fidelity-notes.md

RDP, GENECONV, BootScan, MaxChi, CHIMAERA, SISCAN, 3SEQ, and PHYLPRO fidelity notes

Supplied-source mapping

The scanner follows the active RDP5 path reached from the supplied VB analysis-list loop and AlistRDP4/FastRecCheckP. No alternate implementation was used.

Supplied routine or behavior	Port location	Treatment
FillFSSRDP and compressed equality lookup	alignment.cpp, RdpScanner::build_profile_on	Direct byte-state equality preserves the three information-rich-site categories
FindSubSeqPB3 / FindSubSeqPB6	build_profile_on	Excludes missing/ambiguous triplet sites and records pair support plus alignment coordinates
XOHomologyP2	compute_rolling_counts	Uses 2 × floor(window/2) + 1, including circular padding
Average-homology ordering and cycle rotation	ranked_pairs, scan_triplet	Scans middle/low alternatives and applies the additional source rotation below 0.7
FindNextP	append_candidate_signals	Candidate support must exceed both alternative tracks
DefineEventP2	append_candidate_signals	Maps runs to tracts, respects circular/linear termination, and retains the primary EN != BE acceptance gate
ProbCalcP2	binomial_tail, rdp_probability	Log-space binomial tail, opportunity factor, 169-site scaling, and 1e-300 floor
MakeMCCorrection	refresh_active_sequences, frozen per-signal correction_tests	Established from the initial valid scan plan, capped at supplied (255^4)/2, then held fixed while fragment-expanded later-round workloads remain separate
Manual §4.2 / MakeAnalysisListQvR	reference-group UI, refresh_active_sequences, current_triplet, advance_reference_pair	0 denotes a query; positive IDs denote reference groups. The UI supports filter-aware bulk assignment and the supplied auto-group path's final first-input-order compaction. The scheduler retains reference-pair outer/query inner order, admits only one query plus two cross-group references, leaves recombinant inference unconstrained, and preserves the separate group-pair × query correction without materializing AnalysisList
AutoMaskmnu_Click	make_suggested_mask	Iterative closest-pair pruning and diversity-preserving representative choice
EnableAllMnu_Click / MaskAllMnu_Click / DisableAllMnu_Click	dataset bulk curation controls	Replaces every row state in one operation and recomputes the primary opportunity count
MaskSeq states 0/1/2	sequence-state UI, ScanOptions::mask/disabled, active/trace/late/tree paths	Enabled enters every screen; masked skips primary only; disabled skips primary and secondary evidence; all remain bounded tree context
SaveAlign modes 0/5/6	loaded-alignment curation FASTA routes	Full, masked-or-disabled, and enabled row subsets are available before scanning and preserve aligned source rows
FindSubSeqPP / PXoverD	phylpro.cpp, lazy event-review endpoint	Builds the supplied intended polymorphic-column map, rolls paired half-window distance vectors, and displays only the current recombinant/parent roles; the encoded 66/68/72/85 A/C/G/T reset is preserved, while reversed compact indexing in the supplied snapshot is explicitly repaired
MakePDstMat / UpdatePDstMat / PPRegression	TargetWindows, add_position, source_pearson	Three target rows replace the unused all-pairs matrix with identical target coefficients and O(LN) work; source double sums, float output, missing comparison, self option, and zero-variance result are retained

PHYLPRO is a review method, not another discovery stream. The supplied manual/source route does not implement its proposed permutation significance test, so the browser never assigns a PHYLPRO p-value or inserts a PHYLPRO signal. Circular windows follow the literal wrap path; linear analyses use complete half-windows only. See native-phylpro-review-trace.md for the compact-index repair, manual/source column-selection divergence, context rules, and optimization proof boundary.

Cyclic discovery and fragment state

The supplied workflow does not retain every first-pass hit as an independent final event. The port now performs complete strongest-first cycles:

1. Screen every eligible working triplet with enabled RDP, GENECONV, BootScan, MaxChi, CHIMAERA, SISCAN, and 3SEQ discovery and retain significant method-labelled signals for that round.
2. Select the lowest corrected probability, then local probability, then stable signal order.
3. Group its detectable support using two shared original identities and strict > 0.3 symmetric tract overlap.
4. Build all three role hypotheses and select the current complete co-recombinant group.
5. Erase the event tract from that group and begin a fresh complete triplet pass.
6. Stop when no unexplained signal remains, no new working sites can be erased, or fewer than three distinct active origins remain.

Below the supplied < 100000 alignment-length condition, each erased tract also becomes a gap-padded working fragment. working_origins_ and working_fragment_events_ preserve identity and event provenance. Two working copies of one original can never occupy the same triplet. Copies with fewer than max(5, window, ceil(length/100)) usable states and exact same-origin duplicates are discarded. A visible 256-fragment cap bounds browser memory and combinatorial growth; reaching it is reported rather than hidden.

Fragment-assisted signals are mapped back to original sequence IDs for event reporting. During downstream evidence calculation, a saved fragment context selects the matching working copy with the most usable event-tract sites. Other fragment aliases are not all placed in each tree panel; that is a declared performance adaptation.

In query-vs-reference mode, working fragments inherit the query/reference group of their original sequence. The exact schedule counts every valid working reference-record pair and query record, while the correction counts active reference-group pairs and unique query origins. This keeps same-origin fragment aliases from multiplying the statistical opportunity count, an explicit browser cyclic-state interpretation that requires a native fragment fixture because the supplied MakeAnalysisListQvR code uses its current QNum desktop record count.

Secondary event reconciliation

Supplied routine or rule	Port location	Treatment
Strongest remaining probability	finish_detection_round	Corrected p, local p, then stable signal ID choose the next unexplained anchor
Two shared triplet sequences	triplet_pair_keys, round pair index	Each signal is indexed by its three unordered original-sequence pairs
FillSetsP* overlap	tract_overlap	Symmetric 2 × intersection / (size1 + size2), strict > 0.3, including origin-spanning tracts
FindSets role closure	refresh_role_hypotheses	Builds three direct detectable sets and repeatedly adds a sequence to the remaining role when the other two contain it
Masked follow-up	refresh_trace_evidence	Rechecks each masked sequence; trace-only and corrected-significant calls remain distinct
MakeBPosLR, VSN = 60	correlation_boundary, build_correlation_regions	Four circular walks with source boundary inclusion/exclusion
MakeSDMP2	correlation_region_profile	Parent-one, parent-two, and parent-agree/candidate-differs fractions over five regions
FillRmat	correlation_vector	Two breakpoint pairs plus outside-average/tract six-value pair
CalCR Pearson and relabelling	pearson_six_two_sided_p, best_category_correlation	Direct correlation, three pair swaps, and both cyclic relabellings; cycles share native class four
RCorrWarn	correlation_warnings, distance-triangle guard	Dominant-category warnings plus the source distance-triangle branch and XOR tract/outside warning behavior
MakeGoodC	breakpoint_overlap_sites	Strictly more than ten non-gap candidate sites across either breakpoint span
MakeINList / MakeACOR	source_in_list, acceptable_affinity	Closest outside/inside representative pair maps to the active topology-affinity inequalities; raw midpoint-ultrametric topology rank is preferred with JC fallback
MakeRList	distance evidence finalisation	Preserves the exact first-two-correlation corc == 2 override quirk and warning-adjusted positive P-score aggregate
StripDupInv	stripped_inverse_only final membership	Inverse r > 0.83 remains diagnostic, while active inverse-only rows are removed from the distance set
Opening FinalTrim duplicate pass	duplicate_filtered, duplicate_cleaned_support	Warning/inversion-cleared direct pairs above 0.83 are counted across all three role lists; once a sequence/pair count exceeds one, that pair is suppressed for every occurrence in every role list, matching the source's pair-wide second pass
FinalTrim OKSeq 6	nearest-membership fixed-point block	Reconstructs ascending MakeRList order plus StripDupInv swap-last deletion, then repeats raw/collapsed nearest-nonrecombinant limits, strict 0.83/0.95/0.99 correlation tickets, unfound-event deletion, and paired four-breakpoint JC veto until the source last-index stop condition holds
FinalTrim OKSeq 7/8	FinalTrimMatrixEvidence tree-position scores	Applies the source's four anchor/parent and candidate/parent comparisons to collapsed and raw outside/inside topology-rank matrices, including strict/tied rewards and role modifiers
FinalTrim OKSeq 9/12/13	FinalTrimMatrixEvidence distance scores	Ports whole-tract and both breakpoint-pair JC branches, native warning/<3 gates, positive downweighting, weaker-positive suppression, and negative quarter/half/full penalties; scores feed active CScore
FindActualEvents / MakeMatchMatX2P, OKSeq 14	SourceDetectedTractGrid, match-matrix prefix path	Uses the active direct-event catalogue, pre-inversion-strip membership, inverse-parent gate, strict >1/3 overlap and best tract, circular/intersected JC distance with native saturation, all signed branches, and the source's bare-CompMat index lookup
CheckPatternX, OKSeq 3	source_pattern_scores	Counts source 1/0.5 informative-state patterns across both breakpoint spans and the tract, then emits each role's share with its representative sentinel
FinalTrim ordered expansions / OKSeq 15	active final-list block	Retains the fixed-point nearest-nonrecombinant thresholds, appends correlation-gated candidates in ascending order, applies strict raw-tree parent bounds in the second pass, then preserves selected-role swap-last pruning and its inherited third-list loop index before recording OKSeq 15
ConsensusOK completed score	ConsensusScoreEvidence	Maps corrected correlation P, event overlap, set/pattern/pre-strip/duplicate/nearest/final membership, RCorrX, representative sentinel/zero cells, and the active matrix multiplier into the score used for membership
ConsensusOK NS As Long	source_vb_long, completed score	Applies half-to-even narrowing after the averaged 7/8 term and every nonzero 9–14 addition; the native three-assignment sub-one branch narrows its final 2^-1 value to zero. The UI separately shows the readable raw subtotal
CalcMatchY OKSeq 17/18	source_calc_match, CalcMatchEvidence	Preserves source flank reconstruction/bounds, VB half-to-even window selection, signed match map, circular single-precision smoothing, regional product, six checkpoint samples, standard breakpoint thresholds, and the opening ConsensusOK raw-tree topology filter; available values are active grouping inputs
ConsensusOK list rebuild	primary/equivalent/straggler membership block	Clears all three lists, applies score/regional/class thresholds and collapsed/raw topology fallbacks, snapshots each widening pass, preserves exact six-distance equality and direct/raw straggler comparisons, and restores all old lists if any rebuilt role is empty
Shared selected-role conservative cleanup	post-rebuild tree cleanup block	Accumulates direct whole-region movement sums as Single, ranks all three representatives, applies native raw/direct outlier and clade-equality constraints with swap-last removal, then adds bounded or strict four-matrix inliers; preserves the FAMatSmall <= HDF and always-true Or x = x branches
Six event subalignments	build_phylogenetic_regions	Both breakpoint flanks plus whole outside/inside tract partitions
Jukes–Cantor matrices	phylogeny.cpp	Supplied FastBootDistIP6 single-precision valid/difference accumulation; identity ≤0.25 or no comparable sites uses distance 10
Event tree path	build_tree_region_evidence	Installed-5.93 x87-compatible JC storage; literal packed bundled Clearcut float NJ; five-decimal writer and Tree2ArrayP2 parser/midpoint rank matrices; active zero-replicate branch copies raw ranks into the nominal collapsed family
Manual tree displays	event_trees_json, EventTreeInspector	Reuses source-serialized five-decimal saved edges for all six reconciliation trees while identifying the separate analytical rank matrices; compares each region pair and exposes zero-replicate and fragment provenance
Large-set tree bound	select_tree_sequences	All three anchors plus the closest active originals, capped at 100; omitted candidates use labelled JC affinity
CheckBSTree grouping shape	phylogenetic evidence loop	Candidate-anchor distance must beat both anchor-parent and candidate-parent distances in both trees of a pair
Manual “two of three” rule	complete_two_of_three_set	Detectable, distance, and phylogenetic membership are counted independently; two memberships include the sequence
Ordered correction/rejection	update_event, set_event_review_state, reconcile_after	Accepted corrections are re-erased; rejected tracts are restored; fixed earlier events remain while all later events are rediscovered
Manual group repair	update_event_group	Keeps automatic and current groups separate; forces the recombinant, excludes current parents, and invalidates the downstream chain
Mid-repair save/reload	worker project restore	Drops stale later events/signals, compacts anchor IDs, restores the valid prefix, and preserves the first-changed-event marker
Manual sequence display	event_alignment_json, EventAlignmentInspector	On-demand immutable-alignment windows around both breakpoints prioritize current roles and relevant group/evidence rows, wrap circular coordinates, bracket expected major/minor informative-state transitions, and colour parent/recombinant patterns without moving full sequences to the UI
Native RDP CheckEnds uncertainty	breakpoint_uncertainty, uncertain_erasure_event_ids, event/alignment JSON, CSV, and UI warning	Reconstructs current triplet information positions after prior erasures, the source's ten-character input MissingData baseline, distinct beginning/ending DLL ranges, strict linear edge gates, and literal wrap comparison; erased-event attribution and immediate contact remain separate
BURT / PolishBP statistical confidence	burt_confidence.cpp, settings, result/alignment JSON, CSV, and review card	Full supplied three-state Viterbi/posterior/range/matching/repositioning chain is active with a local seeded Microsoft-CRT random adapter; the supplied flag defaults on, disabled runs preserve coordinates, and parent-transition brackets remain separate review-only evidence
MaxChi MCXoverF exploratory discovery	maxchi_discover, combined cyclic scheduler, method-aware JSON/CSV/review	Three variable-site pair profiles, native critical gate and raw tie order, source-shaped growth, FindSide, optimized left/right breakpoint, tract construction, literal SmoothChiValsP, completed/rejected peak destruction, and three-wasted/100-attempt bounds are active; a heap replaces repeated full-profile maximum scans while preserving intended peak priority
CHIMAERA AlistChi / FastRecCheckChim / CXoverA exploratory discovery	chimaera_discover_prepared, combined cyclic scheduler, method-aware JSON/CSV/review	Each triplet member becomes the target in turn; cached pair equalities form the source information-rich binary string; one rolling χ² profile per target enters source-shaped growth, flank choice, optimized tract construction, literal smoothing/destruction, three-wasted stop, and a per-target 100-attempt cap. Target and provisional parents remain subject to shared late role consensus
GENECONV FindSubSeqGCAP6 / GetFragsP / GetMaxFragScoreP / CalcKMaxP / GCCalcPValP2 / GCXoverD exploratory discovery	geneconv_discover_prepared, combined cyclic scheduler, method-aware JSON/CSV/review	Cached pair equalities form three inner and three outer signed tracks. Source mismatch penalties, strict critical score, lambda/K and KA tail branches, stable lowest-P ordering, ignored indels, provisional roles, and overlap allowance are active. Discovery uses the active triplet/analysis-plan MCCorrection, retaining raw KA P separately; the distinct pair-count GCMCCorrection display path remains a validation boundary. Prefix/range queries replace quadratic extension; a bounded bisection fallback is used only when native Newton iteration is numerically unsafe
SISCAN SSXoverC / GetSSOL / Get3Score / GetPScores2 / DoPerms3 / MakeZValue2 / DoSums / FindMaxZ / ShrinkRegionC discovery	siscan_discover, combined cyclic scheduler, method-aware JSON/CSV/review	Preserves the nearest distinct-origin fourth sequence on a round-cached source WPGMA/cophenetic context, gap-stripped default pattern categories, exact category group tables, the QuickCheckB control-flow quirk, Microsoft-CRT MakeVRand flat-prefix vertical permutations, pair-switch regions, source NormalZ, and every length/window/project probability adjustment. Primary discovery is optional and native-golden unvalidated
3SEQ FindSubSeqTS / Seq3PVals / GetTSPVal / CheckwrapC / TSXOver exploratory discovery	threeseq_discover_prepared, combined cyclic scheduler, method-aware JSON/CSV/review	Rotates targets in supplied 3,1,2 order; filters three ±1 information-rich walks from the cached pair equalities; chooses the strictly lower-P orientation; applies source low-information and zero gates; and uses exact finite hypergeometric tails with an explicitly labelled supplied SiegmundDiscrete fallback. Circular/linear CheckwrapC, method-specific Dunn–Šidák correction, and later-round FindSubSeqTS2/CheckSplit3Seq/SubPVal missing-run trim, reverse-orientation retry, and corrected-P re-gate are active. Manual permutation envelopes remain a boundary
MaxChi FastRecCheckMC2 strongest-peak recheck	maxchi_recheck, event/list result evidence, CSV, and review cards	The same profile family with native missing/erasure and linear-edge bans, strict critical-difference gate, rolling χ² maximum, ChiPVal2P, source-shaped growth, and raw/within/project probability scopes remains a separate corroboration pass; it does not reposition an already reconciled event
GENECONV ordinary-kernel representative/final-list recheck	geneconv_recheck_prepared, event/list result evidence, CSV, and review cards	Reuses MaxChi's prepared triplet equality/missing-data workspace, applies the active six-track KA/overlap route, and retains the stable best fragment, provisional recombinant, raw/corrected P, skew/workload/overlap/fallback evidence without changing reconciled coordinates. This does not claim permutation, manual, alternative-indel, or full late event-reconstruction parity
SISCAN fixed-region representative/final-list recheck	siscan_recheck, event/list result evidence, CSV, and review cards	Reuses the round WPGMA context and seeded flat random prefix, scores the reconciled event region with the supplied final permutation count, and retains outlier, pattern count, score family, Z, every probability stage, and hit status without changing coordinates
3SEQ TSXOver(1) representative/final-list recheck	threeseq_recheck_prepared, event/list result evidence, CSV, and review cards	Reuses the compact target walks and missing/erasure map; preserves the initial selected-orientation gate, forces CheckSplit3Seq for both orientations under FindallFlag = 1, applies the post-swap correction branch, and counts each accepted record plus its supplied inverse-parent/inverse-interval copy. The best retained orientation is audit evidence only and does not change reconciled coordinates
Manual alignment outcomes	FASTA export methods	Accepted group tracts are gapped; fragment export appends aligned event fragments in event order

Project JSON identifies this as reconciliationTier: "detectable-distance-phylogenetic" and retains individual sets, complete groups, discovery method and MaxChi/CHIMAERA/GENECONV/SISCAN/3SEQ trace, round-specific correction factors, fragment provenance, tree-panel/fallback status, bootstrap summaries, correlation gates, and role metrics.

Role recommendation boundary

The role panel exposes nine inspectable metrics:

Metric	Native contribution represented here
PhPr	JC profile correlation, weight 8
TreePhPr	Raw-tree profile correlation, weight 18
CollapsedTreePhPr	Collapsed-tree correlation with the native SubPhPr condition, weight 20
SubPhPr	Leave-one-role-out JC correlation, weight 10
TreeSubPhPr	Leave-one-role-out raw-tree correlation, weight 8
SubDist	JC outside/inside displacement, weight 2
TreeSubDist	Raw-tree outside/inside displacement, weight 10
TrpScore	Weighted triplet-ordering change, weight 8
ThreeSetSupport	Display-only group-size context, weight 0

The mapped voting metrics use the supplied full contribution for a best role and half contribution for a role better than one alternative, with the collapsed-tree special condition handled separately. Confidence is the top-minus-second contribution margin divided by all contributions. The full desktop method battery is not present, so JSON deliberately keeps nativeWeightParity: false.

Representation and performance choices

• Internal vectors are zero based; emitted alignment coordinates remain one based.
• A/C/G/T/U are unambiguous byte states. Gaps, ambiguity, and missing markers are excluded.
• Probability tables become direct log-space evaluation.
• Per-triplet buffers, breakpoint layouts, overlap counts, and reference distances are reused.
• Scans yield after 512 working combinations; same-origin combinations skip the numerical path.
• Query/reference mode retains only active query and reference index vectors plus three cursors; cross-group triplets are generated lazily rather than stored in the supplied growing AnalysisList(2, T) array.
• Sequence curation preserves the manual's three states: only enabled rows enter primary triplets, masked rows can enter secondary evidence, disabled rows cannot enter event roles/groups, and all three states remain candidates for the bounded phylogenetic panel.
• The supplied CheckQueryReference auto-assignment clears MaskSeq. The browser keeps reference assignment orthogonal to the visible enabled/masked/disabled editor and never silently unmasks a curated row; users can explicitly choose “Enable all” when they want the desktop side effect.
• Fragment retention is bounded at 256 and event tree construction at 100 sequences.
• Tree distance matrices load a column's states once before updating all pairs.
• Six tree families are built once per event and reused across all role hypotheses and metrics.
• Graphical breakpoint context is generated only when opened and is capped at 64 rows, 100 sites on either side, and two windows; no full sequence row crosses the worker boundary.
• CalcMatchY reconstructs no more than three alignment lengths, retains no more than 160 variable sites, and reuses its match/smoothing scratch vectors for every ordinary candidate.
• MakeMatchMatX2P builds three valid/difference prefix pairs in O(NL) and answers all role/tract intersections without the source's repeated whole-alignment rescans; CheckPatternX is O(NL).
• OKSeq 6 reuses saved tree/JC matrices. Its order-sensitive fixed point can be O(N²) distance checks in the one-removal-per-pass case but performs no new alignment scan or tree construction.
• MaxChi keeps rolling left/right totals for all three pair tracks, so profile construction is O(V) rather than O(V × H). Discovery heapifies positive raw profile cells once and lazily discards cells erased by smoothed peak basins; the independent confirmation path retains only its strongest peak. The native whole-triplet maximum gate precedes smoothing and retry accounting.
• CHIMAERA filters each target profile from the same cached equality/coordinate pass, uses one capacity-retaining target workspace, and heapifies only positive raw cells. This preserves the mandatory three rotations without multiplying alignment-byte reads or allocating a profile per candidate.
• GENECONV constructs all six signed run lists in one category pass from that cached input. It preserves the supplied circular initial-run plus terminal-wrapped-run ordering and only the source's positive-outer coalescing. Next-lower prefix bounds and rightmost range maxima preserve the negative-stop/latest-tie semantics in O(R log R); stable lowest-P overlap selection uses lazy range coverage in O(F log V). The on-demand plot uses an interval sweep rather than fragment-by-position painting.
• 3SEQ filters all three target walks from that same equality profile in O(V). Exact tails use a bounded rolling survival-state dynamic program, narrow to the supplied Single result, and are cached by (plus, minus, excursion) across triplets; the 8192-entry cache and eight-million- transition per-evaluation ceiling bound memory and latency. Larger states take the supplied Siegmund route, and all three plot walks are reconstructed only on demand.
• SISCAN constructs direct and WPGMA/cophenetic matrices once per affected round, then reuses them for all nearest-outlier queries. Its seeded vertical-randomization template is an extensible flat prefix reused across triplets and later rounds; only the context is invalidated after erasure. Existing unchanged-triplet shortlist replay bypasses the method entirely.

Known parity risks requiring golden comparison

1. Informative-site upper-bound conventions can create one-site boundary or probability deltas.
2. Average-homology tie perturbations use stable pair identities rather than every native lookup table.
3. Circular tracts at coordinate one require focused DefineEventP2 comparison.
4. The erasure helper now follows active ModSeqNumY's inclusive linear/wrapped loops, including the source's full-alignment traversal when both endpoints are equal; exact behavior still requires a native golden fixture.
5. Text readers normalize metadata and do not cover every desktop format.
6. Dot-gap and missing-data arrays are separate natively; compact state zero treats them uniformly.
7. Exact FAMat/SMat categorical normalization around MakeACOR can change marginal affinity calls.
8. Full contradictory-inversion handling beyond active StripDupInv remains incomplete.
9. The active FinalTrim/ConsensusOK grouping path through shared selected-tree cleanup and the primary-RDP post-group recheck are ported but unvalidated. Exact floating-point equality branches, literal sequence-zero fallback, collapsed-matrix availability boundary, Single movement ranks, list ordering, empty-role restoration, and the browser mapping of the native widened LowestProb threshold need focused native fixtures.
10. The 100-sequence tree panel, 256-fragment cap, and omission of non-anchor fragment aliases from event trees are explicit browser adaptations.
11. The source-shaped Clearcut tie/negative-branch behavior, split support, and ranked matrices have a deterministic host regression but still require authorized native golden tree fixtures.
12. Several native recombinant-identification method families remain absent despite the mapped weights for the implemented subset.
13. The manual's RDP-specific uncertainty flag, supplied CheckEnds range construction, active inclusive ModSeqNumY erasure, and distinct BURT/PolishBP path are implemented but unvalidated. Fragment-to-original role mapping, the local Microsoft-CRT random adapter, odd-length circular expansion, signed ranges, accumulated missing-data repositioning, and final active source quirks require native fixtures. Other breakpoint probability-distribution families remain outside this primary-RDP slice.
14. Project replay (including pending-prefix compaction), rejected-event tract restoration, manual group repair, final-export gating, and cyclic termination need native fixtures.
15. Enabled/masked/disabled replay and the exact influence of disabled tree-context rows on native late distance pruning need focused fixtures, especially when the 100-row browser tree cap binds.
16. MaxChi discovery and confirmation are source-shaped but unvalidated. Native lookup-table rounding, zero/length BanWin aliases, MaxX = 0, the literal smoothing off-by-one, missing- boundary growth ordering, left/right optimization indices, destruction basins, accepted-region inclusion, retry counters, preliminary role assignment, and RDP/MaxChi cross-method ordering all need golden comparison. The bounded port deliberately avoids the source routines' unsafe out-of-range destruction reads. Permutation/manual-doublet modes remain unported.
17. Compact projects retain signal statistics and fragment provenance but not every full historical working-profile vector. On-demand later-round graphs therefore reconstruct the original triplet and label that context; golden comparison should use exported scalar evidence until historical profile persistence is deliberately added.
18. Automated query-vs-reference scheduling follows the active MakeAnalysisListQvR triplet constraint and initial ordering, but needs native golden comparison for masked or empty groups, fragments added after erasure, configured-versus-active RefNum, correction factors, and cases where a reference is ultimately the recombinant. The editable REF-<group> parser implements the manual's documented naming convention but does not claim parity with every heuristic branch in the desktop CheckQueryReference auto-grouping routine. Its deliberate no-silent-unmask behavior must also be represented explicitly in workflow validation.
19. CHIMAERA discovery is source-shaped but unvalidated. Target-map indexing, FindSubSeqDP3/6 position aliases, window fallback, missing-boundary mapping, equal raw peaks, growth and tract side, boundary optimization, destruction basins, inside/outside parent-one role ordering, and combined RDP/GENECONV/BootScan/MaxChi/CHIMAERA/SISCAN/3SEQ event order require native golden comparison. The three-target strongest-peak representative/finalized-list recheck is active and auditable but unvalidated; CHIMAERA permutation and full late event-reconstruction modes remain unported.
20. GENECONV ordinary triplet discovery is source-shaped but unvalidated. Compressed-table ambiguity behavior, circular first/last runs, equal score/P ordering, float narrowing, Newton edge cases, coordinate aliases, the supplied sibling-path ten-visit repeated-P escape and terminal-sentinel endpoint behavior, the overlap-array indexing quirk, six-way role mapping, and combined seven-method event order require authorized native fixtures. Exact P-value ties use the source's RDP → GENECONV → BootScan → MaxChi → CHIMAERA → SISCAN → 3SEQ method-major priority. The port interprets overlap coverage over polymorphic positions and bounds an otherwise unbounded Newton loop; both adaptations remain explicit. The same ordinary kernel now supplies auditable, non-coordinate-changing representative/final-list corroboration. Permutation probabilities, manual pairs, alternative indel modes, minimum-filter variants, and full native late event reconstruction remain unported.
21. 3SEQ ordinary discovery is source-shaped but unvalidated. The target/pair-slot rotations, four-site floor, exact-table equivalence and Single narrowing, eight-million-transition cutoff, scaled-exact fallback, strictly lower-P orientation swap, pre-CheckwrapC probability versus post-wrap boundary excursion, 1e-15 Dunn–Šidák/product branch, low-information exits, circular prefix extension, beginning increment, and linear complement mapping require authorized native fixtures. The active later-round walk excludes erased/missing states and applies the supplied inclusive FindSubSeqTS2 map plus CheckSplit3Seq/SubPVal trim and re-probability step; its endpoint aliases, wrapped missing runs, and reverse-orientation retry still need golden comparison. The representative/final-list TSXOver(1) Findall path is active with both split orientations and inverse list copies, but its list-cap/event-catalogue interaction still needs golden comparison. Permutation display envelopes and full late event reconstruction remain unported.
22. Primary distance-mode BootScan is source-shaped but unvalidated. SEQBOOT2's two warm-up draws, generated-but-discarded tail replicate, strict closest-pair ties, padded circular windows, FindBeginBS/FindEndBS gap and edge aliases, provisional FindDaughter roles, XPosDiff invariant-boundary indexing, VB6 rounding, 169-site scaling, and ProbCalc accumulation require authorized native fixtures. The browser's log-domain binomial summation is a deliberate numerical-stability improvement over factorial products and may preserve a positive tail after the desktop arithmetic underflows. The 64 MiB pair-profile cache changes reuse and storage, not the intended pair/window values; it is invalidated after every erasure. Tree/similarity/permutation/manual BootScan modes and full late catalogue reconstruction remain unported.
23. SISCAN discovery and fixed-region confirmation are source-shaped but unvalidated. Authorized fixtures must cover WPGMA tie resolution, missing/disabled/fragment outlier eligibility, QuickCheckB jumps, exact partition and summed category maps, MakeVRand bytes, population variance, zero-variance behavior, final-pair selection, ShrinkRegionC aliases, roles, linear window endpoints, and all four probability stages. The browser's same-origin outlier exclusion makes the supplied TraceSub intent explicit, and double arithmetic may retain a tiny positive P-value where a desktop intermediate underflows. Random/most-distant outlier, alternative gap and category modes, and manual SISCAN remain unported. The browser review plot deliberately condenses the manual's 15 category plus 9 sum curves into three signed sister-pair curves by retaining the greatest-absolute eligible Z per pair/window; the complete manual display is still an explicit boundary.

Until those cases are validated, the project provides a complete review/export path for combined ordinary-triplet RDP, GENECONV, BootScan, MaxChi, CHIMAERA, SISCAN, and 3SEQ discovery plus the seven active late corroboration routes, but does not claim native result parity.