The Auto Exports pathway runs the full processing pipeline — parse → QAQC → corrections → export — as a single guided run, with file selection, destination folders, and structural-decision defaults configured up front. Use it when survey files arrive in a known location, when destinations are stable across days, or when you want a one-button flow that lands corrected exports in pre-set folders without manual review at each stage.

Triggering

Click the 📋 Run Auto Exports button at the top-right of the upload row. The Auto Exports popup opens.

Source file

Three modes, selected via the segmented control at the top of the popup:

• 📁 Search folder — point at a folder; the app picks the most recent dated file matching the configured extension. Useful for daily survey workflows where files land in a known directory.
• 📄 Select file — pick a specific file directly. Useful for re-running auto exports on a known input.
• ✓ Use Working Dataset — available whenever a dataset is loaded in the app. Skips the file reload and runs the pipeline — QAQC included — against the dataset already in memory, whether that's a raw upload or the engine's latest output after corrections.

The default folder-search extension is .xml. To change it, open More Options and switch the Folder search file type segmented control to .gpx, .jxl, or .csv.

Destinations

Three independent export streams, each with its own segmented control to pick a destination mode:

• GPX Export — folder, Downloads, or none. Folder mode writes the GPX file directly to the chosen folder.
• SHP zip Export — folder, Downloads, or none. Folder mode writes the shapefile ZIP. An optional secondary folder destination is available — the same ZIP is written to both.
• Over-write GIS Source File — folder or none (no Downloads option). Overwrites the five shapefile components (.shp, .shx, .dbf, .prj, .cpg) in the chosen folder, matching the existing [prefix]-RTK-Completed_Stations.* filename so a live GIS layer pointed at this folder always reflects the latest data. Secondary destination available.

Any section can be set to ✕ No export to skip that stream entirely.

More Options — auto-confirm toggles

Six toggles control how QAQC handles structural decisions and detected anomalies during the run. All default off.

• Auto-confirm origin offsets — every detected S-origin offset group is confirmed without review. Off → offset lines appear in the OFST subpanel for inspection.
• Auto-merge L-offset grids to parent — grids detected with a separate L-numbering origin are merged into the parent grid. Off → grids appear as candidates in the NSD merge UI.
• Auto-fix grouped format names — grouped non-standard names (e.g. L7400E3750, L7400 S3750) are renamed to canonical L[n]S[n] form. Only fires on grouped deviations (>10 points). Off → format suggestions surface as FMT flags.
• Auto-confirm station infill — infill stations (sub-station-spacing points within a line) are confirmed during the run and suppressed from the flag table. Off → they appear in the Infill Stations panel for review. Mutually exclusive with Assume all stations are primary.
• Auto-confirm infill lines — infill lines (fractional-L lines between standard lines, e.g. L5300 on a 200 m line-spacing grid) are confirmed during the run; LABEL/DUPE are suppressed on their points. Off → they appear in the Infill Lines panel for review. Independent of the station-infill toggle and of Assume all stations are primary.
• Assume all stations are primary — any grid that first-pass QAQC flags as having infill is re-resolved at the densified spacing. Every point becomes a standard anchor of the finer lattice; no infill classification is applied. Use for fully-densified surveys where the base + infill split is an artefact of detection, not a real survey-phase distinction. Applies on every Auto Exports run, whatever the source. Per-grid override via the Survey Spacing popup; mutually exclusive with Auto-confirm station infill.

Generate

Click ▶ Generate. The app:

1. Loads the source — the selected file, or the dataset already in memory for the Use Working Dataset source.
2. Runs QAQC — grid detection, affine fit, flag computation.
3. Applies any enabled auto-confirm toggles to the structural-decision subpanels and flag table.
4. Checks whether the result is fully resolved.

Flag-review interlude

If unresolved flags remain after the auto-confirm pass, the popup minimises and the QAQC flag table appears for review. Accept suggestions, type corrections, or reject — the same review surface as the manual pathway. Your configured destinations are remembered while you work.

When you're done reviewing, click Apply + Auto Export in the QAQC apply row to continue.

Apply + Auto Export

The app applies your actioned corrections, re-runs QAQC against the corrected dataset, then writes the configured export files in sequence — GPX, then SHP zip, then GIS source overwrite. A summary appears in the popup listing each destination that was written.

When QAQC passes cleanly

If no flags remain after the auto-confirm pass, the run completes without surfacing the flag table. Exports write directly; the summary appears at the bottom of the popup.

End-to-end data flow from file input to export outputs. Each stage maps to a discrete processing phase in the app's JavaScript. waypointData[] is the single mutable array that carries points through every stage; pointsCache[] holds the immutable originals and is never modified after load.

Leica XML (LandXML)

Uses DOMParser to read LandXML documents, targeting CgPoint elements under the http://www.landxml.org/schema/LandXML-1.2 namespace with a fallback to namespace-less lookup. Common Leica export variants where the name appears as ="" or n="" are sanitised to name="" before parsing. Concatenated multi-file exports are recovered by wrapping in a <LandXMLCollection> envelope.

Extracted per point: name (name, n, or id attribute), lat/lon (WGS 84 decimal degrees), elevation (elevation, elev, or body text), timestamp (normalised to ISO 8601), role/description. The CQ3D quality metric and all non-core attributes are captured in _extra for optional export. Points with invalid coordinates trigger a warning modal.

Trimble JXL (JobXML)

Parses the <Reductions> section of Trimble JobXML files, extracting <Point> elements. Position is read from the <WGS84> child (Latitude/Longitude/Height in decimal degrees); if absent, falls back to <Local>. Points with only <Grid> or <ECEF> coordinates are skipped (no projection transform available in-browser). Unnamed points (no Name attribute) are assigned UNKNOWN.

All Trimble metadata is captured in _extra: SurveyMethod (Fix, Float, Autonomous, etc.), Classification (Normal, Control, Check, etc.), Code, Grid projected coordinates (North/East/Elevation), QualityControl1 (HorizontalPrecision, VerticalPrecision), and QualityControl2 (variance-covariance matrix: CovarianceXX through CovarianceZZ). These flow through to CSV and SHP exports when Include all source fields is enabled.

GPX

Imports <wpt> elements only — track points (<trkpt>) and route points (<rtept>) are silently ignored, as they lack station names. Namespace-aware lookup is used when the GPX declares an xmlns. Extracted per waypoint: lat/lon attributes, child elements <name>, <ele>, <time>, <desc>/<cmt>. Unnamed waypoints are assigned UNKNOWN.

CSV

Requires Name, Latitude, and Longitude columns (header matching is case-insensitive, trims whitespace, and recognises common aliases like Station, Easting, Northing). Elevation, Timestamp, and Description are optional. Any columns beyond the recognised set are captured in _extra. A CSV exported by this app can be re-imported — column structure is preserved for iterative field-office QA workflows.

Working state

pointsCache holds the original parsed records (including _extra) and is never modified after load. waypointData is rebuilt from pointsCache on every regeneration, with date filters, deduplication, and QAQC corrections reapplied in sequence. This ensures toggling any option always produces a deterministic result from the original source data.

Unnamed grids

When a dataset has fewer than 6 points with L[n]S[n] names but contains at least a dozen unnamed coordinate-bearing points, the engine attempts to infer the grid lattice from geometry alone and synthesise names before QAQC begins. This handles surveys collected without grid names in the field — typical of free-style RTK collection on a planned grid where the crew assigned generic point IDs. The bootstrap runs as the first action inside QAQC, before the parseable-points filter: it estimates the two grid axes (their bearings and spacings), assigns each point a line/station index, detects 1/2-fraction infill if present, resolves per-line station origin offsets, and writes a synthesised L#S# name onto each point that indexes cleanly. The result is a fully named dataset that flows into the normal pipeline — affine fit, flag generation, exports — with no special branching downstream.

If one or more grid-named anchors are already present, the cleanly-indexed anchors pin the labelspace by majority agreement so synthesised names match the existing convention — a single poor-geometry anchor is outvoted rather than trusted; otherwise a default origin of L5000S5000 is used. Points the engine cannot index cleanly remain nameless and surface in the standard non-parsed table. Non-grid datasets — track logs, scattered control points, anything where two coherent axes cannot be established — abort the bootstrap and fall through to the standard message that too few grid-named points were found for grid analysis to run. The current implementation handles a single grid per dataset and 1/2-fraction infill only; the original name is preserved on each point so the synthesis is reversible.

In order, on every regeneration: (1) date range filter removes points outside the selected window; (2) coordinate deduplication (pre-QAQC pass) — when the Remove duplicates (identical coordinates) toggle is on, removes points that share an exact latitude/longitude pair, retaining the earliest-timestamped record per coordinate; (3) QAQC corrections from the last Apply are re-applied as name substitutions; (4) preview renames from the current session are re-applied (keyed by stable _originalName|time so they survive the rebuild); (5) soft-deleted points are removed; (6) if QAQC has been run, it reruns on the cleaned dataset to keep grid models and flag lists in sync; (7) a second pass applies the Post-QC proximity deduplication (radius-based, configurable in the Post-QC Processing panel) and exact-name deduplication if those options are enabled.

Only points whose names match /^L(\d+)S(\d+)/i are analysed. The L and S numeric values are parsed directly from the name as floating-point numbers; zero-padding width is recorded separately and used when formatting suggested names. All other points pass through to exports unmodified.

Phase 1 — Spatial pre-clustering

Parseable points are projected to a local metre-scale coordinate system using a centroid-anchored cosine correction for longitude. Two independent 1D gap scans are then run — one along the northing axis, one along the easting axis — each looking for consecutive inter-point gaps exceeding 600 m. A new geographic cluster is declared only when both axes independently agree there is a gap at approximately the same location. This dual-axis requirement prevents false splits on highly rotated grids that span a wide latitudinal range but remain geographically connected.

Phase 2 — Namespace gap split

Within each spatial cluster, the unique L-values (and separately S-values) are inspected for a bimodal gap. The mode of small consecutive gaps is computed as the typical label spacing; a split is accepted only if the largest gap exceeds 3× that typical spacing and the left subgroup contains at least 6 points and the right subgroup at least 2. This catches grids that share geographic space but use non-overlapping label ranges (e.g. Grid A uses L8000–L12000 and Grid B uses L18000–L20000). Namespace splitting is applied exactly once at the top level and deliberately not called recursively — recursive calls produce false S-axis splits when grids have different line lengths.

Phase 3 — Structural clustering (per sub-cluster)

After spatial and namespace splitting, each remaining sub-cluster is examined for multiple grid orientations within the same geographic and label space. The approach exploits the fact that consecutive stations on the same survey line share a consistent bearing and distance that uniquely fingerprints their grid.

1. Quick single-grid check: fit an affine to all points; if the full-dataset RMSE is less than 2× the tolerance the cluster is treated as one grid and the pipeline stops here.
2. Intra-line vector extraction: points are grouped by L-value. Within each L-group, consecutive S-sorted pairs are evaluated as inter-station vectors. Pairs are accepted only if the physical distance is between 15 m and 200 m (the 15 m minimum filters GPS-noise micro-vectors). For each L-group with ≥ 2 raw vectors, the median bearing is computed; vectors deviating more than 20° from the median are rejected. This achieves very high same-grid purity even when two grids share the same label namespace.
3. Mean-shift clustering in orientation space: each surviving vector is encoded as (cos θ, sin θ, log(dist) × 0.4). Mean-shift is run on this 3D feature space with bandwidth 0.10 (tuned to resolve grids separated by as little as 25° in azimuth). Converged seeds within 60% of the bandwidth are merged into modes. Modes with fewer than 15 contributing vectors are discarded as spurious.
4. Seed affine fits: for each surviving mode, the point-pairs contributing to it are collected as seed points. An initial affine is fit to these seeds using robust 3σ rejection.
5. Iterative reassignment: starting from the seed transforms, every parseable point is assigned to its minimum-residual grid, each grid's transform is refit on its newly assigned points, and the process repeats until no point changes assignment or 20 iterations elapse.

Phase 4 — Grid merge (auto-detect mode only)

After structural clustering, candidate grid blocks whose bounding boxes overlap or are adjacent (within ~500 m) are checked for similarity. Two blocks are merged if their directed station-axis azimuth difference is less than 15° (mod-360, not mod-180 — the mod-180 form aliases opposite-facing grids) and their line-spacing ratio is less than 1.25. Merging uses iterative refinement until no further merges are possible.

Phase 5 — User-count forcing (when grid count is specified)

If the user has entered a specific grid count, the pipeline attempts to reach that target by: splitting the highest-RMSE block (namespace gap split first, then k-means++ on affine residual error vectors as fallback) if too few grids exist; or merging the most similar adjacent pair (weighted by azimuth difference and spacing ratio) if too many exist. A warning is shown when the forced count differs substantially from the auto-detected count.

Phase 6 — Final global reassignment

Once the grid set is stable, a final iterative reassignment pass runs across all grids simultaneously. Every parseable point — including those that were in small sub-clusters — is assigned to its lowest-residual grid, with transforms refit after each round.

Timestamp coherence pass

After geometric assignment converges, points in an "uncertain bucket" (residual between 0.5× and 1.5× their current grid's tolerance) are tested against alternative grids using two signals. First, a timestamp score counts how many same-day neighbours the point shares on each candidate grid (points are grouped by predicted L-line bucket, rounded to the nearest 50 label units). Second, the point's current name is tested against both the current and the alternative grid's predicted node. Three outcomes are possible:

• Alt grid scores ≥ 1 same-day linemate AND the name fails the current grid AND the name fits the alt grid node → reassign to the alt grid.
• Alt grid scores ≥ 1 same-day linemate AND the name fails the current grid AND the name does NOT fit the alt grid node either → keep current assignment, mark with TSCON flag (GPS position, timestamp, and name are mutually inconsistent).
• Alt grid does not score higher on timestamps, or the name already fits the current grid → no action.

The affine transformation maps grid coordinates (L, S) to geographic coordinates (lat, lon):

The 6 parameters are solved by Gaussian elimination with partial pivoting on the normal equations. Grid azimuth is derived from the S-direction coefficients (b, e) — the direction perpendicular to survey lines — using atan2(e × mLonScale, b × 111320). Line spacing in metres is derived from the magnitude of the L-direction vector (a, d). Inlier RMSE is the root-mean-square of residuals after robust rejection. These values are reported in the per-grid section of the QAQC panel.

The robust fitting procedure runs up to 5 iterations: fit all points → compute residuals → exclude points with residual > mean + 3σ → refit → repeat. Iteration stops early if no points are excluded or if fewer than 6 inliers would remain. The inliers from the final pass drive the adaptive flagging threshold.

Line and station spacings used for node snapping are detected from the modal gap in the observed L- and S-value distributions, not from the affine parameters. This makes snapping robust to irregular label numbering (e.g. L values that increment by 50 rather than 1).

Single-line affine (V3.05193)

The standard 6-parameter affine requires at least two distinct L-values. When a grid presents only a single line of stations (degenerate L-axis), the fit is opt-in via opts.allowSingleLine: the engine collapses to a 4-parameter L-on-S affine, stamps the grid with _singleLine = true, and propagates that flag through the residual pipeline so DIST/EXTR thresholds and label suggestions remain geometrically meaningful. When the opt-in is off, single-line grids are detected, reported in the affine summary, but excluded from flagging.

The L-axis infill subsystem teaches the QAQC engine to recognise fractional-L lines — survey lines at a sub-multiple of the grid's base line spacing (e.g. L5300 on a 200 m grid) — as legitimate densification rather than mislabel candidates. It is the structural mirror of the S-axis infill-station subsystem.

Detection (Phase 12.1). _detectInfillLines(grid, gi) runs once per grid immediately after the S-axis _detectInfill, temporally after Phase 12.5 shift-table population (the phase number 12.1 is a name, not a sequence — it runs after 12.5 so the shift-aware residual check is available). The pass short-circuits when infillLineNoInfill[gi] is set. Otherwise it selects a base line spacing from observed-gap candidate modes — or from userSpacings[gi].line when the user has forced one — commits an L-origin into g._lineOrigin, and classifies each observed L-value as standard or candidate-infill. Candidates are gated through three checks: bracketing (which immediate-slot std-lattice lines are observed in the grid above and below the candidate), local coverage of those bracketers (T_local = 0.50 for two-bracketer cases; T_local_edge = 0.67 for single-bracketer cases), and a shift-aware per-point residual check via the node(gi, l, s) wrapper. The no-bracketer relaxation (V4.0005, Q-1): a candidate with neither immediate-slot std-lattice bracketer observed in the grid still passes when (a) ≥ 5 observed stations, (b) the candidate's offset matches a canonical fraction of the base spacing (1/2, 1/3, 2/3, 1/4, 3/4) within tolerance, and (c) every station's S-value lies on the grid's std S-lattice — boundary infill lines at the top or bottom of a grid (where the upper or lower bracketer wasn't surveyed) are now detectable; genuine off-lattice rows still fail. The residual check is OFST-interpolation-aware (V4.0005, Q-2): when bracketers carry non-equal OFST shifts, the candidate's expected position is computed via linear interpolation of those shifts across the candidate's mid-step L — so an L-axis infill line between a clean bracketer and a fractionally-OFST bracketer still stamps cleanly. Points on a PASS-classified line that clear the per-point residual gate are stamped _isInfillLine on the persistent waypointData object; per-grid results are written to qaqcInfillLineGroups[gi] and infillLineDetectedSpacing[gi]. PASS-line points that fail the residual gate are typos — they are not stamped, and fall through to LABEL.

Base line spacing (Phase 9). detectBaseLineSpacing(pts, options) returns { lineSpacing, lineOrigin } — the helper's earlier ambiguous return was removed at V&V C4 (V3.05206) because the authoritative ambiguity write happens at Phase 12.1's commit step, not here. Precedence at the Phase 9 call site: a user-forced Force Line Origin / Line spacing (from the Survey Spacing modal) wins; otherwise a project-wide cached default (qrDefaultLineSpacing) is written into userSpacings[gi] once per grid per file load via _qrCacheAppliedThisLoad before the helper is invoked; otherwise the helper derives lineSpacing from observed-gap candidate modes. Only lineSpacing is destructured at the Phase 9 site, but the same call retains _blsResult.lineOrigin for g._lSnapOffset construction (DR-L6 / V3.05214) — the std-lattice origin consumed by _snapL's fallback path.

Slot-engine integration (Phase 13). _snapL(val, g, labelL, anchorS) consults the runtime _confirmedInfillLineKeys Set: when the key ${gi}:${Math.round(labelL)}:${anchorS} is present, the function returns Math.round(labelL) directly — the point's own fractional-L position — bypassing the std-lattice snap. Otherwise it returns the nearest std-lattice node via g._lSnapOffset + g.lineSpacing. Phase 13's _nodeOccupantsMap consumes the result for bucket keying, so confirmed-infill-line points get their own buckets and stop firing LABEL/DUPE; rejected or unconfirmed candidates remain on the standard lattice and continue to flag.

State & persistence. The user's decisions persist across re-runs — infillLineConfirmedState (per-line gi:line:lineL and per-point gi:pt:idx shapes), userLineOrigins (per-grid Force Line Origin), and infillLineNoInfill (per-grid No L-axis infill opt-out). These reset only at file-load reset sites (currently 10 per V&V C2 / V3.05206). The engine's detections are recomputed on each runQAQC — qaqcInfillLineGroups, infillLineDetectedSpacing, qaqcInfillLineAmbiguous, and the per-grid derived g._lineOrigin. The runtime _confirmedInfillLineKeys Set is cleared once at the top of each runQAQC then rebuilt per-grid during Phase 12.1's pass from infillLineConfirmedState ∩ per-point gate survivors. applyQAQCCorrections clears _isInfillLine stamps on points belonging to ERRONEOUS decisions before exports rebuild (V3.05222 parity fix) — both the per-point gi:pt:idx and per-line gi:line:lineL ERRONEOUS shapes — so the brief Apply→re-run window does not surface stale state in summary tiles or the chart infill series.

L-origin ambiguity. When the observed line distribution is symmetric, two L-origin interpretations can be equally consistent. The engine commits one via a deterministic ordering — snap-consistency against g._lSnapOffset is the primary key (DR-L5 / V3.05213, introduced to keep detection-side origin consistent with the snap-path lattice), with descending std-line count and lex-min std-set comparison as secondary tie-breaks. When the top two eligible origins yield equal std-line counts, qaqcInfillLineAmbiguous[gi] is set, surfacing the ⚠ AMBIGUOUS badge in the Infill Lines subpanel header. Forcing a Line Origin in the Survey Spacing modal short-circuits the eligible-origins walk entirely (gated by the UI guard that disables the origin input until a Line spacing is also set) and removes the ambiguity.

Auto-exports. The Auto Exports pipeline confirms detected infill lines automatically when the Auto-confirm infill lines toggle (qrInfillLineAutoConfirm) is on — independent of the station-infill toggle (qrInfillAutoConfirm) per V3.05211, after the L-3 sub-batch's earlier unified gating was split. The post-run summary log line groups S-axis and L-axis confirmations into a single ${N} infill confirmed count via the shared _infillConfirmedCount counter.

A power-user toggle in the Auto Exports popup (More Options panel) that pre-densifies any grid with detected infill before the QAQC re-run, so every collected point — anchors and what would have been infill candidates — is resolved as a standard anchor of the densified lattice. Default OFF; setting is sticky (localStorage).

What it does

When enabled, the Auto Exports pipeline inserts a Pass 1e step between Run 1 and Run 2 of _qrRunPipeline. For each grid the pipeline runs both axes in parallel:

• S-axis (station spacing). If Run 1 produced a detected S-axis infill spacing (infillDetectedSpacing[gi]) and the user hasn't already forced a station spacing, the pipeline writes the detected sub-spacing into userSpacings[gi].station and sets infillNoInfill[gi] = true. Run 2's Phase 9 picks up the new station spacing; Phase 12's _detectInfill early-exits on the no-infill flag; every downstream station — including what would have been infill — is treated as a standard anchor of the densified station lattice.
• L-axis (line spacing) — V4.0005 (Q-11). If Run 1 produced a detected L-axis infill spacing (infillLineDetectedSpacing[gi]) and the user hasn't already forced a line spacing, the pipeline writes the detected L-sub-spacing into userSpacings[gi].line and sets infillLineNoInfill[gi] = true. Run 2's Phase 9 picks up the new line spacing; Phase 12.1's _detectInfillLines early-exits on the no-infill flag; every formerly-fractional-L line becomes std on the densified L-lattice and no _isInfillLine stamps survive. Closes the prior inconsistency where GPX/SHP [INFILL] tagging (which reads _isInfill only) treated the toggle as effective for S-axis only while chart/summary tile counts (which also read _isInfillLine) still surfaced L-axis-infill points in the infill series.

Either axis can fire independently. The notice surfaces both as separate entries (e.g. L7050 (75m → 50m) for an S-axis densification, L7050 (150m → 75m, line) for an L-axis densification on the same grid). Skip rules apply per axis — see below.

Skip rules

The pipeline respects explicit user instructions and skips each axis's rewrite independently. The S-axis rewrite is skipped when (a) userSpacings[gi].station is already set (the user picked a station spacing in the Survey Spacing modal) or (b) infillUserSpacing[gi] is set (the user explicitly declared an S-axis sub-spacing — suppressing infill would silently override them). The L-axis rewrite is skipped when userSpacings[gi].line is already set (the user forced a line spacing). Both axes are skipped when (c) the grid is about to be NSD-merged-away (its stateKey is in _nsdMergedKeys) — writing to a soon-to-be-deleted gi is moot. User station origins (userStationOrigins[gi]) and L-origin (userLineOrigins[gi]) are intentionally not in the skip predicate — origins are base-invariant in physical metres (Inv-5), so densifying the spacing doesn't conflict with a user-set origin.

Persistence and revert

Side effects persist in-app after the run. The rewrites to userSpacings[gi].station and userSpacings[gi].line are normal Survey Spacing settings — they don't auto-clear when the toggle is later turned off. Revert by clearing the spacing in the Survey Spacing modal for the affected grid(s), or by reloading the file (which resets all user spacings). Each Pass-1e rewrite is recorded in _qrLastRunTreatedAsStandard[] (with an axis: 'S' | 'L' tag per entry) and surfaced as a notice line in the auto-exports completion summary.

Mutex with auto-confirm station infill (V3.05196)

This toggle is mutually exclusive with the Auto-confirm station infill toggle in the same More Options panel — enabling one disables the other. The two are complementary strategies for handling infill: auto-confirm accepts the infill detection and treats those points as confirmed infill; "Assume all primary" rewrites the spacing so infill detection doesn't trigger in the first place.

When to use

Use this mode when the operator knows ahead of time that all collected stations are intended as primary anchors on a densified grid — for example, a survey collected at 25 m station spacing reusing planned data that was generated at 50 m spacing. In this case the "infill" pattern is the intended grid, not an exception, and reviewing each infill grid through the standard subpanel adds no value.

Lattice-geometry DIST / EXTR thresholds

Both gates are derived from the line's operative lattice spacing (g.stationSpacing for standard lines, infillDetectedSpacing[gi] for confirmed-infill lines). A residual at or below distThresh is silent. A residual above distThresh but at or below extrThresh fires DIST (label is plausibly correct, position is moderately off). A residual above extrThresh fires EXTR — the label can no longer plausibly point to the closest lattice node. α = 0.20, β = 0.50 on standard lines and infill at subSp ≥ 50 m; α = 0.40, β = 0.65 on tight infill (subSp < 50 m). The user-minimum is the DIST floor on standard lines (default 15 m); a per-grid override in the Survey Spacing modal replaces the DIST threshold for that grid only. The extrThresh threshold is geometric — β × spacing, a property of the lattice, not GPS noise, and not directly user-set. Note, however, that the Pass 1 gate silences any residual at or below distThresh before DIST/EXTR classification — so raising the Min DIST tolerance (or a per-grid override) above a point's residual suppresses its flag, EXTR included. Inlier-set membership remains the affine-fit gate (Phase 9 / Phase 12.6 RMSE) but is no longer an EXTR signal.

Suggested name generation

For each flagged point: (1) the affine inverse (Cramer's rule on the 2×2 system formed by the L and S columns) gives a continuous predicted L and S; (2) these are snapped to the nearest observed L- and S-values using the detected spacing, producing a canonical grid node; (3) the node's geographic position is back-projected through the forward affine; (4) if multiple points occupy the same node, they are sorted by timestamp and assigned the anchor name (S ending in 0 or 5) for the earliest, then anchor+1, anchor+2, etc. for each subsequent arrival. Zero-padding width is preserved from the original dataset.

Collision detection

A collision is detected when another parseable point either shares the suggested name or lies within 0.5× the minimum grid spacing of the predicted node. All parseable points are considered as potential collision partners — there is no residual restriction. Genuine repeat-station pairs — same L, same anchor-S (where anchor = ⌊S / 5⌋ × 5) — are excluded from collision detection regardless of proximity. If any collision partner is on the same grid, DUPE fires. If any partner is on a different grid, XCOL fires independently — both tags can appear together when a point collides with partners on both grids. XCOL is a structural observation (disabled by default) and does not imply an error on its own; TSCON separately handles cases where cross-grid overlap is accompanied by timestamp or label inconsistency.

Flag tag assignment

Multiple tags can apply simultaneously to a single point:

DIST is the fallback tag — assigned only when no other flag applies. DIST with a green name means the label is correct, only the GPS fix is suspect. XCOL alone (also green) is a structural note that two grids share a node — no rename needed and no action required unless TSCON or LABEL also appear. The current name text colour reflects the worst tag on that row: red for EXTR/TSCON/LABEL, orange for UNNAMED/ORPH/DUPE, yellow for FMT, green for DIST or XCOL only.

Clicking Apply Corrections writes all accepted renames into a qaqcCorrections map keyed by original name. On every subsequent regeneration this map is re-applied to waypointData after parsing, ensuring corrections persist through option changes. Rejected rows keep their original names and are not entered into the corrections map. Soft-deleted points (tracked by a set of unique GPS-coordinate + name keys) are filtered out at the same stage. The four export formats (GPX, Shapefile, CSV, KML) are rebuilt from the corrected, filtered dataset.

The Data Preview table supports two types of in-session edits that operate independently of the QAQC correction system:

Soft-deletes

Tracked in a deletedPtKeys Set keyed by ptDeleteKey(pt) — a stable key using the raw upload name (_originalName||name) and timestamp, so the entry survives QAQC renames. Deleted rows remain visible in the preview table (strikethrough, greyed) — they are not removed from the DOM. On every regenerateExports() call, deletedPtKeys is re-applied as a filter step after QAQC corrections, so the deletion persists through all option toggles. The GPX, Shapefile, CSV, and KML builders each independently filter with !isPtDeleted(pt) (buildGpxFromData, buildShapefile, buildCSV, buildKml), preserving the deletion across format-specific code paths.

Preview renames

Tracked in a previewRenames Map keyed by ptRenameKey(pt) — a stable key using _originalName|time rather than the current mutable pt.name. Each entry stores { original, newName }. On every regenerateExports() call, previewRenames is re-applied after qaqcCorrections, using the stable key to locate points even after waypointData is rebuilt from pointsCache. The revert target is always the post-QAQC name at the moment the first edit was committed — never the raw upload name. When the Include previous name toggle is on, the raw upload name is preserved as the PREV_NAME column in Shapefile exports and the Prev_Name column in CSV exports.

Reset behaviour

Both deletedPtKeys and previewRenames are cleared on every file load (all four load paths: single XML, folder XML, CSV, and QAQC-corrected re-export). They are also cleared in qrClose() so Auto Exports session state does not leak between pipeline runs.

The filter operates on the final waypointData array after all other processing. It is evaluated by the passesFilter(name) function, which applies two prefix lists:

• Include list — a comma-separated list of name prefixes. If the list is non-empty, a point must start with at least one listed prefix to pass. Comparison is case-insensitive and is made against the name with any RTK display suffix already stripped.
• Exclude list — a comma-separated list of name prefixes. A point starting with any listed prefix is always excluded, regardless of the include list.

The filter always controls what appears in the dashboard summary counts, the daily chart, the satellite map, and the data preview table. It does not affect export files unless the Remove filtered stations from export toggle (in the Post-QC Processing panel) is ON. When that toggle is enabled, passesFilter() is additionally called when building each export blob — points that fail the filter are omitted from GPX, Shapefile, CSV, and KML output.

In addition to the prefix filter, the Station Summary counts and the Cumulative Stations chart apply a parseLS() gate: only waypoints whose names resolve to a valid L[n]S[n] grid format are counted. Non-parseable waypoints pass through to the map, preview table, and exports unaffected, but are excluded from all summary statistics and the chart's new production / repeat / infill / cumulative series.

The isGridEnabled(pt) function provides a parallel gate for the per-grid export toggles. Non-grid points (those without a _gridIdx stamp, i.e. non-L[n]S[n] names) always pass this gate regardless of the toggle state.