Guides

Reports

Pick a project’s recorded files in a calendar, open them in charts, or generate a production report: why batches miss the golden, batches that beat it, the change since the last period, capability, stops and maintenance, OEE, alerts, changes, energy, shifts, sensor health and what to do next.

Reports & analysis, in the bar above the charts, reads the files your senders recorded into a project (the collector, Python, the plugins or the browser). Every number is computed by Kymoscope in your browser, and every finding says how sure it is. See Recordings for how files are made and Cycles for how to write a cycle rule.

Pick the files

  1. Open Reports & analysis and pick the Project, then the From and To dates. The calendar lights the days with recorded data.
  2. Press Cycles and pick the cycle (with one in the library, it is already applied). For a value other than the cycle duration, choose Lower is better or Higher is better; for the cycle your scheduled reports use, the project's leads set it once for everyone. The calendar now colours each day by the average value to compare; a grey day has data but 0 cycles. No cycle yet? Suggest a cycle from the data… reads the newest recording, finds the channel that goes up and down regularly and opens the library with the rule filled in, for you to check and save; or Create a cycle in the library… by hand.
  3. Click a day in the calendar; Ctrl+click (⌘+click on a Mac) another day for a range, or pick one in Quick range. Its files are listed with their stream, time, cycles and value; tick the ones to use, or tick none to use them all.
  4. Import to chart opens each ticked file as a dataset; Batches lists their batches and Generate Report makes the report (with no file ticked, both use every listed file).

With no file ticked, the report’s period is the days you chose, midnight to midnight, so Since the last period compares a week with the week before and a day with the day before. With files ticked, the period is the time those files cover within the chosen days.

Cycles are found in your browser the first time a file is read with a cycle, and kept for the next visit by editors and leads; viewers see what is kept and get the rest for their visit.

Batches: flags, golden and CSV

Batches lists every batch in the ticked files, newest first, with its stream, start, duration and value to compare. Tick batches, then:

  • Mark good, Mark bad… or Exclude… (editors and leads). Bad and excluded batches ask for a reason, and they never become the golden. Marking is optional: reports work from the data alone.
  • Make Reference… (editors and leads), with batches of one stream. One batch becomes that stream’s golden, with a tolerance band. Several batches make a band from their mean ± k standard deviations, which becomes the golden only if the stream has none yet.
  • Export CSV downloads every listed batch: stream, value, start, end, duration, flag and reason.

Generate a report

Generate Report finds every batch in the ticked files with the applied cycle and compares each one with the golden of its stream. Each file is measured once and the result kept, so the next report over the same files is quick.

A batch is at golden level when its value is within 2% of the golden’s (or twice the spread of the golden’s own batches), and better than the golden beyond that on the good side. With no golden set, the report takes one from the ticked files: the best value 1 batch in 20 reaches (and at least 5), so neither one odd batch nor a lucky handful becomes it. When an editor or lead generates it, that batch becomes the stream’s golden from then on, and the cover says it was set automatically; for viewers it is only proposed. To use another batch, press Make it the golden on a batch a report proposes, or make a Reference in Batches; the next report measures against it.

Editors and leads save the report in the project; viewers get it for their visit and can download the PDF. In an end-to-end encrypted project the report is sealed with its own key before it leaves your browser: Kymoscope stores it but cannot read it.

What a report contains

A report is short by design: a section only shows when it has something to say. The sections with charts read the same way: the verdict first, one main chart, the findings in order of importance, and the other charts and tables under More detail. Where a section follows a standard it says so under its title (OEE as ISO 22400-2, Capability (AIAG SPC), Reliability data (ISO 14224 style), Energy review inputs (ISO 50001)): the report supports your review, it does not certify anything.

SectionWhat it tells you
00 Executive summaryThe answer on one screen. The share of batches at golden level or better, the sensors to replace and the channels drifting (two different jobs, never one sum); Where the money goes: every loss of the period in one ranked bar (stops, slow batches, bad parts or off-golden batches, the extra energy of off-golden batches), each with its amount, its share and who owns it, and a total that reads “up to” because it is what running without stops at golden level would recover. Bad parts is one row: the line time the batches outside their limits or flagged bad took, plus their material at the cost of a scrapped batch; with no specification limits to say which batches are scrap, off-golden batches are listed at that cost with an Assumed tag, outside the total (never when the value to compare is the cycle time: a slow batch is not scrap). Batches made while a sensor was failing are listed as at risk, outside the total. It is in money once the cost of an hour of line time is set in Process context, and in hours until then; a loss too small to matter is left out. Then What to do first (three actions with their owners) and the period day by day: each day from green to red by its batches off the golden, with stops and stuck sensors marked. What each section found is one press away.
01 Since the last periodThis period against the one before it, of the same length and on the same golden: the share at golden level or better, the value to compare, stopped time and the sensors to replace or drifting. A change is named only when a test says it is real. It reads the cycles kept from the period before, so it appears once that period has stored batches.
02 Why batches miss the goldenWhich channel in which phase explains the gap, and when batches miss it (hour and weekday, after stops, by stream or file, and by shift once shifts are set in Process context). A group is named only when it runs at least 5 points more off-golden than the rest. Under the main finding, Make this an alert (editors and leads) turns it into an alert rule in one press: a warning when that channel goes past the level 98 of 100 batches at golden level stay within, shown in plain words before it is created with the share of the period’s off-golden batches it would have warned on (and of the rest). It is offered only when it would have warned on most of them: a difference in timing or shape that no level catches gets no button. The rule lives in Alerts like any other, and the next report’s section 07 says whether it comes before off-golden batches. Review the batch furthest from the golden opens that batch’s review by exception, also one press from any row of Batches: one line saying what needs a look, then only what is out (the phases outside the golden’s band, on which channels and by how much in plain words, the time each channel with specification limits spent inside them, a sensor that held one value where the golden moves or stopped reading, the alerts raised while it ran) and one line for everything that is in. It supports a batch review; it is not a batch record and releases nothing.
03 Better than the goldenBatches that beat the golden, what they do differently, and a proposed new golden when it repeats.
04 Quality: capabilityPpk of each cycle KPI whose channel has specification limits (LSL, USL) in Process context, from the overall spread of at least 30 batches; capable from 1.33, or from 1.67 (what PPAP asks of a new or critical feature) when you set it in Process context, 5 · Reports. The Ppk is indicative: it assumes a stable process and normally distributed values, which a report does not test. With no specification limits on any channel there is no capability to compute: the section then shows a control chart (each batch against the mean ± 3σ of its own variation) for the feature whose batches leave those limits more than chance would, and only then. It says the process moved, not that a batch is bad. 04.2 lists the batches made while a sensor was stuck or reading beyond its band: batches to check before they ship.
05 Reliability and maintenanceA table of each sensor’s health (stuck, noisy, spikes, gaps, rate drops, out of range) and drifts forecast to the edge of the golden’s band. Several channels of one stream drifting together are one thing to check (the process: wear, a setting), with one action and one maintenance job, and channels that stepped on the same day are one row to watch. 05.4 counts the stops (gaps between recordings, or long runs with no batch) with MTBF and MTTR per stream; with machine units in Process context, per unit. 05.5 turns drifts and wear into maintenance to plan, by date, placed in the next planned stop before it. Shifts and planned stops in Process context keep planned time out of the stops. Without shifts, gaps longer than 2 hours count as time off.
06 OEE and where time is lostWhere planned time is lost (stops, slow batches and bad batches) as the section’s one chart, then availability, performance and quality with the ratios of ISO 22400-2; the table by shift and the daily trend are under More detail. The section says “OEE as ISO 22400-2” under its title: the report supports your review, it does not certify anything. Availability is the planned time not stopped (stops of 5 minutes or more). Performance is the ideal cycle time × batches over the time not stopped: the ideal cycle time you set in Process context, or, left empty, the one inferred from the data (the golden’s cycle time plus the shortest usual wait between batches), and the section says which. Quality is the batches not flagged bad and inside their specification limits; with no limits and the cycle time as the value to compare, nothing says which batches are bad, so quality reads 100% and the report asks you to add limits. Needs shifts in Process context; the cost of an hour of line time adds what the lost hours cost at most (“up to”).
07 Alerts before bad batchesWhich of the project’s alert rules fire, in the hour before a batch or while it runs, before off-golden batches more often than before the rest. Needs alert rules in the project.
08 Changes detectedEach change of a channel whose role is Adjustable setting, read from the data, with the value to compare before and after it; and each new golden in the period. Needs at least one channel marked Adjustable setting in Process context.
09 Energy, cost and CO₂Where the period’s energy goes when the energy channel is a meter (while a batch runs, and between batches, in stops and breaks); kWh per batch and in the period, against the period before once its energy is stored, and how much more off-golden batches use. Needs only the energy channel. The price per kWh adds the cost; the grid CO₂ factor adds CO₂ per batch, labelled for what it is: Scope 2, location-based, with the factor you entered.
10 Other linesThe lines named in Process context side by side, and the one that misses the golden more often. Needs two or more lines.
11 Shifts comparedOne row per shift: batches, the share at golden level or better, stopped time and alerts, and the weakest shift when a test says it trails the others; one line when the shifts perform alike. Needs shifts in Process context. Tick Don’t compare shifts in reports there to leave it out, with every other comparison of shifts.
12 Recommended actionsWhat to do, most urgent first, each with the findings it comes from.
13 Next analysesWhat stayed unexplained and the data that would explain it.
14 Method, evidence and approvalFolded at the end: how each number was computed and what each evidence level means.

Process context

Process context is a tab on the project page, after Recordings. It holds what a report needs that the data can’t say. Nothing in it is mandatory: a report uses what is there, and section 14 names what to add for the rest. Leads and editors fill it in; every member can read it.

  1. Open the project, then Process context, and press Fill from data. It reads the project’s recordings in your browser (a progress line shows what it is doing; Cancel changes nothing) and proposes units, each channel’s expected range, the energy channel, shifts and breaks, weekly stops, comparable lines and machine units. With AI writing on, it also proposes channel roles and a draft of the process text, written from the channel names, units and roles only.
  2. Check the proposals. Each is marked Suggested: correct what’s wrong, confirm each one, or press Confirm all suggested. You don’t have to wait: a report works out the shifts, planned stops, lines, machine units and energy channel from the data wherever nothing is set, uses them, and marks each section that rests on one with Assumed from the data, so you know what to confirm. A report of a day or a week takes the shifts and breaks from the last 4 weeks of recordings, so a break is not counted as a stop.
  3. Add what the data can’t know: the costs, specification limits (LSL, Target, USL) and calibration limits, and what the process is for.
  4. Press Save. Each save is a new version, and each report names on its cover the version it used.

The line at the top says what would put money in the reports, for example “add the cost of an hour of line time (losses in money)”, and under it what else adds a section, for example “add shifts to get 06”. In an end-to-end encrypted project the Process context is sealed in your browser like the rest of the project’s settings.

PartWhat to fill in
1 The processWhat the line does and how it works: the product, the steps, what a good part is, known problems.
2 ChannelsEach channel’s role (adjustable setting, measurement, quality result, outside influence, energy, ignore) and unit; its expected range, specification limits and calibration limit open under the channel. Suggest with AI proposes roles from names and units.
3 LineShifts, and Don’t compare shifts in reports (a comparison of shifts can single out people: ticked, reports leave out section 11, the table by shift in 06 and the shift pattern in 02, and shifts still set the planned time); the ideal cycle time in seconds, from one batch’s start to the next one’s (optional: left empty, reports infer it from the data and say so); planned stops, each once (a date and time), every day or every week on the days you tick (a weekly cleaning, the weekend), which reports don’t count as stops or as lost time; the streams that are comparable lines; and the channels of each machine unit.
4 CostsCurrency, cost of an hour of line time, cost of a scrapped batch, price per kWh and the grid CO₂ factor. All optional: each adds money to the reports, always as “up to”.
5 ReportsThe report style for your industry: Generic, Discrete manufacturing (stamping, machining, assembly: parts instead of batches, OEE and reliability first) Process (chemical, food, batch processes: the golden, capability and energy first) or Regulated (pharma, food: quality and the batch review first, reliability and OEE in the Quality view too; choosing it turns AI writing off, and a lead can turn it back on). It sets the order of the sections, what each role’s view shows, which sections start folded and the words a report uses; the numbers are the same in every style. It starts from the industry you chose when you signed up (Pharma & food starts as Regulated), is kept with the project the first time someone saves the Process context, and can be changed at any time: reports already made stay as they are. Then AI writing, each member’s report view, the reviewer and approver, and scheduled reports.

AI writing

AI writing is included in some plans. A project lead turns it on in Process context → 5 Reports; from then on Kymoscope’s AI words each report: a short verdict per section, at most two sentences under it, and the actions in plain words.

  • Every number and finding is still computed by Kymoscope. The AI is sent the findings and the Process context only, never your data, and Kymoscope keeps nothing of what it is sent.
  • Every sentence names the findings it rests on, and may only use numbers those findings contain; a sentence that doesn’t is left out.
  • Without AI writing the report has the same sections and findings, worded from fixed templates. If the AI can’t be used for a report, it is made with the fixed wording. Section 14 says which wording the report has.

Evidence levels

LevelMeaning
MeasuredRead directly from the data. No model involved.
Confirmed by trialA deliberate change with a before/after comparison.
StatisticalA difference between groups of batches, with its p-value: the chance of a difference this large if there were none.
CorrelationTwo signals move together: a lead to test, not a proven cause.
Confirmed by engineerAn engineer checked the finding and signed it in the report.

Views, PDF and saved reports

  • Read it as shows the Full report or only the sections for Management, Quality, Maintenance or Production. The full report tags who each section is for.
  • Download PDF gives the same report as a document, the same for everyone who opens it.
  • Reports in the bar lists the project’s saved reports beside the open one. Its author or a project lead can move a report to the trash.

Sign-off

A report can be signed by a reviewer, then an approver. A project lead names them once; every saved report then goes to them in turn.

  1. Open the project, then Process context → 5 Reports, and pick a Reviewer and an Approver from the members. Either can be left empty; with neither, reports stay Prepared.
  2. Save a report. The reviewer gets an email; its link opens the report in Reports & analysis.
  3. In the approval table of section 14 the reviewer presses Review or Reject (Reject asks to confirm). Once reviewed, the approver gets an email and presses Approve or Reject the same way.
  4. The author gets an email when the report is approved or rejected.
  • A report saved before a reviewer or approver was named is sent with Send for sign-off, by its author or a project lead.
  • The Reports list shows each report’s status: In review, Awaiting approval, Approved or Rejected.
  • A rejected report stays as it is. Make a new one with the changes.
  • Each signature is shown with the signer’s name, the date and what it means: Reviewed, the signer read the report and found its content correct; Approved, the signer accepts it as final. A signed report’s Download PDF adds a sign-off page with the same, and the content hash of the report that was signed.
  • Optional: in 5 Reports a lead can tick Ask the reviewer and the approver for their password again before they sign. It is off unless you tick it. With it on, a signature counts only within 5 minutes of the signer proving who they are (the password field beside Review and Approve; with Google sign-in, sign out and in again first), and the sign-off page says so.
  • Kymoscope records who signed which step and when. It does not certify compliance with any regulation.

Scheduled reports

A project lead can have reports made on a schedule, so the customer’s team gets them without asking.

  1. Open the project, then Process context → 5 Reports.
  2. Give each member a report view: Management, Quality, Maintenance, Production or None.
  3. Pick the Cycle for scheduled reports and choose Lower is better or Higher is better. Until a cycle is picked, the schedule can’t be ticked.
  4. Under Scheduled reports, tick Daily, for the day before, Weekly, on Mondays, Monthly, on the 1st or Yearly, on 1 January, and pick the Time zone. The tab shows when the next one is made.
  5. Choose Email every report or Email only when something needs attention, and whether to Put the verdict and the first actions in the email.

A lead can also open any report and press Get this report every Monday: the weekly report then uses that report’s cycle and direction.

Reports are made at 07:00 in that time zone for the period that just ended: the week Monday to Sunday, the calendar month or the calendar year. Nobody needs to open Kymoscope: Kymoscope report service makes them, and the cover reads Made by Kymoscope report service. A daily report covers the day before, midnight to midnight.

  • What it reads. Recordings sealed with the project’s key, kept by Kymoscope or in the customer’s own S3 storage. Recordings sent with a stream password, or kept in a folder on a computer, are not read; their cycles and measures already saved still count, and the report says how many recordings weren’t read. In a project that isn’t end-to-end encrypted, scheduled reports use the cycles and measures already saved.
  • Who gets it. Each member with a report view gets an email; its link opens the report on their view. With a reviewer or approver named, the report goes through sign-off like any other.
  • Only when something needs attention. With that choice, the email goes when the report has an urgent action or a significant change since the period before. Otherwise the report is still made and listed in Reports & analysis, with no email.
  • The verdict in the email. With that choice, the email shows the report’s verdict and its first three actions, so many readers don’t need to open the report. In an end-to-end encrypted project, that text travels in plain text, outside the encryption.
  • If it can’t be made. The service tries three times. Then the project’s leads get an email with the reason.