Open Tools · Closurework

Closurework user guide

Use Closurework for traverse misclosure and Bowditch adjustment, levelling reduction, coordinate geometry, staking tables, parcel areas and network adjustment. To use the tool itself, open Closurework.

Professional use. Results are for education and preliminary design. They must be checked by a qualified professional, and they are not a substitute for code-compliant design or licensed engineering judgement.

Closurework reduces survey observations in a local planar coordinate system. It supports traverse closure, differential levelling, coordinate geometry, construction staking, straight-boundary parcels and weighted survey-network adjustment. It runs in your browser. Pick a workspace along the top, enter lengths in metres and angles in degrees, and the results, chart and tables update as you type.

Getting started

Choose a workspace: Coordinates, Traverse, Levelling, Staking, Parcels or Network. The quickest start is Examples, which opens a complete worked calculation with invented numbers. For a first traverse, enter the known start and end points, then the ordered horizontal distances and azimuths of the legs. Choose "None" to inspect the raw misclosure, or Bowditch to distribute it in proportion to leg length. Check both the raw misclosure and the gap after adjustment before exporting. Bowditch does not identify a bad observation or adjust observed angles.

The Verify page runs reference checks that are mathematical verification, not published field-validation cases. A signed run token is required for worker calculations and Verify. Editing, JSON Save/Open and reading this guide do not require a token. Computation runs in the browser; the only service exchange is the domain-gate token request.

Units and coordinate conventions

All lengths, coordinates, heights, rod readings and distance standard errors use metres. Angles are typed in degrees: decimal (123.4567), degrees minutes seconds (123 45 30.5 or 123-45-30.5) or, wherever a bearing is asked for, a quadrant bearing such as N 30 15 E. Angular standard errors are typed in arc-seconds. Saved files and CSV store radians, the engine unit. Angles shown as text are the nearest decimal to the stored radians, so opening a file and saving it again can change an angle in its 16th digit, and a quadrant bearing you typed is saved as the equivalent azimuth. Easting is e, northing is n, height is positive upwards. Azimuth starts at north and increases clockwise, in [0, 2π). A right offset is positive when looking from the baseline start toward its end. Grade and cross-slope are dimensionless signed rise per horizontal run. Chainage is measured from the baseline start; negative chainage and extension beyond the endpoint are permitted and explicitly retained.

Convert other units before entry; this release has no foot or survey-foot selector. Heights and bearings must be on one consistent user-selected datum. No projection, scale-factor, geoid, curvature, refraction or slope-distance correction is applied. Inactive height values in a horizontal network remain in the saved model and are not adjusted.

Coordinate geometry

Forward uses a start, nonnegative horizontal distance and azimuth. Zero distance returns the start. Bearing conversion turns a quadrant bearing into an azimuth. Inverse returns distance and azimuth; coincident points have an unavailable azimuth (null), rather than an invented direction. Quadrant bearings accept N/S, E/W and a nonnegative angle up to π/2. Intersections are between infinite lines: signed along-line parameters can be negative. Parallel or poorly separated directions produce a diagnostic.

Traverse and metes-and-bounds

Traverse input contains start, desired end, ordered calls and method none or bowditch. Outputs include raw and corrected coordinates per leg, correction vectors, total length, vector and scalar misclosure, length/misclosure precision and endpoint residual. Precision is unavailable at exact closure; no accuracy classification is assigned. Bowditch is an approximate coordinate-balancing rule, separate from weighted network adjustment.

Metes-and-bounds input is an ordered list of straight bearing-distance calls. Closure must be within 1e-10 of total perimeter; the original closing gap is reported, and area uses the polygon closed from the final retained vertex to the start. Open calls are refused; explicitly adjust a traverse first if that is appropriate. No natural-language deed parsing or curved boundaries is provided. A geometric description cannot determine ownership, monuments, title or a legal boundary.

Differential levelling

Each row is one backsight minus foresight height difference, with a positive segment distance. Start and known end heights define the misclosure. adjust distributes the closing error by segment distance. The table retains raw difference, raw height, correction and adjusted height. Enter turning points as successive rows; this release does not model intermediate sights, instrument calibration, collimation or sight-distance imbalance corrections. Empty books with equal start/end are valid.

Construction staking

A baseline start/end, start elevation and signed grade define a straight sloping design line. Enter chainage/right-offset rows and an instrument line-of-sight elevation. The table gives easting/northing, design height and grade rod (instrument elevation minus design height). Offset does not change design height: crossfall requires a separate model. A negative rod is retained as a signed mathematical difference; it does not claim a physical reading is possible.

The daylight operation intersects a design line through an edge offset/elevation with straight terrain defined by elevation at zero and signed slope. Choose side +1 or -1 relative to increasing cross-section offset. A parallel pair or an intersection on the opposite side is refused. This is a local straight cross-section calculation, without terrain surfaces or roadway templates.

Parcels, subdivision and strips

Enter 3–500 unique boundary vertices in order, without repeating the first at the end. The boundary must be simple and its area resolved. The engine reports signed area, absolute area, orientation, centroid and perimeter. Coordinates are translated and scaled for area integration; intersection classification uses the exact sign of represented coordinates when floating-point orientation is ambiguous.

Subdivision requires a convex polygon, a cut-normal azimuth and a positive target area below the total. It returns the two clipped polygons, their areas, the cut origin and signed normal offset. Bisection stops within 1e-10 relative target-area error, or returns a diagnostic. A straight strip also requires a convex polygon: supply a reference line origin/azimuth and low/high right offsets. The result is the parcel portion inside that strip; a disjoint strip returns an empty polygon and zero area. This is geometric clipping, without easement rights, cadastral topology or legal approval.

Network adjustment

Choose horizontal or level. Point rows contain unique id, e, n, height and fixed. Fixed points are exact control; uncertain control should be represented by observations in a suitably designed network. At least one component must be unknown. Horizontal observations are distance, azimuth, deltaE or deltaN; level observations are heightDifference. Each observation has a unique id, from/to ids, measured value and positive standard deviation. Observations are independent; correlated measurements are unsupported. Initial coordinates must give meaningful noncoincident distance/azimuth endpoints. Adequate control and geometry must constrain the datum.

The solver whitens by the user standard deviations and uses column-scaled pivoted Householder QR (with row equilibration and restored original covariance weights for square systems) with iterative linearization and bounded backtracking. The iteration limit is 1–30; relative coordinate-step tolerance is 1e-12–1e-4. Success requires a small step plus a small normalized weighted gradient or weighted residual. Hitting the limit remains not-converged even if the last applied step appears promising. A not-converged result contains provisional coordinates and local diagnostics only; do not treat them as a completed adjustment.

Inspect iterations, relative step, normalized gradient, sum of squared weighted residuals, degrees of freedom, condition estimate and memory estimate. Rank deficiency and condition estimates above 1e12 are refused. The condition estimate is a conservative QR and equilibration diagnostic, not a rigorous matrix condition bound. A flat objective can stop before the iteration cap without declaring convergence. Changing starting coordinates may locate a different solution; global uniqueness is not established.

Residual sign is computed minus observed; angular residuals use the shortest signed difference across north. The covariance is a priori: it uses the supplied sigmas, without multiplying by the residual variance factor. Standard errors and horizontal ellipses therefore also use a priori sigmas. Ellipse axes are one-sigma semiaxes; the major-axis azimuth is clockwise from north and unavailable for a circular ellipse. These ellipses are not stated-confidence regions. Posterior variance factor is weighted residual sum of squares divided by observations minus unknowns; it is unavailable when degrees of freedom are zero. No p-values, confidence limits, accuracy certification or automatic outlier removal are supplied.

Observation redundancy is 1 minus its weighted leverage, with rounding bounded to [0,1]. Normalized residual divides the weighted residual by the square root of redundancy; it is unavailable when redundancy is <=1e-10. This release does not use normalized residuals as an automatic acceptance test. The sum of redundancies should equal residual degrees of freedom, and the converged residual should be weighted-orthogonal to the design columns.

Controls, editing and shortcuts

Each workspace has a form on the left and live results on the right. Tables (legs, setups, corners, stakes, points, observations) add and remove rows with buttons, page at 25 rows, and accept pasted spreadsheet lines: tab, comma or semicolon separated, with a tick box when the first line is a heading. Results are hidden while the inputs are invalid and marked out of date while a new calculation runs, so numbers from earlier inputs are never shown beside new ones. Renaming a network point renames it in every observation.

Undo and redo cover every edit, including pasted rows and opened files, and take you back to the workspace where the change was made. Edits to one field within 0.8 seconds are one step. Your typing is kept automatically in this browser (local storage) and read back strictly: anything that does not match is discarded rather than guessed. Shortcuts: Ctrl or Cmd + Z, Shift + Z or Y, S (save), O (open), Enter (recalculate), Alt + 1 to 6 (workspaces), ? (list), Esc (close).

Files and exports

Save needs valid inputs but not a calculation. Save/Open uses original UTF-8 JSON with format: "closurework", version 1, copyright, terms and one workflow job. All supported fields are explicit; unknown fields, coercions, missing fields and unsupported versions are refused. Input schema admission is shared with the engine; geometry and rank are checked when running. Opening JSON does not establish that a model has a solvable geometry. Projects are limited to 4 MiB. Files contain user data and should be handled according to the project's privacy requirements.

CSV is a rectangular path/value result table, with the tool attribution and professional-use notice. Numeric cells retain JavaScript round-trip precision; unavailable values become n/a. Text cells normalize fullwidth characters, neutralize formula prefixes and remove embedded control/newline characters. CSV is a report export, not a project-import format. Chart exports are SVG or PNG with the attribution and professional-use notice, drawn from the same code as the screen. Print uses the browser print dialog. No vendor project formats or binary instrument files are supported.

Resource limits, limitations and troubleshooting

Traverse, level and stake tables accept up to 5,000 rows. Networks accept up to 5,000 observations, 100 points and 60 unknown coordinate components. Estimated working memory must stay below 128 MiB. Network backtracking is capped at 16 trials per iteration. These are admission limits, not speed claims. The engine does not silently truncate data. Coordinate magnitudes and supported lengths are bounded by 1e12 m; positive physical inputs and sigmas must be at least 1e-12 in their unit. Area targets use square metres and a separate 1e-24–1e24 m² admission window; area outputs scale with length squared.

A displacement too small to resolve at a large origin receives a RESOLUTION diagnostic. Translate to a nearby local origin and retain the translation separately. A singular network requires better control or independent geometry, not a larger iteration budget. Check observation ids, mode, angular units and sigmas before retrying. Empty traverse/level books are valid only if their endpoints agree.

Calculations run in a module worker. Cancel terminates that worker, because a queued message cannot interrupt synchronous JavaScript; the next run starts a fresh one, and late replies from a stopped run are ignored. The engine also checks a cancellation callback during QR and iteration.

How we verify

One register of cases supplies both the automated tests and the Verify page, where Run all checks shows each reference value, computed value, difference and tolerance, and the headline reads as passing only when the whole run completed and every ready case passed. The page prints as a report. It records input, expected result, units through the workflow, tolerance, provenance class, basis and method family. Ready cases are synthetic mathematical verification. Published survey validation is pending source and is neither run nor counted as a pass. Multiple output checks from one construction are reported as one method family, rather than multiple independent validations.

Separate tests compare pivoted QR against normal-equation elimination on 240 seeded small, well-conditioned networks. Geometry uses independently derived triangles and rectangles. Survey-book tests check endpoint balance, weighted residual orthogonality, redundancy sums, sign reversal, coordinate translation and scale families. Tests probe non-finite inputs, singularity, iteration limits, worker parity, hard cancellation, file round trips and unsafe CSV text. The evaluator rejects non-finite references and tolerances, combined tolerance overflow, and zeroed tiny reference values. Tests do not establish field accuracy or jurisdictional acceptance.

Compared with other tools

Dedicated desktop surveying systems cover larger sparse networks, field controllers, instrument formats, CAD drawing, correlated observations, 3-D geodesy and cadastral management. Closurework currently provides a bounded local-coordinate core with open files and explainable calculations. It does not replace field procedures, site-specific accuracy assessment, licensed boundary work or a professional review. Published field validation against openly available worked survey examples is still outstanding and is stated as pending on the Verify page.

Professional use

For education and preliminary design. Results must be checked by a qualified professional. Not a substitute for code-compliant design or licensed engineering judgement. No legal-survey certification is provided.

Closurework user guide: traverses, levelling and adjustment | Simulations4All