`SiblingLayerExt::iter_minorants_batch`].
-pub(crate) struct MinorantBatchIter {
+pub struct MinorantBatchIter {
     inner: MinorantIter,
     batch_size: usize,
 }
@@ -131,7 +131,7 @@ fn collect_batch<I: Iterator>(inner: &mut I, batch_size: usize) -> Option<Vec<I:
 /// (`LayerData`) rather than implemented once per matrix kind: kmer
 /// iteration doesn't depend on the data payload, and `TypedLayer<D>` already
 /// carries `iter_kmers`/`hash_batch`/`fill_sub_matrix_carries` for every `D`.
-pub(crate) trait SiblingLayerExt {
+pub trait SiblingLayerExt {
     /// Zip this layer's k-mers with their sibling-annex entry, in iteration
     /// order. `annex` must have been built from this same layer (its length
     /// must match the layer's k-mer count).

src/obikphylo/src/siblings/mod.rs

@@ -20,14 +20,18 @@
 //! reverse.
 
 pub mod algorithms;
-pub mod extensions;
 mod entropy_annex;
+pub mod extensions;
 mod helpers;
-mod iter;
+pub mod iter;
 mod siblingannex;
 
-pub(crate) use entropy_annex::{ENTROPY_ANNEX_FILE_NAME, EntropyAnnex, EntropyAnnexBuilder};
+pub(crate) use entropy_annex::ENTROPY_ANNEX_FILE_NAME;
-pub(crate) use siblingannex::{FamilyMask, SiblingAnnex, SiblingAnnexBuilder};
+pub use entropy_annex::{EntropyAnnex, EntropyAnnexBuilder};
+pub use iter::{
+    MinorantBatchIter, MinorantIter, SiblingBatchIter, SiblingEntry, SiblingIter, SiblingLayerExt,
+};
+pub use siblingannex::{FamilyMask, SiblingAnnex, SiblingAnnexBuilder};
 
 pub use algorithms::{
     BasePairTally, CardinalityTally, EntropyBias, RawSnpDistanceOutput, SankoffBundle,