This trace began with the Session 8 primary-RDP translation and now records the MaxChi, CHIMAERA, GENECONV, and 3SEQ corroboration addenda, using only the supplied RDP4/RDP5 DLL and VB sources. It is an audit aid, not a claim of runtime parity. No alternate implementation was used.
The active VB orchestration in Module3.bas has two mutually exclusive RFF=0 shapes:
- When all three role-compatibility results are available,
RetrimFlagis set near lines 45340–45343 andFinalTrimis called near line 45390 withRFF = 0andRWinPP = 4. That pass computes the late evidence without selected-role pruning. A later call near line 45755 usesRFF = 1after role selection. - Otherwise the first call is skipped. The call near line 45755 receives
RFF = 0and the selectedRWinPP < 3, so it performs the complete RFF=0 path including selected-role pruning.
The browser collapses those two orchestration shapes around its explicit role reorder.
refresh_role_hypotheses calculates the role vote, the event is reordered if that vote changes the
recombinant, and the routine is rerun with role zero as the current selected role. The complete RFF=0
list build runs there, followed once by the selected-role block shared below the RFF guard. The
primary-RDP part of the later signal/probability revalidation then runs over the finalized lists.
The browser also exposes bounded MaxChi, CHIMAERA, GENECONV, and 3SEQ corroboration over the same
finalized rows; full method-origin dispatch and global event-catalogue integration remain a
documented boundary. Together with primary RDP, these are five active method-family rechecks.
The supplied late path does not use the public display order for its three representatives.
ISeqs is daughter, minor parent, major parent. Session 25 restores that internal ordering in
FinalTrim/ConsensusOK and BURT inputs while leaving JSON, UI, CSV, and project roles in the
public recombinant, major parent, minor parent order.
After cyclic erasure, TraceSub resolves original identities to the retained working fragments
that produced cached XOverList evidence. The port now retains exact working-triplet provenance in
the signal shortlist, assigns strongest-support aliases for all original candidates, and gives the
anchor triplet precedence. This is an identity mapping rule, not a cycle-number rule. Restored old
projects without exact working triplets keep the earlier fragment-event lookup.
The selected-role movement/rank block iterates through source NextNo, including working fragments.
Session 25 therefore uses the expanded population and denominator rather than only the original
alignment rows. Candidate membership remains keyed by original identity; comparisons that skip the
role representative use the reported identity rather than its transient working-fragment index.
Module2.bas defines FinalTrim at line 23342 and gates this family with If RFF = 0 at line
23390. The browser mapping is:
| Supplied lines | Native state transition | Port treatment |
|---|---|---|
| 23398–23477 | Copy correlations, clear warned/inverted values, suppress duplicate candidate/pair values, set OKSeq 5 |
raw_correlations, cleaned_correlations, duplicate_filtered, duplicate_cleaned_support |
| 23479–23832 | Snapshot OKSeq 6; repeatedly rebuild nearest-nonrecombinant lists and remove rows until the last index stabilizes |
nearest_lists fixed-point block, preserving MakeRList ascending order and StripDupInv swap-last order |
| 23835–23868 | Mark current rows and all representatives; append ascending candidates inside both retained nearest-neighbour limits with any cleaned r > 0.83 |
first_expansion_added and ordered finaltrim_lists append |
| 23873–24300 | Populate OKSeq 7–14, including explicit-zero 10/11 and event-region 14 |
FinalTrimMatrixEvidence, SourceDetectedTractGrid, and match-matrix prefix calculations |
| 24303–24323 | Append remaining ascending candidates strictly closer to the anchor than both parents in both raw whole-region trees | second_expansion_added |
| 24331–24420 | If closest pairs differ and a role is selected, prune the three mapped lists with collapsed/raw thresholds and an all-three-r < 0.99 removal ticket |
active selected-role block and selected_role_pruned |
| 24425–24429 | Mark surviving rows as OKSeq 15 |
finaltrim_membership |
| 24430–24442 | Build RCorrX as the maximum un-warned, non-inverted correlation |
maximum_direct_correlation |
| 24444–24459 | Run CalcMatchY, then clear/rebuild lists through ConsensusOK |
source_calc_match, topology class grid, completed score and three rebuild passes |
Two order-sensitive details are intentionally retained:
- List removal copies the final row into the removed position. The port does not replace this with stable filtering before the next native stage.
- After pruning
INList(1), the source does not resetYbefore theINList(2)loop at line 24393. The third loop therefore begins at the final size of the second list. The port preserves this inherited index.
The collapsed first-list inside comparator at line 24340 is spelled
SCMatSmall(INList(1), ISeqs(INList(0))); the fallback spelling uses the opposite row. The matrix is
symmetric, but the port retains the active spelling in code and documentation.
ConsensusOK begins at line 22079.
| Supplied lines | Native behavior | Port treatment |
|---|---|---|
| 22120–22177 | Score nonrepresentatives from OKSeq 0–6, multiply by 2 for 15, apply RCorrX, then the NS As Long matrix sum; own representative is 1000 and the other two are zero |
ConsensusScoreEvidence.final_score; source_vb_long narrows after every native assignment |
| 22178–22187 | Save all three lists and clear them | finaltrim_lists retained while consensus_lists start empty |
| 22207–22297 | Reject inconsistent OKSeq 18 rows using both raw whole-region trees |
consensus_match_class topology pass over every ordinary sequence |
| 22313–22483 | Primary ascending rebuild from CScore, regional match, class, raw dominance, collapsed topology, exact raw position, and zero-distance fallbacks | primary_membership |
| 22488–22534 | Snapshot membership; add exact six-distance equivalents or low-score direct neighbours | equivalent_membership with a fixed pass snapshot |
| 22545–22655 | Snapshot again; collect raw/direct distance stragglers | straggler_membership with a fixed widened snapshot |
| 22705–22715 | If any role is empty, restore all three saved lists | consensus_fallback_restored |
The role-zero exact-position fallback at line 22447 uses literal full-matrix row 0, while roles one
and two use ISeqs(role). The port deliberately retains that literal sequence-zero behavior.
ConservativeGroup is treated as its active default zero; no user option in the supplied workflow
enables the conservative branch.
After the RFF=0 guard closes at line 24473, both native call shapes converge when RWinPP < 3 at
line 24510. The active ConservativeGroup = 0 block is mapped as follows:
| Supplied lines | Native behavior | Port treatment |
|---|---|---|
| 24540–24571 | Sum all direct outside/inside distances into Single arrays and rank each representative by total movement | movement_distance and representative_rank, narrowing every accumulation back to float |
| 24576–24829 | Mark rebuilt-list rows that fail raw/direct outlier, clade equality, sixfold-distance, region-direction, or extreme-rank constraints | selected_tree_cleanup_pruned; all comparisons retain strict/equal source operators |
| 24830–24845 | Remove marked rows by copying the final entry into the current position | swap-last mutation of each consensus_lists row |
| 24866–24918 | Admit candidates within the remaining raw/direct cluster bounds and closer in both raw trees | first selected-tree admission branch |
| 24919–24944 | Admit strict direct+raw four-matrix inliers | unconditional fallback branch; the source condition ends with Or x = x and is therefore always true |
Line 24895 compares outside FAMatSmall with HDF, the maximum direct outside distance. The port
retains that mixed-matrix comparison rather than substituting FMatSmall.
The distance matrix is symmetric. Session 25 evaluates each unordered pair once, adds it to both movement rows, and reuses the completed rows for representative totals. With ascending lower/upper indices, every row still receives contributions in ascending sequence order, so Single narrowing and source rank thresholds are unchanged. Pair-cache calls are roughly halved and six representative re-sums disappear; symmetric caching already prevented duplicate expensive JC calculations.
The native recheck block begins at line 25000. The primary-RDP branch can emit widened-threshold records, while the six other discovery methods currently represented in the browser retain bounded, non-coordinate-changing corroboration from their supplied ordinary statistics:
| Supplied lines | Native behavior | Port treatment |
|---|---|---|
| 25008–25015 | Temporarily widen LowestProb to at least LowP * 100000 and the corrected project threshold |
post_group_local_cutoff = min(1, max(project cutoff, best local P * 100000)) |
| 25087–25214 | For every role list, skip its own representative and rerun the event's originating program against the other two representatives | finalized consensus_lists; the RDP path calls triplet_signals(..., enforce_cutoff=false), while BootScan, MaxChi, CHIMAERA, GENECONV, SISCAN, and 3SEQ retain their separate bounded evidence records |
| 25137–25139 | Dispatch XOver when ProgramFlag = 0 |
primary RDP profile/candidate path; no substitute implementation is used |
| 25220–25353 | On the selected list, run enabled programs other than the event's originating method | distance-mode BootScan, FastRecCheckMC2-shaped MaxChi, three-target FastRecCheckChim, ordinary six-track GCXoverD, fixed-region SISCAN, and three-rotation/two-orientation TSXOver(1) corroboration are retained without inserting new events; full native dispatcher/event reconstruction and unported families remain pending |
| 26092–26120 | Copy emitted records back into the persistent event catalogue | compact per-sequence recheck evidence records emitted/candidate/overlap counts and the best tract/probabilities; the ordinary cyclic scan remains the authoritative global signal catalogue |
The browser distinguishes four things that the native global XOverList representation combines:
all RDP records inside the widened threshold, records naming the list candidate as recombinant,
records overlapping the reconciled event by more than 30%, and records also passing the ordinary
corrected project cutoff. This keeps loosened-threshold traces from being mislabeled significant.
An unavailable information-rich profile is retained explicitly rather than treated as a negative
signal.
The supplied disk-only RDP5ExcludeList substitution path after line 25500 has no browser analogue:
the browser does not offload alignment records to an external exclude file. Every loaded sequence is
already considered in memory, while the existing masked-sequence trace path remains separately
auditable.
The browser now applies the supplied FastRecCheckMC2 statistic to the representative event triplet
and to finalized nonrepresentative distance-list rows. This is intentionally narrower than calling
the complete native program dispatcher in lines 25087–25353: it reports the strongest admissible
MaxChi peak and all probability scopes but does not insert a MaxChi-discovered event into the global
catalogue, reposition the RDP event, or run the native smoothing/DestroyPeakP retry loop.
The line-level numerical mapping and its explicit deviations are recorded in
native-maxchi-recheck-trace.md. Result metadata therefore uses
source-shaped-strongest-peak-unvalidated and keeps nativeMaxChiFullRecheckComplete false.
Session 12 added the supplied three-target FastRecCheckChim statistic over the same prepared
equality/missing-data workspace. Session 13 added the supplied ordinary six-track GENECONV
fragment/KA kernel over that workspace. Both are exposed beside MaxChi for the representative
triplet and every finalized nonrepresentative distance-list row. None of these corroboration passes
moves a reconciled breakpoint or silently writes a new discovery event.
Session 14 added ordinary and cyclic 3SEQ discovery to the shared event scheduler. Session 15 adds
the supplied post-erasure CheckSplit3Seq/SubPVal discovery re-probability and the distinct late
TSXOver(1) representative/finalized-list route. That route preserves the initial selected-
orientation gate, forces split evaluation of both Findall orientations, applies the post-swap
correction branch, and accounts for the inverse-parent/inverse-interval list copy per accepted
orientation. It remains non-coordinate-changing audit evidence rather than global event-catalogue
reconstruction.
Session 21 adds the supplied fixed-region SISCAN route. It chooses the nearest eligible fourth
sequence from the round-wide source WPGMA context, scores the already reconciled region with the
final permutation count and shared MakeVRand prefix, and retains normal-tail, region-, window-, and
project-adjusted probabilities. It cannot move event coordinates. See
native-siscan-discovery-trace.md.
Session 22 adds PHYLPRO only as the supplied graphical event-review diagnostic. It is not inserted
into this post-group dispatcher or the global event catalogue because the supplied RDP5 route has
no implemented PHYLPRO significance test. Its three current-role curves and explicit source repairs
are traced separately in
native-phylpro-review-trace.md.
| Native matrix | Browser value |
|---|---|
FMat / SMat |
direct whole-outside / whole-inside Jukes–Cantor distance |
FAMat / SAMat |
raw whole-outside / whole-inside midpoint-ultrametric topology rank |
FCMat / SCMat |
weak-node parent-promoted whole-outside / whole-inside topology rank |
*MatSmall(role, y) |
distance from the working representative for role to ordinary candidate y |
Candidate-to-candidate checks in the equivalence and straggler passes can revisit the same pair many
times. Session 8 memoized symmetric (sequence A, sequence B) values separately for the two regions
and three distance families. Saved tree-panel values remain constant-time; a candidate outside the
bounded panel falls back to a cached direct distance. This changes repeated-work cost, not branch
ordering or thresholds.
- MaxChi strongest-peak, CHIMAERA strongest-target, ordinary-kernel GENECONV, and 3SEQ Findall corroboration are active. Their full native method dispatch and late event-catalogue reconstruction are not.
- Post-group rechecks for other method families not represented by this checkpoint also remain inactive.
- The 100-sequence event-tree panel and labelled direct-distance fallback are explicit browser bounds rather than native full-matrix storage.
- Compilation, runtime execution, and native golden comparison were intentionally not performed.