Method
What the Boundary Explorer compares, and what it does not.
A short read for anyone who wants to know what is being computed, what the boundary line means, and what is honestly outside the instrument's reach.
01 · The shape of the comparison
A baseline and one or more proposed alternatives.
The Boundary Explorer compares the current way of doing a recurring job with one or more proposed alternatives over a stated useful lifetime. The current way is treated asbaseline — its future setup is sunk (zero), since the work is already being done. Each alternative is a proposed way whose setup cost is treated as a future investment.
Each option is described with the same five fields:
- Future setup
- One-time minutes to bring the alternative up. Sunk for the baseline.
- Human minutes / run
- Recurring human effort each time the task runs.
- Failure probability
- Likelihood a run needs additional repair. Between 0 and 1.
- Repair minutes / failure
- Additional time spent on a failing run beyond that option's ordinary per-run handling. Each option's repair estimate is counted against its own per-run cost, not against the baseline's.
- Maintenance / month
- Ongoing upkeep minutes per month — prompt tweaks, model updates, drift handling.
Plus two fields shared across all options:frequency (how often the task runs, in runs per month) and useful lifetime (how long this comparison is meant to apply to, in months).
Each option carries a small fixed tag — its origin: observed, user estimate, illustrative, or unknown. The math does not change with origin; the origin is metadata for the reader, not a weight.
02 · A short worked example
Drafting a recurring AI task.
The default illustrative example on the home page and the tool. Replace any number with your own estimate; the comparison re-runs.
Baseline · current way
Manual today
- Future setup
- 0 min
- Human minutes / run
- 30
- Failure probability
- 0.1
- Repair min / failure
- 0
- Maintenance / month
- 0
Alternative · proposed way
Assisted reusable workflow
- Future setup
- 120 min
- Human minutes / run
- 6
- Failure probability
- 0.2
- Repair min / failure
- 18
- Maintenance / month
- 45
Frequency
15 runs / month
Useful lifetime
8 months
Preferred · payback
assisted · 0.5 months
Baseline cost over the horizon: 3600 minutes. Alternative cost: 1632 minutes. Boundary: at 2.94 runs/month, the preferred option flips.
03 · What a boundary means
The exact value at which the preferred option changes.
The boundary is not a confidence interval. It is the exact value of one variable at which, holding every other variable fixed, the comparison switches which option is cheaper over the horizon.
In the worked example above, the boundary is onfrequency. At about 2.94 runs per month, the manual way and the assisted workflow cost the same. Above that, the assisted workflow is cheaper; below it, the manual way is cheaper. Holding everything else fixed.
The boundary can also be on a single field of one option — for example, the human minutes per run of the assisted workflow. The engine searches the variable you select and reports the flip within the searched range. If no flip exists in that range, the instrument says so explicitly rather than guessing.
When the comparison contains more than two transitions in the searched range, the explorer shows each one as an adjacent interval — not a single averaged line.
04 · What a scenario is — and is not
A scenario is not a probability.
A scenario is a named override applied to the comparison: a different frequency, a different repair time, a different human-minutes-per-run. The engine re-runs the comparison under that scenario and reports whether the preferred option changes.
A scenario is not a probability. It does not combine with other scenarios. It does not weight by likelihood. Adding "the frequency doubles" and "the repair time halves" as two separate scenarios does not produce a "both at once" answer — those are two distinct worlds to inspect.
Every scenario carries a provenance string — where the override came from. A scenario with no provenance is not accepted; the explorer will refuse to add it.
05 · Quality and eligibility assumptions
Marking an option ineligible is a deliberate act.
Each option carries an eligibility flag and anineligibility reason. When you uncheck eligibility, the alternative is removed from the cost comparison and the explorer asks for a reason — a vendor that failed a governance review, a tool that does not meet the data-handling bar, anything that ought to take this option off the table.
Eligibility is therefore a different kind of decision from cost. The comparison does not average across eligible and ineligible options; it computes cost only among the eligible ones and reports the rest as out of scope.
The instrument also assumes that the quality of the alternative's output is acceptable to the operator. A workflow that returns the wrong answer faster is not a cost saving; it is a different problem. The Boundary Explorer does not test for quality.
06 · Honest limits
What the instrument does not establish.
- Whether the workflow will actually produce an acceptable output.
- Permission to delegate consequential judgment to a tool.
- A vendor or model recommendation.
- A guarantee of future savings, accuracy, or quality.
- An answer about a decision that is not a recurring time-cost trade-off.
- Where the preferred option flips under held-fixed assumptions.
- A defensible next step in the current state.
- A side-by-side reading of two named ways of doing a recurring job.
- A scenario, with provenance, that you can re-evaluate under changed assumptions.
- An honest record of the inputs you used to reach a given verdict.
A filled form does not establish acceptable quality, safe delegation, or that a workflow is worth building. The instrument shows where the answer flips under held-fixed assumptions; it does not propagate uncertainty through distributions.
It is one decision aid for one recurring question. It is not a general-purpose intelligence surface.