Changelog

Period Compare — Month-End Comparison Fix

If you run month-over-month comparisons, this one matters. On the last day of a month, “one month ago” did not land in the previous month at all.

• ✕Offset periods overflowed at month end. From 31 March, “1 month ago” produced 3 March; from 31 May it produced 1 May — the current month compared against itself. From 29 February, “1 year ago” produced 1 March. Month and year offsets now clamp to the last valid day of the target month. Calendar-mode periods (“last month” as a whole) were never affected.
• ✕Day and week offsets drifted across daylight-saving changes. They subtracted a fixed 86 400 000 ms, which lands an hour off the wall clock and misaligns every sample when two periods are overlaid. They now step the calendar. No effect in regions without DST.
• ~The stddev summary used the population formula (÷n) while Smart SQC and Histogram both use the sample formula (÷n−1) — the same data gave two different numbers across the three symbols. Period Compare now matches them. With a single point it reads “—” instead of 0.
• ~Histogram and Alarm Table were audited alongside and needed no calculation changes. Histogram’s statistics, percentiles, skewness/kurtosis, normal approximation and short-term sigma all check out; Alarm Table’s deadband, chatter counting and Pareto ranking all check out. 48 checks, run against the shipped obfuscated build.

Duval Triangle 1 — Fault Zone Correction

If you use Triangle 1 for DGA screening, update and re-check any recent result that came back DT. Both errors below pointed the same way — towards the milder diagnosis.

• ✕D2 was reported as DT over 9.6% of the diagram — the region %C2H2 29–60 with %C2H4 40–71. D2 is high-energy discharge (arcing); DT is a mixed thermal/electrical fault. The D2 zone extends past %C2H4 = 40 once %C2H2 reaches 29, and the rule stopped at 40 unconditionally.
• ✕T3 was reported as DT over 1.4% of the diagram — the region %C2H2 13–15 with %C2H4 ≥ 50. The discharge test ran before the T3 test and swallowed the 13–15 sliver of T3 (thermal fault above 700 °C).
• ~Triangle 1 is now verified against the published zone polygons across the whole diagram — 125,751 sample points, zero disagreement, on the shipped obfuscated build. The zone shading is drawn from the same function that classifies your sample, so the picture and the verdict cannot disagree.
• !Triangles 2, 4 and 5 are unchanged and not yet verified to the same standard, so the symbol now says so where you will actually see it: a small note under the diagram, and a marker on those three buttons in the Format panel. Triangle 2 currently reuses Triangle 1’s boundaries with different labels and is missing its X1 zone. They stay selectable — existing displays keep working — but confirm their verdicts against your own reference. Triangle 1 is the one to rely on.

Every Remaining Symbol Audited

This completes a functional audit of the whole catalogue. Two findings below are worth acting on even if you have not noticed a problem — Correlation Pro could report an impossible correlation, and Manual Data Entry could accept a reading without writing it.

• ✕Correlation Pro v4.7 — r was wrong on large-magnitude tags. The single-pass correlation formula loses its significant digits when values sit at a large offset with small variation, which is exactly what totalisers, energy counters and epoch-second tags look like. Measured against a known r of 1.000: at an offset of 10⁸ it reported 1.061 — a correlation above 1, which cannot exist — and at 10⁹ it reported 0.000, i.e. “no correlation”. Now computed mean-centred; both cases return 1.000.
• ✕Correlation Pro v4.7 — “0.00” was also used for “cannot be computed”. No overlapping data, a single point, or a flat-lined tag all produced r = 0.00, which on a correlation chart reads as the finding “these two are unrelated”. Those cases now show “—” in a distinct grey, the rolling chart leaves a gap instead of drawing a confident line through zero, and a display with exactly one window of data no longer renders blank.
• ✕Manual Data Entry v1.1 — Save could do nothing at all, silently. If the symbol could not reach PI Web API, Save returned without writing, without an error, and without a banner: the operator typed a reading, pressed Save, and had no way to know it was never recorded. It now says so and keeps the entries on screen. Everything else in the write path was verified against a stubbed PI Web API — partial failures, HTTP 403, and unresolvable tags were all already reported correctly.
• ✕Manual Data Entry v1.1 — a grid interval of 1mo was read as one minute when matching logged values to rows, so a monthly entry grid came back empty.
• ~Alarm Table v1.1 · Histogram v1.1 — CSV/JSON export now reports failures instead of doing nothing, works in Firefox, and no longer leaks the exported file in memory. A blocked alarm chime says so rather than staying mute.
• ~Nav Menu v1.1 — internal trace output no longer prints in every customer’s console on page load (set window.PIFORGE_DEBUG = true if support asks for it).
• ~Gantt v1.2 · Sankey v1.1 · Period Compare v1.1 · Web View v1.1 — silently discarded errors now write one clear console line each. This includes the call that moves the display in time from a Gantt zoom, which previously failed with the chart redrawing as if it had worked.

Playback Pro v2.5 & DateTimePicker v1.2

The same functional audit applied to the two time-navigation symbols, verified against a 27-case suite that drives the actual buttons and checks what time range the display is asked for.

• ✕Playback — event markers were missing on historical playback. The event-frame query always asked for “the last 90 days from now”, ignoring the range actually being played. Replaying last January found nothing and reported “no events in lookback window”, with no hint that the query had never looked there. The query now always covers the range being played.
• ✕Playback — “Link copied!” when nothing was copied. The clipboard call’s result was never awaited, so a browser that refused the copy still reported success and the user pasted stale content. It now waits, and offers a manual copy box when the clipboard is blocked.
• ✕Playback — A-B loop ignored a range marked backwards. Marking B before A lit both buttons and silently produced no loop. The two marks are now taken in either order.
• ~Playback — “events failed (HTTP 404)” no longer advises switching to https; 404 means the database or element was not found, which sent people after the wrong problem.
• ~Error reporting — 12 more silently discarded errors across both symbols now write one clear console line each. This covers the calls that move the display in time, which previously failed with the scrubber still sliding and the clock still counting.

Calculation & Time-Range Accuracy Release

A full functional audit of the analytics symbols. Every issue below was a silent one — the symbol drew a chart and reported a number, just the wrong one. Updating is recommended for anyone using Cpk, spreadsheet formulas, or month/year time ranges.

• ✕Smart SQC Pro v2.1 — Cpk was computed against the control limits, so it reported ≈1.000 no matter what the process did. Cpk now uses real specification limits, configurable as USL / LSL; with neither set the gauge reads “—” instead of a fabricated number.
• ✕Relative time ranges — *-3mo was parsed as 3 minutes and *-1y as 1 minute. Month and year ranges now use calendar arithmetic. Affects Smart SQC Pro v2.1, Combo Chart v1.1, DateTimePicker v1.1, Playback v2.4, and Web View v1.1.
• ✕Datasheet v1.1 — six formula-engine defects: multi-column ranges such as SUM(A1:B4) dropped values, COUNTA always returned 0, MOD gave the wrong sign for negative numbers, the & concatenation operator was missing from the parser, -3^2 evaluated to 9 instead of −9, and ISBLANK/ISNUMBER/ISTEXT tested a substituted value rather than the cell.
• +Smart SQC Pro v2.1 — Specification limits (USL/LSL) added to the config panel, separate from control limits.
• ~Datasheet v1.1 — Error messages are now in English, and the SheetJS setup message gives the correct %PIHOME64% install path.
• ~The formula engine is now covered by an 83-case regression suite run against the shipped, obfuscated build.
• ~Error reporting — Combo Chart, Smart SQC Pro, and Datasheet used to discard 36 internal errors without a trace, so a failure looked identical to “nothing happened”. Anything that can degrade the symbol now writes one clear line to the browser console (once per session, so nothing floods), and a failed CSV export tells you instead of doing nothing. Combo Chart’s internal trace output no longer goes to every customer’s console — set window.PIFORGE_DEBUG = true if support asks for it.

PIForge Platform Update

• +Added Lemon Squeezy card payment (Visa/Mastercard)
• +Download limit system (10 downloads per license)
• +Visitor analytics dashboard for admin
• +Newsletter subscription
• +EULA/ToS agreement recorded at purchase (legal compliance)
• ~Improved docs page with compatibility table and FAQ
• ~Updated legal policies with anti-piracy enforcement clauses

Smart SQC Pro — Major Update

• +Added PI Vision 2022 R2 compatibility
• +New dark theme support
• +Improved rendering performance on large datasets (>10,000 points)
• +Added CUSUM (Cumulative Sum) chart type
• ~Refactored control limit calculation logic for accuracy
• ~Reduced bundle size by 30%
• ✕Fixed tooltip display bug on Firefox
• ✕Fixed chart not re-rendering on PI Vision display resize

PIForge Platform — Security Update

• ~Upgraded password hashing from SHA-256 to PBKDF2 (100k iterations)
• ~Added rate limiting on login and purchase endpoints
• +Email verification required at signup
• ✕Fixed session token not expiring after 24h

Smart SQC Pro — Patch

• ✕Fixed license key validation failing on certain server locales
• ✕Fixed symbol not loading after IIS reset in PI Vision 2021
• ✕Fixed sigma lines misaligned when data contains null values

Smart SQC Pro — Feature Update

• +Configurable Western Electric Rules (Rules 1–4)
• +Export chart as PNG
• +Configurable number of subgroups displayed
• ~Improved mobile/tablet rendering in PI Vision kiosk mode
• ~Smoother animation on data refresh

PIForge Platform — Launch

• +PIForge marketplace launched
• +User accounts, email verification, and license key delivery
• +PayPal payment integration
• +Cloudflare Pages + D1 + R2 infrastructure

Smart SQC Pro — Launch

• +Initial release of Smart SQC Pro
• +X̄-R chart, X̄-S chart, Individual-MR chart
• +Auto control limit calculation (3σ)
• +PI Vision 2019, 2020, 2021, 2022 compatibility
• +Real-time data streaming from PI tags
• +License key activation system