The Cockpit is SKL's control tower. In under 60 seconds it answers three questions: Will we hit this quarter's objectives? What's slowing execution? What's waiting on me? The switcher at the top opens a department view: Marketing, Sales or Finance.
2How it makes everyone more efficient
Manage by exception. What's on track stays quiet; only gaps take up space. Attention goes where it changes the outcome.
Project, don't just observe. Every KR shows its forecast landing and required pace. We correct in week 4, not week 13.
One signal, one action. Every alert has its button: decide, follow up, reassign. Nothing is left hanging.
Visible bottlenecks. Blockers, team load and pending decisions, including the founder's, are all measured.
Zero prep. Saturday's Cockpit is the weekly review agenda.
3How to use it
Each morning, scan the KPI row: cycle time, promise reliability, blockers, load, decision speed.
Handle “Waiting on you” first: these are blockers only you can clear.
Move to “What's slowing execution” and trigger the suggested action.
Review the KRs: a red KR requires a recovery plan within 48 h; a KR with no check-in for 7 days gets a follow-up.
Check explorations: a used-up timebox forces a decision (continue, pivot or stop).
On Saturdays, walk through the Cockpit as the weekly review agenda.
{{eyebrow}} {{facilitator}}{{statusPill.label}}
Step {{cur.n}} of 7 · {{cur.min}}
{{cur.title}}
{{cur.why}}
{{cur.count}}
{{c.name}}
{{c.owner}} · last: {{c.last}}
{{c.badge.label}}
Key resultQuarter trajectoryRatioForecast · next milestoneStatus
No weekly milestone due this week or last week. They are set in the Roadmap, one person, one week, one verifiable result.
{{m.eyebrow}}
{{m.formula}}
{{m.meta}}
{{m.badge.label}}
{{m.count}}
{{m.hint}}
No promise was recorded last week. They start with step 7 of this review.
{{p.owner}}
{{p.text}}
{{p.hint}}
Promises are due by Saturday {{nextSat}}. At most 3 per person. Load above 70% means something else must give.
{{p.name}}{{p.role}}
{{p.load}}
{{p.loadText}}
{{i.text}}
{{i.meta}}
No promise yet
Promise reliability
{{reliability}}
{{reliabilityText}}
{{b.pct}}
{{b.label}}
WeekDateKRs at closingPromises keptAlerts triagedNew promises
No review closed yet.
Promises of the week before · facilitated by {{h.by}}
{{x.who}}{{x.text}}{{x.res}}
Promises made for the next week
{{x.who}}{{x.text}}
{{h.gaps}}
Page method
Weekly Review
1What this page is for
The Weekly Review is the Saturday ritual that keeps the quarter on track: 45 minutes, the whole team, always the same 7 steps. Everything is read from the shared data (key results, alerts, decisions, tasks) and everything decided in the room is written back to it. Each closed review is kept in History with a snapshot of the key results and the promises made.
2How it makes everyone more efficient
Numbers before opinions. Every KR is checked in before it is discussed.
No red KR without a plan. A red or untracked KR leaves the room with a plan, an owner and a date.
Nothing stuck survives Saturday. Every Critical and High alert still open gets an outcome in front of the team.
Milestones closed on facts. Each weekly milestone is judged on its mini goals: all ticked is done, anything else is not done with its cause (estimation, dependency, priority, skill, commitment).
Promises you can count on. Each person commits to at most 3 things, checked against their load. Next Saturday they are scored kept or missed: that score is the team’s promise reliability.
3How to use it
Check-ins are done on Saturday: each KR owner saves the check-in in step 1 (or in My Day) before the review. The red count in the sidebar shows the KRs still missing.
On Saturday, the facilitator runs the 7 steps in order on a shared screen. The left rail shows what is done.
Step 5 ticks the mini goals of each weekly milestone, then marks it done or not done. Steps 6 and 7 are required. Steps 1 to 5 can stay open: the gaps are recorded with the review.
Close the review. It moves to History, and next Saturday its promises come back in step 6.
Board
Target lead time: 48 h if reversible · 7 d if irreversible
{{col.title}}
{{col.count}}
{{col.hint}}
Nothing here
Decisions
Archive
Every decision taken more than 2 weeks ago leaves the board and lands here automatically. {{archiveCount}} archived.
DecisionOwnerOption chosenDecidedAge
Page method
Decisions
1What this page is for
This is the backlog of every SKL decision: pending, made and up for review. Each decision has its record: the question, the cause to address, the options, the assumption, the expected outcome, a single decision owner and a deadline. Each record is linked to the key results it serves, the tasks it blocks or triggers, and the deliverables it produces or changes.
2How it makes everyone more efficient
No more issues that drag on. 48 h for a reversible decision, 7 days for an irreversible one. Past that, an alert is raised.
One decision owner. Every decision has a named owner: no more re-debating the same topics every week.
Full traceability. Links show what a decision unblocks, which result it moves and which documents it changed. Nobody works on a task that waits for a decision without knowing it.
Decisions that get better. The 30- and 90-day reviews compare the actual outcome to the expected one and capture the lesson.
3How to use it
Read the board left to right: To decide, Decided, To review. Click a card to open its full record on the right.
Every decision follows 5 steps: the problem, 3 weighted variables (weights add up to 100), the context of each variable, 3 options scored on each variable, then the roadmap (tasks, key result, deliverables).
Decisions taken more than 2 weeks ago move to the Archive, next to “New decision”.
Any question that blocks at least one task becomes a decision.
Frame it as a choice, with at least two real options, and state whether it is reversible.
Link it: the key result it serves, the tasks it blocks, the deliverables it will produce or change.
The decision owner decides before the deadline and writes down the expected outcome.
Every exploration that reaches the end of its timebox is closed out here: continue, pivot or stop.
In “To review”, compare actual to expected, note the lesson and update the linked playbook.
Page method
CRM
1What this page is for
The sales CRM, taken as is from ALKIMIY Lite: every prospect from first contact to cash collected, with its next action, its calls, its transcripts and the reason it was lost.
2How it makes everyone more efficient
No prospect falls through. Every lead carries a next action with a date; late ones stand out.
The pipeline is honest. Stages, rounds and loss reasons are explicit, so the forecast rests on facts.
Cash, not just deals. Closed deals show invoiced and collected amounts side by side.
3How to use it
Pipeline: drag a card to move a prospect to its new stage.
Leads: filter, sort, pick your columns and act on several leads at once.
Sales cycle: one line per live lead, from the booked call to the first Whop payment, against the Show, Decision and Cash targets. Each morning, read Critical, then Sales must act, top to bottom; Calendar shows today and the next calls, Since booking compares cycle lengths; open a line to read every interaction, or the lead record for every message, call and payment.
Click a prospect to open its file: calls, next action, note, transcripts and analysis.
Log the next action before you close a prospect's file.
IP SKL intellectual property · click a deliverable to open it
Marketing
{{count.marketing}}
New
Nothing filed yet
Sales
{{count.sales}}
New
Nothing filed yet
Delivery
{{count.delivery}}
New
Nothing filed yet
Product
{{count.product}}
New
Nothing filed yet
Finance
{{count.finance}}
New
Nothing filed yet
People
{{count.people}}
New
Nothing filed yet
Operations
{{count.ops}}
Frameworks
New
Nothing filed yet
Page method
Library
1 · What this page is
The home of SKL's validated deliverables and intellectual property. It is organised in the same domains as the Mechanism map: Marketing, Sales, Delivery, Product, Finance, People and Operations. Operations holds the four frameworks of the Iteration system: Iteration system, Mechanisms, Registers and Organs.
2 · Why it matters
One reference. Everyone reads the same method, in the same words, at the same version.
Protected intellectual property. Each framework is SKL IP and is filed as such.
Theory next to practice. The Mechanisms group applies these frameworks to Pre-call and Sales.
3 · How to use it
Find a deliverable in its domain column.
Click it to open it in full: definitions, diagrams, tables and rules.
Close with the × button, the Close button, a click outside, or Escape.
“New deliverable”, top right, files a new piece: import its file (any format), pick its type (framework, memo, playbook, template…), its domain, owner and status, and tag it IP if it is unique to SKL. Open shows the file; PDFs and images also preview in the window.
Modify, in an open deliverable, changes its type, domain, title, owner or status, or replaces its file.
Drag a deliverable onto another domain column to move it there.
Framework 01 · Knowledge System · IP
Iteration system
The base sits on top: mechanisms are what you are trying to make more efficient, and everything below exists to serve them. The loop reads the base, works left to right, and every result returns into it.
4 steps3 exits10 blocks
What it is
one sentence, four steps
An iteration system is a closed loop that reads a base of mechanisms and registers, turns each friction into a decision someone can make, tests one lever at a time, and writes every result back — amending, correcting or extending the very base it reads.
The base sits on top. The loop reads it, works left to right, and every result returns into it.
An open loop is not an iteration system. Running tests without writing the results back produces activity, not learning: every cycle restarts from the same level of ignorance and the same arguments come round again. The write-back is what makes the loop reinforcing — richer base, better-founded hypotheses, fewer inconclusive tests, capacity freed, more iterations closed.
Every block of the loop
role, how it works, what it is not, rules
the base
Mechanisms
Its role
What runs and produces a result: chains and systems. This is what everything else in the arrangement exists to make more efficient.
How it works
Each mechanism is modelled with 21 organs, serving five functions. The loop reads them at diagnose and writes to them at write back — those are the only two points of contact.
What it is not
Not a foundation you pour once. If the write-back never touches it, the loop is open and every cycle restarts from the same level of ignorance.
Rules to apply
A mechanism is ready for the loop once all five functions are served and the two laws are checked. Below that, diagnosing produces noise.
One primary result per mechanism. Two competing results mean two mechanisms, or a trade-off nobody made.
No two mechanisms produce the same thing. If they do, the boundary between them is badly drawn.
A mechanism is only worth formalising if its delay exceeds your patience. What answers in two hours is steered by eye.
the base
Registers
Its role
What belongs to no mechanism in particular. Four shared lists: interfaces, resources, events, owners.
How it works
A register holds an object exactly once. Mechanisms populate them; the loop reads them at diagnose to check the two blockers that no single mechanism can see.
What it is not
Not a preamble. Create them before modelling so shared objects land in them, but their content comes from the mechanisms — populate resources first and you list what you believe is scarce rather than what your mechanisms contend for.
Rules to apply
What belongs to nobody is watched by nobody. That is the whole reason a register exists.
Events and owners start immediately: a date cannot be reconstructed, and authority is a fact you already know.
Resources and interfaces fill themselves as mechanisms arrive. An interface cannot be written before two mechanisms touch.
Position decides where an object lives, not its dynamics. A shared stock — cash, context base, audience — belongs in resources.
layer
Measure
Its role
Make quantities observable, continuously. It sits under the base rather than inside the loop.
How it works
Sensors read on their own cadence, independent of whether an iteration is running. Every value carries its evidence level and its date: measured, benchmark, hunch.
What it is not
Not a step of the loop. Making it one implies you measure once per cycle, which is exactly what the noise rule warns against — you need a series, not a reading.
Rules to apply
Instrument the accumulations, not only the output. A swelling stock precedes the effect instead of recording it.
Measure noise over a period without intervention, otherwise you measure your own corrections.
Three readings minimum before trusting a trend.
A value past its shelf life changes level: a six-month-old measurement is a historical benchmark.
step 01
Diagnose
Its role
Turn the base into a finite list of named problems. The only step that runs entirely on structure — no judgement, only comparison.
How it works
Two laws are checked first: cadence against total delay, dead band against sensor noise. Then the ignorance map ranks what you do not know by what it carries. Then eleven friction rules read the organs — conservation, clock mismatch, saturation, blind gap, runway, and the rest.
What it is not
Not a review meeting. Nothing here needs an opinion. If a rule requires judgement to fire, it is not a rule, it is a discussion.
Rules to apply
Laws before heuristics. A broken law makes the reading of every other friction unreliable.
A friction that points at no organ is not a friction.
Rank by ignorance weighted by what it carries, never by expected gain.
Diagnose across the whole base. Two of the four blockers are invisible from inside a single mechanism.
step 02
Decide
Its role
Convert a friction into a call someone is allowed to make. This is where diagnosis becomes action.
How it works
Each friction carries a blocked question, two or three options with their cost and their owner, and the caller taken from organ 05. The reversibility lane is set here: reversible under a week goes light, committing goes through the full gates.
What it is not
Not a prioritisation exercise. Ranking comes after elimination — a decision blocked by a saturated resource or a slow stock is removed before anything is scored.
Rules to apply
No decision without a named caller. A decision everyone owns is one nobody makes.
Two options minimum, and one of them is always « accept and document ».
Eliminate before ranking. The four blockers are checked before any scoring.
The lane is chosen before the test, never after seeing the result.
step 03
Iterate
Its role
Move one lever and read what happens, under conditions where the reading can mean something.
How it works
One lever, one window, one measurement comparable to the one before. The test runs at least as long as the turnover of the stock it crosses. The decision criterion is written before it starts and applied unchanged afterwards.
What it is not
Not an experiment for its own sake. A test that cannot come back negative is not a test, it is a rollout with extra steps.
Rules to apply
One lever per cycle. Two make attribution impossible and the cycle is wasted.
Duration is set by the stock, never by patience.
The criterion is written before the start and applied unchanged after.
Log every context change during the run. That log separates a wrong mechanism from a polluted test.
step 04
Write back
Its role
The only step that closes the loop. It is what makes the whole thing compound rather than repeat.
How it works
Three exits: amend a value, fix the structure, extend the base. Whatever the verdict, something is written — a failed test still moves the prior.
What it is not
Not documentation. Documentation records what happened; the write-back changes what the next diagnosis will read.
Rules to apply
Write back whatever the verdict.
Fixing structure invalidates every earlier diagnosis on that mechanism. Re-run them.
Extend only when the clock test forces it. If the lever is merely outside the boundary, redraw the boundary instead.
Any event during the run goes to the events register, with its date.
exit 1
Amend values
Its role
The ordinary case, and the most frequent. A measured value replaces a hunch, a delay is finally known, a prior moves.
How it works
The value and its date go into organ 19 of the mechanism concerned. Its evidence level changes: hunch becomes measured, or measured becomes a historical benchmark once past its shelf life.
What it is not
Not a correction of the model. The structure stands; only a number moved.
Rules to apply
Nothing you concluded before becomes wrong. The next diagnosis simply reads better numbers.
A value error is self-healing: measure again and it moves toward the truth.
Always carry the date. A value without one cannot be aged out.
Record the posterior even when the verdict is inconclusive.
exit 2
Fix structure
Its role
The test revealed the model itself was wrong: a missing stock, a boundary drawn too narrow, an assumption of independence that fell.
How it works
Correct the organ concerned — 01 for the boundary, 08 for a stock, 20 for an assumption — then re-run the diagnosis on that mechanism from scratch.
What it is not
Not an amendment. A shape error is a different animal from a value error and must not be treated as one.
Rules to apply
Every earlier diagnosis on that mechanism becomes suspect. Frictions you dismissed may now fire.
A shape error never self-corrects: each iteration refines the parameters of a wrong form and confirms it with growing confidence.
This is why the ignorance map ranks shape above value, despite being far rarer.
One structural fix outweighs fifty amendments.
exit 3
Extend base
Its role
A mechanism that did not exist has to be created. The library grows from evidence rather than intention.
How it works
Almost always because the clock test forced a split: two halves running at speeds differing by more than a factor of five, so one cadence cannot pilot both.
What it is not
Not the answer to a lever sitting outside the boundary. That calls for redrawing the boundary, not for a new mechanism.
Rules to apply
Extend only when the clock test forces it. Otherwise redraw.
Record why: split, or redrawn boundary. Nothing else explains why a mechanism exists.
New interfaces appear and resources gain one more claimant. Update both registers.
Multiplying mechanisms because a lever was out of reach is how a library stops being maintainable.
One turn, in practice
what a Monday morning looks like
1
Open the control screen
Two laws first: cadence against total delay, dead band against sensor noise. If either is broken, stop — every other reading is unreliable until it is fixed.
2 minutes
2
Read the ignorance map, not the friction list
It is ranked by what each unknown carries: an unserved function, then shape, then bound, then value. A shape assumption outranks a wrong number even though it is rarer.
5 minutes
3
Take the top item to the decision backlog
Write the blocked question, two or three options with their cost, and the caller. Then set the lane: reversible under a week, or committing.
10 minutes
4
Check the four blockers before launching
Slow stock, blind sensor, saturated resource, busy owner. Any one of them and the test cannot produce a usable answer — drop it and take the next item.
2 minutes
5
Run one lever, for at least the turnover of the stock it crosses
Write the verdict criterion before you start. Log any context change during the run — that log is what will separate a wrong mechanism from a polluted test.
days to weeks
6
Close it, whatever the verdict
Record the prior and the posterior. Then take one of the three exits: amend the value, fix the structure, or extend the base. An iteration closed without a write-back leaves nothing.
10 minutes
Everything before the run takes under half an hour. The expensive part is the test itself, and its duration is set by the stock it crosses rather than by your patience. If a turn is taking longer than that on the desk work, the model is being rebuilt rather than read — which means it was not ready to enter the loop.
Where the gap comes from
the loop has two entry points
First pass
You name it: an observed value, a target, a date. This is phase one of modelling, not a step of the loop — it happens once, when the mechanism is built.
Every pass after that
The previous write-back hands it to you. A value moved, an assumption fell, a structure was corrected — and the next diagnosis reads a different base, raising frictions the last one could not.
Why it matters
« Starting from the mechanism models what is available. Starting from the gap models what matters. »
What the loop measures about itself
built into the decision backlog
The flow layer
Work in progress, cycle time, throughput, inconclusive rate. Little's law: cycle time equals work in progress divided by throughput, so every extra decision held in test lengthens all the others proportionally. The WIP limit is the only lever that shortens cycle time without changing anything else.
The reversibility lane
A change you can undo in a week and one that commits three months should not carry the same overhead. Reversible ships and gets logged; committing goes through the full gates with a named caller. Without the split the method is too heavy everywhere, which is how method tools die by month three.
Confidence, not verdict
A binary verdict throws information away — two inconclusive tests can carry very different evidence. Recording the prior and the posterior turns a failed test into a partial answer, and lowers the confidence of every future iteration resting on the same mechanism.
The iteration system is itself a mechanism. It has stocks — decisions open, in test, pending — a delay, and noise. It is the one thing in the company that would otherwise be exempt from its own method, which is why the backlog carries these numbers rather than a task list.
Framework 02 · Knowledge System · IP
Mechanisms
A mechanism is a chain or a system. Same organs in both — what separates them is a single return path, and everything else follows from that.
2 shapes4 roles6 phases
Definition
one sentence
A mechanism is a chain or a system — a set of elements that interact to produce an intended result, regularly enough that it can be steered rather than rebuilt every time.
Interact
This is what separates a mechanism from a sequence. If the elements follow one another without ever influencing each other back, you have a chain of steps, and its behaviour is deducible from its parts.
test — is there at least one closed loop?
Intended result
Intended, not pursued. A mechanism wants nothing: it produces a result, and that result is not always the declared one. The distance between the two is an object in its own right — organ 03.
test — what does it actually produce when pressure rises?
Regularly enough
This is what makes steering possible. You only steer what repeats: on a one-off event there is nothing to tune, only a decision to make once.
test — does it start again, at an identifiable cadence?
The mechanism is not the model. The mechanism exists and runs whether you wrote it down or not: your sales mechanism is turning today with its sales reps and its CRM. The model is your representation of it, and it can be wrong. Conflating the two makes it impossible to say a model is mistaken, hence impossible to correct it.
The two shapes, same organs
same notation as the model graph
Chain — matter goes one way and never comes backSystem — one effect returns onto its own cause
matter flowrule governing a flowmeasurementcommand
Only stocks carry matter. A rule does not sit in the path — it governs a flow from the side, which is why it is drawn with a dotted line onto the arrow rather than as a box between two stocks. Same for levers and the policy: they command, they do not transport.
The same matter path, plus four organs that close the loop. Sensors, gap, policy and levers are what the system has and the chain does not — and all four are constructs: they do not exist in the real thing, you build them. A chain does not become a system by being more complicated. It becomes one when you build the loop it was missing.
Chain or system
both are mechanisms, they differ in depth
Chain
System
What defines it
Steps follow one another; no effect comes back onto its cause.
At least one effect returns onto its own cause.
Its behaviour
Deducible from its parts. Three steps at 80, 70 and 60 per cent give 33.6 per cent.
Not deducible. Every element behaves as predicted and the whole contradicts the prediction.
What surprises you
Nothing. Improve one step and the result moves proportionally.
Gains that eat themselves, problems that return despite corrections, two-stage behaviour.
Modelling depth
The five functions are quickly served. Gains add up.
The full twenty-one earn their keep. Surprises come from how the loops compose.
The test
Follow an effect. It never comes back.
Follow an effect. It comes back onto its cause.
The loop is a deliverable, not an entry requirement. A chain enters your library as a candidate. The second role — absorbing variation without mobilising anyone — needs a regulation loop: a dead band, a cadence, a policy. So modelling takes something that already runs and grafts on the loop it lacks. That is why four organs are typed « construct »: sensors, gap, policy and owner do not exist in the real thing, you build them.
What is not a mechanism
five things people model by mistake
Not this
Why, and where it belongs instead
The question that settles it
A tool
A CRM, a spreadsheet, an automation. It carries no matter of its own — it is where other mechanisms' sensors live. Operations register, or organ 15.
does it produce a result, or does it hold someone else's data?
A team
People are an owner and a capacity, not a mechanism. Modelling « the sales team » rather than « the sales call » draws a boundary around an org chart instead of around a loop.
would the boundary survive a reorganisation?
A project
It has an end. A mechanism starts again. A launch is a project; what it leaves running afterwards may be a mechanism.
does it recur at an identifiable cadence?
An event
It does not repeat, so it cannot be steered — only logged. Anything that returns every year is not an event either: it is an unmodelled input.
will it happen again in the same form?
A goal
« Growth », « quality », « being the best ». A result is not a mechanism; the thing that produces it is.
can you name what enters and what leaves?
The common thread is that none of them has an input and an output. That is the cheapest test available: name what enters, name what leaves, and say at what cadence it starts again. If you cannot do all three, whatever you are looking at belongs in a register, in an organ of some other mechanism, or nowhere.
The four roles
each with its test
1
Produce its result repeatably
The first term of the definition put to the test. A project ends, a mechanism starts again. Until the result has a name, a unit and a cadence, you have an activity, not a mechanism.
test — can you name what it produces, in what unit, at what cadence?
Checked by · answered by organs 10 and 02 — outputs give the unit, the declared objective gives the cadence
2
Absorb variation through its own interactions
The role everyone forgets, and it follows directly from the word « interact ». A well-tuned mechanism takes small variations on its internal loops without mobilising anyone. If it demands a human call on every gap, it is not a mechanism: it is a queue of decisions dressed up as a process.
test — what happens when the gap stays small? If the answer is not « nothing, it is written down », the role is not met.
Checked by · answered by organ 17 — the dead band, and the inaction written next to it
3
Signal its own drift before the result shows it
Interactions produce delayed effects: by the time output falls, the cause is often weeks old. A swelling stock precedes the effect instead of recording it, which is why a mechanism must be instrumented on its accumulations, not only on its output.
test — what signal appears first when this mechanism starts to degrade?
Checked by · answered by organs 08 and 15 — a stock with a sensor on it, not just a sensor on the output
4
Be runnable by someone other than you
A mechanism only you can run is not a mechanism, it is a personal skill. This is the role that decides whether the company can grow: delegating assumes a named owner, a written cadence and a fixed dead band.
test — how long before someone else takes it over, and what do they need that is not in your head?
Checked by · answered by organs 05 and 17 — a named owner who can hold a refusal, and a policy decided in cold blood
Ready for the loop
the gate between building and iterating
A mechanism enters the loop once all five functions are served and the two laws are checked. Not twenty-one organs.
The five
Frame, Matter, Machinery, Control, Epistemic. Every organ serves exactly one, and nothing in the diagnosis works while one is unserved.
Below the gate
Diagnosing produces noise. The rules have no inputs, so the frictions they raise are artefacts of what is missing rather than of what is wrong.
Above the gate
The loop turns with most organs still empty — and it fills them itself, through the write-back. Completeness is an output of iterating, not a prerequisite.
The gate is the same for a chain and for a system. Depth of modelling differs — a chain rarely needs every organ, a system usually does — but the five functions are what the rules read, and no diagnosis runs without them either way.
Modelling
6 phases, each producing its share of the model
1
Frame the gap
An observed value, a target, a date. Until all three are written, you do not have a problem but a concern, and nothing downstream can trigger a decision.
produces · Organ 16 · Gap, and its target term, organ 02
2
Draw the boundary
Draw it around a complete loop, never around a team. Then check the clock ratio: two halves running at very different speeds are two mechanisms.
produces · Organ 01 · Boundary, and its complement organ 21
3
Trace the matter
Outputs first, then stocks, then flows, then inputs. That order is deliberate: starting with inputs models what is available rather than what matters.
produces · The model graph — organs 06, 08, 09, 10
4
Open the machinery
Transformation rules, the split between levers and frozen parameters, delays. Delays come last because they depend on the stocks identified in the previous phase.
produces · Organs 04, 05, 11, 12, 13, 14
5
Close the loop
Sensors with their latency and noise, gap, policy. The policy is set last because its cadence is computed from the total delay established just before.
produces · The control apparatus — organs 15, 16, 17, 18
6
Qualify
Evidence level for every value, structural assumptions. This phase produces no new component: it annotates what exists, and it is the only one that stops a model being relitigated forever.
produces · The ignorance map — organs 19 and 20
Library rules
once you have several
Each mechanism has one primary result. Two competing results mean either two mechanisms, or a trade-off nobody wanted to make.
No two mechanisms produce the same thing. If they do, the boundary between them is badly drawn, and you will have permanent attribution fights over results.
A mechanism is only worth formalising if its delay exceeds your patience. On something that answers in two hours, trial and error beats the model.
Framework 03 · Knowledge System · IP
Registers
A mechanism holds what is inside it. A register holds what is between mechanisms, or above all of them. Four registers, and the reason they exist is blunt.
4 registers3 positions2 blockers
What a register is
one sentence
A register is a single shared list of what belongs to no mechanism in particular — because what belongs to nobody is watched by nobody, and described inside each sheet a shared object is entered twice and the two copies diverge.
The failure this prevents, in one example. Your pre-call mechanism reports 61 leads leaving. Your sales mechanism declares 90 arriving. Twenty-nine units vanish between them, and neither sheet is wrong on its own — each is impeccable inside its own boundary. Nobody does the subtraction because there is nowhere the two numbers meet. In an interface register there is one number, and the discrepancy is impossible to miss.
Where an object lives
position decides, not dynamics
Inside a mechanism, between mechanisms, above all of them.
Position
What sits there
Held by
Why
Inside a mechanism
Stocks, flows, levers, sensors. They belong to one mechanism and live in its organs.
organs 06 to 18
No register. Duplication is impossible — there is only one owner.
Between mechanisms
Interfaces, when a flow passes from one to another. Resources, when several draw on the same pool without passing anything.
interfaces · resources
Belongs to neither. Written inside each, it is entered twice and the copies diverge.
Above all mechanisms
Events, which break comparability everywhere at once. Owners, whose authority cuts across sheets.
events · owners
Belongs to no mechanism at all. Only a shared list can hold it.
The four registers
what each holds and detects
Interfaces
the point where what leaves one mechanism enters another
What it holds
Interface point, sender or senders with their share, receiver or receivers, mode — duplication or split — rate, unit of time, and the agreed definition.
What it detects
Orphan leaks. If one mechanism reports 90 leaving and the next reports 61 arriving, twenty-nine units vanish between them — and neither sheet is wrong on its own.
Filter it hands to iterations
« This iteration changes what the neighbour receives: warn them, and measure on the slower of the two clocks. »
The point to remember
It belongs to neither mechanism. Described inside each, it is entered twice and the two copies diverge. In the register it exists exactly once.
Resources
what several mechanisms compete for without passing it between them
What it holds
Name, nature among the five — throughput, threshold, depleting stock, stock threshold, presence — unit, capacity, load, alert threshold, and who draws on it.
What it detects
Contention. Two mechanisms slow down together with no visible internal cause, and nobody connects them because there is no arrow to follow.
Filter it hands to iterations
« Do not iterate to raise volume here: the resource is saturated and the gain will be absorbed by the queue. »
The point to remember
Cross-cutting stocks live here too — cash, the context base, audience, reputation. Position decides where an object lives, not its dynamics.
Events
what breaks comparability
What it holds
Date, the event, what it breaks, and the mechanisms affected. Four columns, no modelling.
What it detects
Nothing, and that is deliberate. It produces an invalidation rather than a friction: measurements before and after are no longer comparable.
Filter it hands to iterations
« Do not compare across this date. »
The point to remember
Anything that returns every year — seasonality, holidays — is not an event. It is an unmodelled input and belongs in organ 06.
Owners
who holds authority over which lever
What it holds
Person, role, the levers under their authority, the mechanisms involved, and whether they can hold a refusal.
What it detects
The real human constraint. It is usually the attention of two or three people, rather than budget or time, that bounds the whole.
Filter it hands to iterations
« This iteration has no available owner, whatever its priority. »
The point to remember
Authority cuts across mechanisms. Duplicated inside each sheet, the concentration on one person stays invisible.
What is not a register
three cases that look like one
Case
Where it belongs
Nuance
A stock inside one mechanism
It lives between two flows of that mechanism. Organ 08, not a register.
unless it is shared — cash, context base, audience — in which case position wins over dynamics and it moves to resources
An invalidated assumption
It concerns one mechanism's structure. Organ 20 holds it.
it is the only asset a failed test produces, but it is not transverse
A metric definition
« Lead shown » means what exactly? It belongs to the sensor that produces it, organ 15.
arguably the weakest call in this classification — two teams comparing numbers on different definitions never reconcile
The test is position, not dynamics. A stock and a resource behave differently — one accumulates and circulates, the other bounds without circulating — but that is not what decides where they live. What decides is whether more than one mechanism touches it. A shared stock goes to the resources register and keeps its stock nature there, which is why that register carries the five natures rather than a single one.
Filling order
create all four, fill them at different times
Create all four before you model. Not to fill them, but so shared objects have somewhere to land instead of being duplicated inside each sheet. A register that does not exist when you draw your first mechanism will never catch up.
Events start today. A date cannot be reconstructed afterwards, and a missing event ruins a conclusion months later — you will not know whether a mechanism was wrong or a test was polluted.
Owners start today too. Authority is a fact about your organisation that you already know without modelling anything.
Resources fill from the second mechanism. Populate them before and you list what you believe is scarce rather than what your mechanisms actually contend for.
Interfaces fill from the third. You cannot write one before two mechanisms touch — it is the only register that is strictly derived.
Two of the four blockers live here. A saturated resource and a busy owner are the reasons not to iterate that no single mechanism can see. Which also means these two registers only start paying once three or four mechanisms are modelled — below that, the slow stock and the blind sensor carry the load alone.
Framework 04 · Knowledge System · IP
Organs
The 21 organs on one map. Matter flows left to right, control comes back right to left. Each card gives the function, the control question, and what breaks when it is missing.
21 organs5 functions10 proposals settled
Structural model
21 organs on one map
Matter flows left to right, control comes back right to left.
function, control question, what breaks when it is missing
Frame · 5 organs
01
Boundary
Function
Cuts what belongs to the mechanism from what stays outside. Everything else in the canvas depends on that cut.
Control question
What is inside, what is outside, and exactly where does the line run?
If the organ is missing
Effects it does not produce get attributed to it, and the objective becomes unreachable by construction.
Filled in when
You can name a borderline element and say which side it falls on, without hesitating.
02
Declared objective
Function
Sets the measurable target that serves as reference for the gap comparator.
Control question
What value, on what metric, by what date?
If the organ is missing
No gap is computable, so no steering is possible. Only opinion remains.
Filled in when
A number, a unit, a deadline. Not a direction.
03
Revealed objective
Function
Exposes what the mechanism actually rewards, regardless of what it announces.
Control question
Looking at last month's decisions, what do they maximise?
If the organ is missing
You correct levers while the structure pulls elsewhere. Gains dissolve without explanation.
Filled in when
You can quote a recent decision that contradicts the declared objective, or attest that there is none.
04
Constraints
Function
Bounds the space of admissible solutions: budget, legal, capacity, ethics.
Control question
What limits can you not cross, even by paying?
If the organ is missing
You simulate impossible scenarios and plan on gains that will never be executable.
Filled in when
Each constraint has a limit value and a reason.
05
Owner
Function
Assigns decision, execution and exposure to the result to identified people.
Control question
Who decides, who executes, who lives with the result?
If the organ is missing
A policy with no owner never applies. The mechanism stays theoretical.
Filled in when
One name per role, not a team.
Matter · 5 organs
06
Inputs
Function
Feeds the mechanism with resources, signals or units to process.
Control question
What, how much, at what cadence, with what variance?
If the organ is missing
You blame the process for a drop that comes from upstream, and tune an engine starved of fuel.
Filled in when
An average rate and a spread, not a running total.
07
Disturbances
Function
Represents the shocks the mechanism takes without choosing them: seasonality, platform, market.
Control question
Which shocks, how often, of what amplitude?
If the organ is missing
The model looks more stable than it is. Real variance surprises you every cycle.
Filled in when
Each disturbance has an estimated frequency and a quantified effect on a flow or a stock.
08
Stocks
Function
Accumulates between two flows. The mechanism's memory, and the thing that reveals bottlenecks.
Control question
What piles up, at what level, for how long?
If the organ is missing
The bottleneck goes invisible: conversion rates stay fine while the queue grows.
Filled in when
A level and an average age, recorded on a date.
09
Flows
Function
Carries matter from one step to the next, at a given rate and pass-through ratio.
Control question
What rate between each step, and what pass-through ratio?
If the organ is missing
You reason on global volume without knowing where it jams. Corrections become random.
Filled in when
Every edge on the canvas carries a rate and a ratio.
10
Outputs
Function
Produces the measurable result to be compared with the objective, and which sometimes feeds another mechanism.
Control question
What result, measured how, going to whom?
If the organ is missing
The mechanism runs with no proof it serves anything. Activity gets mistaken for production.
Filled in when
A metric, a unit, a destination.
Machinery · 4 organs
11
Levers
Function
Variables genuinely actionable over the chosen horizon, with their range and their cost.
Control question
What can you move over this horizon, within what range, at what cost?
If the organ is missing
The diagnosis leads to no action. You know what is wrong and can do nothing with it.
Filled in when
Each lever has a current value, a range and a cost of movement.
12
Frozen parameters
Function
Fixes what will not move over the horizon, and therefore enters the simulation as a constant.
Control question
What is frozen for this horizon, and for what reason?
If the organ is missing
You plan on imaginary levers. The plan is unworkable from day one.
Filled in when
Each frozen parameter carries a reason and a review date.
13
Transformation rules
Function
Links each step's input and output through an explicit function.
Control question
What function links input and output at each step, with what coefficients?
If the organ is missing
The simulation has no engine. You stack ratios without knowing why they take those values.
Filled in when
You can compute a step's output from its input without looking at history.
14
Delays
Function
Shifts the effect relative to the cause. This is what sets the maximum steering cadence.
Control question
How long between the action and its visible effect on the output?
If the organ is missing
You judge too early, stack corrections, and the mechanism starts oscillating instead of converging.
Filled in when
One delay per step, in time units, measured rather than assumed.
Control · 4 organs
15
Sensors
Function
Makes a quantity observable, with an instrument, a latency and a noise level.
Control question
What quantity, measured by what instrument, with what latency?
If the organ is missing
The mechanism is endured, not steered. You react to visible consequences, always late.
Filled in when
Every stock, output and leak has at least one named sensor.
16
Gap
Function
Compares measured against target. The only signal allowed to trigger an action.
Control question
What value, over what window, with what trend?
If the organ is missing
You react to anecdote instead of signal, and every wobble becomes grounds for intervention.
Filled in when
A signed number, together with its observation window.
17
Policy
Function
Converts a gap into action, following a rule decided in cold blood.
Control question
What rule triggers what action, from what threshold?
If the organ is missing
Every gap triggers a different improvisation. No learning is possible between cycles.
Filled in when
A default action is written for every signal on the control table.
18
Loops
Function
Closed paths in the mechanism. A B loop pulls back to target, an R loop amplifies the gap.
Control question
Which closed loops exist, and which are balancing, which reinforcing?
If the organ is missing
You treat as linear a mechanism that feeds itself. Compound effects arrive unanticipated.
Filled in when
Every loop is drawn on the canvas and labelled B or R.
Epistemic · 3 organs
19
Evidence level
Function
Qualifies how solid each value in the model is, and flags where the risk concentrates.
Control question
Is this value measured, estimated or assumed, and on what sample?
If the organ is missing
An assumption acquires the status of fact. The model looks solid while resting on hunches.
Filled in when
Every simulation variable carries an evidence tag.
20
Structural assumptions
Function
Lists the causal links accepted without testing, and the observation that would invalidate them.
Control question
Which causal links do you take for granted without having tested them?
If the organ is missing
When the model fails, you do not know what to fix: the structure or the parameters.
Filled in when
Each assumption has a decisive test written next to it.
21
Excluded scope
Function
Explicitly records what was left out of the model, and why.
Control question
What did you deliberately leave out, and under what condition should it come back?
If the organ is missing
Surprises arrive with no known origin. Every anomaly restarts an investigation from scratch.
Filled in when
The list exists and gets re-read before any counter-intuitive conclusion.
The ten proposed organs, settled
eight absorbed, two still open
Proposed
What it was
Verdict
Where it went
Horizon and cadence
A parameter is only « frozen » relative to a duration.
absorbed
Organ 17 — cadence is already one of the policy's three settings, and the horizon is the objective's.
Binary success criterion
The threshold that turns an objective into a test.
absorbed
Organ 02 — the acceptable minimum field does exactly this.
Resources consumed
Budget, time and attention burnt each cycle.
absorbed
The resources register — it is transverse, so it never belonged inside a mechanism.
Leaks and scrap
Drop-offs, churn, dead leads, content never shipped.
absorbed
Organ 10 — induced outputs, and the conservation gate makes them impossible to skip.
Capacity and bottleneck
The maximum rate of each step.
absorbed
Organ 04 — the throughput nature, with its degradation threshold.
Non-linearities
Thresholds, saturations, diminishing returns.
absorbed
Organ 13 — the non-linear flag, and a gate that fails when none is declared.
Actuator
What actually executes the decision, with its own reliability and delay. A policy with no actuator changes nothing.
open
Genuinely missing. The owner says who may decide; nothing says what carries it out, or how long that takes.
Measurement latency
Data age, noise, sensor bias.
absorbed
Organ 15 — all three are now mandatory fields of a sensor.
Known failure modes
The ways this mechanism breaks, with the leading signal for each.
open
Nothing holds it. It would make monitoring predictive rather than reactive.
Decisive test
What observation would invalidate a structural assumption.
absorbed
Organ 20 — the test field next to each assumption.
The derivation settled eight of the ten. Once the five functions were named, most of these turned out to belong to something that already existed rather than new objects. Two survive: the actuator, and the list of known failure modes. Both are real gaps, and neither has a home yet.
The map of the mechanisms that keep a company alive, by domain: marketing, sales, delivery, product, finance and people. Each card shows whether the mechanism is modelled, how many of its five functions are served, and whether a cycle is running. Arrows show what feeds what, within a domain and between domains, and the two reinforcing loops that make a company compound.
2How it makes everyone more efficient
One picture of the company. Everyone sees which mechanisms exist, which are modelled and which feed which.
Model in the right order. Upstream mechanisms come first, so each model has real inputs to read.
Cycles only where they can work. A mechanism is ready for a diagnose cycle only once its five functions are served.
3How to use it
Click a mechanism to see what it is, its domain, what feeds it and its state.
Not modelled: “Model this mechanism” opens it on the Organs tab.
Modelled: “Open the model” opens it on the Model tab. Pre-call and Sales open their own pages.
Ready: “Start a diagnose cycle” opens the Control tab.
No notifications from 20:00 to 08:00, local time of each person. Critical alerts still go through.
Daily digest
Low-severity alerts are grouped into one message at 08:30 instead of being sent one by one.
Escalation
A critical or high alert not acknowledged within the delay below is sent to Romain.
{{esc.label}}
Example notification
SKL OS · Task overdue · HighShooting brief — masterclass #3 is 4 h past its committed date (Thu 15).To: Ilya · Open in My Day
Alerts
Click an alert to set its threshold, severity, recipients and channels
{{grp.label}}{{grp.count}}
Alert · {{ed.group}}
{{ed.name}}
{{ed.stateLabel}}
{{ed.stateHint}}
Condition
{{ed.before}}{{ed.valueLabel}}{{ed.after}}
{{ed.defaultHint}}
Severity
{{ed.sevHint}}
Notify
Channels
When it fires{{ed.summary}}
Page method
Alerts & notifications
1What this page is for
This is where the OS alert system is set. Each alert watches one condition, such as a task going overdue, a key result turning red or a decision past its deadline. For each alert you choose the threshold, the severity, who gets notified and through which channel. Delivery rules decide when notifications arrive and what happens when nobody reacts.
2How it makes everyone more efficient
The system pushes, nobody has to check. A slip reaches the right person as it happens, not at the next meeting.
The right person, not everyone. Each alert goes to the owner first. Romain only gets what needs him, or what nobody handled in time.
No noise. Severity, quiet hours and the daily digest keep low-value alerts from interrupting deep work.
Rules that match how you work. Thresholds follow the OS rules (8 h tasks, 3 tasks in progress, 48 h decisions) and can be tuned as the team matures.
3How to use it
Switch an alert on or off with the toggle on its line.
Click the alert's name to open its settings.
In the popup, adjust the threshold with − and +, for example how many hours after the committed date a task counts as overdue.
Pick the severity (Low alerts go into the daily digest), who is notified and the channels. The dark box shows exactly what will be sent.
Set the delivery rules: quiet hours, daily digest and the escalation delay.
Save. Review the settings at each quarterly planning: an alert that nobody acts on should be tuned or switched off.
Registers hold what several mechanisms share: the interfaces where one mechanism hands units to the next, the resources they compete for, the events that break comparisons, and the people who own the levers. A mechanism sheet describes one mechanism; a register describes what sits between them.
2How it makes everyone more efficient
No orphan leaks. If Pre-call sends 61 shows and Sales counts 90 bookings, the Interfaces register shows the gap nobody owns.
Hidden contention becomes visible. Two mechanisms that slow down together often share a saturated resource with no arrow between them.
Honest comparisons. The Events register tells you which before/after comparison is no longer valid.
The real human bottleneck. The Owners register shows how many levers depend on the same two or three people.
3How to use it
Pick a register with the four tabs at the top.
Read the four boxes: what the register holds, what it detects, how it filters iterations and when to fill it in.
Settle the highlighted lines first: they are open questions written as TO ARBITRATE, CONFIRM or FILL IN.
Watch the Resources load bars: a resource past its alert threshold blocks any iteration that adds volume.
Log an event the day it happens: it takes thirty seconds and saves a wrong conclusion.
Click a mechanism name to open its model in the Knowledge System.
Page method
Pre-call mechanism
1What this page is for
This page models the Pre-call mechanism with the Iteration system: 21 organs spread across 5 functions (frame, matter, machinery, control, epistemic). From that model the page computes the stocks and flows, checks 5 gates, and lists the frictions that should become decisions.
2How it makes everyone more efficient
You act on causes, not symptoms. The diagnosis points to the organ that limits the result, so effort goes to the one lever that matters.
No iteration on a broken model. The gates stop you from testing a lever while the model leaks units or corrects faster than the system can answer.
Facts are separated from hunches. Every value carries its evidence level, so nobody defends a guess with the confidence of a measurement.
One shared picture. Anyone can read how the mechanism works, who owns which lever and what is deliberately left out.
3How to use it
Build: follow the steps to construct the mechanism, one organ at a time.
Model: read the graph left to right. Click a node to edit it; the bar on top holds what frames the mechanism.
Organs: the 21 organs by function, each with its explanation and filling rules.
Control: the gates, the ignorance map and the notifications. Fix what blocks first, then open a decision.
After each test, update the organs it changed: the model improves with every cycle.
Page method
Sales mechanism
1What this page is for
This page models the Sales mechanism with the Iteration system: 21 organs spread across 5 functions (frame, matter, machinery, control, epistemic). From that model the page computes the stocks and flows, checks 5 gates, and lists the frictions that should become decisions.
2How it makes everyone more efficient
You act on causes, not symptoms. The diagnosis points to the organ that limits the result, so effort goes to the one lever that matters.
No iteration on a broken model. The gates stop you from testing a lever while the model leaks units or corrects faster than the system can answer.
Facts are separated from hunches. Every value carries its evidence level, so nobody defends a guess with the confidence of a measurement.
One shared picture. Anyone can read how the mechanism works, who owns which lever and what is deliberately left out.
3How to use it
Build: follow the steps to construct the mechanism, one organ at a time.
Model: read the graph left to right. Click a node to edit it; the bar on top holds what frames the mechanism.
Organs: the 21 organs by function, each with its explanation and filling rules.
Control: the gates, the ignorance map and the notifications. Fix what blocks first, then open a decision.
After each test, update the organs it changed: the model improves with every cycle.
Page method
Operating System mechanism
1What this page is for
This page models SKL OS as two mechanisms, one per clock, built with the same Iteration system as Pre-call. Execution steering (daily, owner Romain) turns task and decision alerts into closed alerts and decisions taken. Results steering (weekly, owner Amin) turns KR readings into plans kept or changed, and sends recovery actions back to execution. Switch between them at the top right. Each model uses 21 organs spread across 5 functions (frame, matter, machinery, control, epistemic). From that model the page computes the stocks and flows, checks 5 gates, and lists the frictions that should become decisions.
2How it makes everyone more efficient
You act on causes, not symptoms. The diagnosis points to the organ that limits the result, so effort goes to the one lever that matters.
No iteration on a broken model. The gates stop you from testing a lever while the model leaks units or corrects faster than the system can answer.
Facts are separated from hunches. Every value carries its evidence level, so nobody defends a guess with the confidence of a measurement.
One shared picture. Anyone can read how the mechanism works, who owns which lever and what is deliberately left out.
3How to use it
Build: follow the steps to construct the mechanism, one organ at a time.
Model: read the graph left to right. Click a node to edit it; the bar on top holds what frames the mechanism.
Organs: the 21 organs by function, each with its explanation and filling rules.
Control: the gates, the ignorance map and the notifications. Fix what blocks first, then open a decision.
After each test, update the organs it changed: the model improves with every cycle.
Page method
SKL Map · Journey
1What this page is for
The map of all of SKL: the individual's journey from Point A (E0) to Ambassador (E13), the transition cells between states, and every need that moves the individual forward, in production, conversion, support and at the meta level, with their means, memories and tools.
2The principles of the map
One model, several views. Journey, Events, Build order, Simulator and Model read the same data. Nothing is entered twice.
The need comes first. Every card is a need. AI agents, humans and systems are means attached to the need.
Gaps are computed. Coverage, complete cells, contracts and agent sheets are computed by rules, never entered by hand.
Base model plus overlay. Model v5 is fixed; every change the team makes is stored on top of it and can be undone or restored.
3How to use it
Pick the reading at the top left: Needs, Means or Fully AI target.
Drag to move; Ctrl or ⌘ + wheel to zoom; 0 to fit all. The minimap at the bottom right is a shortcut.
Click a card: everything not linked to it fades and its detail opens on the right, where everything can be edited.
Filters (F key): lanes, links, coverage, add a need or a tool, jump straight to an act or a lane.
Search (/ key); Undo at the bottom left or Ctrl/⌘ + Z.
Page method
SKL Map · Events
1What this page is for
Every handoff read as an event, Event Storming style: sending need, event, data, receiving need and its means, with the hot spots that still block.
2The principles of the map
One model, several views. Journey, Events, Build order, Simulator and Model read the same data. Nothing is entered twice.
The need comes first. Every card is a need. AI agents, humans and systems are means attached to the need.
Gaps are computed. Coverage, complete cells, contracts and agent sheets are computed by rules, never entered by hand.
Base model plus overlay. Model v5 is fixed; every change the team makes is stored on top of it and can be undone or restored.
3How to use it
Read the groups by act, chain, meta or support.
Start with the red labels: they are the hot spots of the receiving need.
Click a label to open it on the Journey.
Page method
SKL Map · Build order
1What this page is for
The build order computed from the model's dependencies: every tool, memory, orchestrator or need that is not finished becomes a build item, sorted into waves.
2The principles of the map
One model, several views. Journey, Events, Build order, Simulator and Model read the same data. Nothing is entered twice.
The need comes first. Every card is a need. AI agents, humans and systems are means attached to the need.
Gaps are computed. Coverage, complete cells, contracts and agent sheets are computed by rules, never entered by hand.
Base model plus overlay. Model v5 is fixed; every change the team makes is stored on top of it and can be undone or restored.
3How to use it
Start with “What unblocks the most”.
A wave starts only once its prerequisites exist.
In each wave, items tied to objectives come first, then those that unlock the most.
Click an item to open it on the Journey.
Page method
SKL Map · Simulator
1What this page is for
The journey as stocks and flows: exposed people per week, the rate of each cell, recovery of no-shows and undecided prospects, the referral loop. It computes clients per month and net revenue.
2The principles of the map
One model, several views. Journey, Events, Build order, Simulator and Model read the same data. Nothing is entered twice.
The need comes first. Every card is a need. AI agents, humans and systems are means attached to the need.
Gaps are computed. Coverage, complete cells, contracts and agent sheets are computed by rules, never entered by hand.
Base model plus overlay. Model v5 is fixed; every change the team makes is stored on top of it and can be undone or restored.
3How to use it
Each value is marked Measured, Scenario or Assumption.
Change a rate to build a scenario; it stays in this browser and does not touch the measures.
The levers show the revenue gain if a cell's rate rises by 5 points.
Page method
SKL Map · Model
1What this page is for
All the map's data in tables: states, cells, needs, means, agent sheets, handoffs, memories, tools and deleted items.
2The principles of the map
One model, several views. Journey, Events, Build order, Simulator and Model read the same data. Nothing is entered twice.
The need comes first. Every card is a need. AI agents, humans and systems are means attached to the need.
Gaps are computed. Coverage, complete cells, contracts and agent sheets are computed by rules, never entered by hand.
Base model plus overlay. Model v5 is fixed; every change the team makes is stored on top of it and can be undone or restored.
3How to use it
Click a row to open the item on the Journey.
“Reset the whole map” clears every change the team made and returns to model v5.
Page method
Agent System · Overview
1What this page is for
The agent cockpit at a glance: what your agents do, what waits for your decision and what is drifting.
2How it makes everyone more efficient
Agents propose, humans decide. Every output an agent is not yet trusted to send alone waits here for a person, and every decision teaches the agent.
Autonomy is earned. An agent moves up a level only when its figured criterion is met: tests passed and outputs approved without edits.
One owner per agent. Every agent has a person who answers for it and a person who receives what it must not handle.
Everything is traced. Every run keeps its trigger, result, duration and cost.
3How to use it
Read the five indicators; the dark one is what waits for you.
Check the agent health table; click an agent to open its sheet.
Handle the items in “To handle”: a click opens them in Decisions.
Read the alerts: an agent below the quality threshold or in error.
Page method
Agent System · Decisions
1What this page is for
The human control of the agents: approve, edit or reject what they propose, and handle the escalations.
2How it makes everyone more efficient
Agents propose, humans decide. Every output an agent is not yet trusted to send alone waits here for a person, and every decision teaches the agent.
Autonomy is earned. An agent moves up a level only when its figured criterion is met: tests passed and outputs approved without edits.
One owner per agent. Every agent has a person who answers for it and a person who receives what it must not handle.
Everything is traced. Every run keeps its trigger, result, duration and cost.
3How to use it
Pick an item in the list on the left.
Read why it needs you, the agent's proposal, its destination and the sources it read.
Approve to send it, or reject it: the next pending item opens on its own.
Decided items stay in the list, faded, with the decision recorded.
Page method
Agent System · Agents
1What this page is for
The registry of SKL's AI agents: what each one does, who answers for it and at what level of autonomy it works.
2How it makes everyone more efficient
Agents propose, humans decide. Every output an agent is not yet trusted to send alone waits here for a person, and every decision teaches the agent.
Autonomy is earned. An agent moves up a level only when its figured criterion is met: tests passed and outputs approved without edits.
One owner per agent. Every agent has a person who answers for it and a person who receives what it must not handle.
Everything is traced. Every run keeps its trigger, result, duration and cost.
3How to use it
Click a card to open the agent sheet: how it works, its autonomy ladder, allowed actions and guardrails, performance and accountability.
From the sheet, pause or reactivate the agent, or rerun its tests.
“New agent” starts the creation of an agent.
Page method
Agent System · Triggers
1What this page is for
The orchestration: which event wakes which agent, what it produces, who decides what happens to it, and the connectors it uses.
2How it makes everyone more efficient
Agents propose, humans decide. Every output an agent is not yet trusted to send alone waits here for a person, and every decision teaches the agent.
Autonomy is earned. An agent moves up a level only when its figured criterion is met: tests passed and outputs approved without edits.
One owner per agent. Every agent has a person who answers for it and a person who receives what it must not handle.
Everything is traced. Every run keeps its trigger, result, duration and cost.
3How to use it
Read each row from left to right: when, agent, output, decision.
Click an agent's name to open its sheet.
Page method
Agent System · Log
1What this page is for
The traceability of the agents: every run with its trigger, result, duration and cost, most recent first.
2How it makes everyone more efficient
Agents propose, humans decide. Every output an agent is not yet trusted to send alone waits here for a person, and every decision teaches the agent.
Autonomy is earned. An agent moves up a level only when its figured criterion is met: tests passed and outputs approved without edits.
One owner per agent. Every agent has a person who answers for it and a person who receives what it must not handle.
Everything is traced. Every run keeps its trigger, result, duration and cost.
3How to use it
Scan the Result column first: “Error” and “Escalation” need a look.
Compare durations and costs between runs of the same agent.