What the Flyt Format Covers Beyond Flowcharts

Flyt 0.1 was published on 11 August 2026 and describes one diagram: a process flowchart captured from a photograph rather than authored in a tool. Version 0.2 followed on 12 August and adds the swimlane profile, which is the second diagram read from the same document without conversion.

Three other diagrams come off the same walls. A room that has finished mapping a process sorts the problems into groups, argues about which cause produced which effect, and walks the floor tracing how far a part travels. Each of those is photographed by somebody who was absent when it was drawn, days later, in poor light, with a note or two missing.

The apparent next step is a format for each. The arithmetic argues against it.

What four formats would duplicate

The difficult half of a capture format is the record of how the capture knows what it claims:

• provenance — which photograph an element came from, and which region of it
• confidence — how sure the reader was, per element, because a blurred sticky note is a weaker fact than a clear one
• the legend — a color and a shape carry meaning on a given wall, and the document carries that mapping or the typing cannot be checked
• the coordinate space — what the numbers in a position mean, and whether they can be compared across two photographs
• determination — whether the document records enough to be argued with

Each of those is common to all four diagrams. An affinity diagram requires every one, and requires position more urgently than a flowchart does. Four formats means four implementations of that machinery, and separate implementations diverge: the cause-tree reader arrives at a different interpretation of confidence than the flowchart reader, and the divergence surfaces years later, when somebody tries to reconcile two documents.

One core with a profile for each family avoids that. The core stays fixed.

The four families

Sorting by diagram type produces a list that grows without limit. The useful division is by what carries the meaning, because that is what a reader has to process differently.

Sequence. The edges carry the meaning as order. Flowchart, swimlane, value stream. A traversal of the edges describes a route the work takes.

Grouping. Membership carries the meaning, and the document has no edges. Affinity diagrams, card sorts, themes in buckets. Proximity is the content.

Causality. The edges carry the meaning as claims. Five whys, fishbone, interrelationship digraph.

Space. Coordinates carry the meaning, in real units. Spaghetti diagrams, floor layouts, walk routes. The distance is the finding.

Why sequence and causality are kept apart

The two are indistinguishable in a data structure. Both are nodes joined by directed edges. A traversal of a causal diagram assembles an argument, and a tool that costs that traversal as though it were a route reports a cycle time for a fishbone while giving no sign that anything went wrong.

Flyt already carries an annotation edge kind for a smaller version of the same problem. It exists because the wall behind the 0.1 specification turned out to have 23 of its 93 relations be notes stuck beside steps rather than flow between them, and a cost computed across that map passes through a problem statement on its way out.

What the core already carries

The core requires little change to hold the other three families:

That reads as foresight. It was accumulated by meeting one wall and being wrong about it. The annotation kind exists because of the 23 relations above. The rule that a document deriving an edge from position must record that position exists because the wall in question carried 91 sticky notes and no drawn arrows at all — every connection in the capture was a conclusion somebody reached from where things sat, and the first draft recorded the conclusions while discarding the evidence for them.

What one wall taught is what the other three families require.

What the swimlane profile settled

A swimlane diagram sorts the steps into rows, one per person, department or system. Drawing tools treat that as a layout feature. A room draws the lanes to make the boundaries countable, and the profile defines the count: a handoff is a non-annotation edge whose endpoints share no lane, derived and never stored.

Two rules in it exist because an implementation gets them wrong quietly. A step sharing any lane with its successor is not a handoff — a step owned by planning and scheduling that hands to scheduling never left the room. And a step in no lane is excluded from the count rather than treated as a lane of its own: on the worked document published with the profile, one unassigned step out of ten reports seven handoffs where there are three.

Lane membership is itself derived. On a captured wall, which lane a step sits in was read off where the note sat relative to a ruled line, and the handoff count is computed from lane membership alone. A capture that assigns lanes by eye without saying so produces a number that reads as a measurement.

0.2 is four optional keys. A 0.1 document becomes a 0.2 document by changing its version string and nothing else, and each version's schema is served at its own address, so a document written against 0.1 keeps a schema to be read against.

What the value stream profile settled

A value stream map is a flowchart with numbers on it and a timeline along the bottom, setting the time a part spends being worked against the time it spends in the building. Making that computable turned out to require two refusals.

A missing measurement makes a total undefined. The arithmetic is addition, and addition has an identity that looks exactly like a measurement of nothing: a step whose cycle time was never recorded contributes zero, and the total is then printed with a unit on it, inside a percentage, beside a target. On the worked map published with the profile, rubbing out two numbers and summing anyway gives a process cycle efficiency of 0.0167 % against a true 0.0117 % — overstated by 43 %, from a map that still reads as complete. A total is defined only when every member of the set carries the measurement, and a floor may be reported so long as it is named a floor.

A monthly demand cannot become a daily demand without a working calendar. Demand is written on a wall per month and every division that follows needs it per day. Thirty calendar days gives that map a takt time of 88.0 seconds; twenty-one working days gives 61.6. The weld cell runs at 62 seconds, so the first number says the line has room and the second says it cannot meet demand. The invented divisor reverses the answer, and takt is what every cycle time on the map is read against.

The second rule was not in the draft. It appeared while implementing the first.

What the affinity profile settled

An affinity diagram is the first of these where the content is the grouping. In a flowchart the steps exist whether or not anybody draws them. A cluster is a conclusion somebody reached by moving paper, so the document keeps three layers apart that the picture merges: what was written, how it was grouped, and what each group was called — made by the room, by the room, and by one person afterwards, in descending order of reliability.

It is also the purest case of the rule about derivation. A swimlane has a ruled line to check a reading against; a cluster has nothing but a region where some notes sit closer to each other than to anything else. A capture that keeps the clusters and discards the coordinates has destroyed the only evidence its claim ever had.

A count of notes is a measurement of the room rather than of the process. It measures how many people found a thing easy to articulate. A room writes six notes about something everybody has words for and one about something harder to say, and the one-note cluster is routinely the expensive one. On the worked wall published with the profile, the room put eleven dots on a three-note cluster and four dots on a five-note one, so reading the diagram by size would have put the room's own answer fourth. Where the vote and the count disagree, the disagreement is the finding.

Release order

Six releases, each a profile with a worked example:

1. 0.1 Flowchart — published 11 August 2026
2. 0.2 Swimlane — published 12 August 2026. The same document, unconverted, read as a second diagram
3. 0.3 Value stream — published 12 August 2026. Measurements with units, so a captured wall becomes an input to a calculation
4. 0.4 Affinity — published 12 August 2026. Nodes, groups, no edges, with position doing all the work
5. 0.5 Causal — five whys and fishbone, with the rule against costing a traversal as a route
6. 0.6 Spatial — spaghetti and floor layout, in meters

Three things would change that order, stated so that changing it later is visibly a decision. A customer with a real wall of one of these kinds: every correction to 0.1 came from meeting an actual capture, and none came from planning. A second implementer finding something hard, whose difficulty outranks this list. Or a family that requires something the core cannot express, which would be worth publishing on its own.

Terms

The specification is CC BY 4.0. The schema is CC0. Implementation is open to anyone, for any purpose, and requires no agreement with anybody. A format that cannot be cited is one company's file layout with a longer document attached.

The specification, the schema and the family are at svend.ai/flyt.

This bulletin is published by SVEND, which also publishes the Flyt specification described above.