- 01
The company's information
Under a confidentiality agreement, read with the public record.
Rules outA figure with no source.
- 02
Options generated
Every family of move crossed with every way of making it, each empty cell with its reason recorded. On the method's test company: 96 possibilities, 23 strategies beside the plan, 5 valued in full, 1 recommended.
Rules outA review that tests only the moves already on the board's agenda.
- 03
Each priced on one basis
On the return to the owners, with the assumptions beside each figure.
Rules outOptions measured in different ways, and a plan that escapes the test the alternatives face.
- 04
Combinations netted and bounded
Counted once, and held to the evidence.
Rules outHeadline figures added together that cannot all be true at once.
- 05
Automatic checks
On every page and every figure. The checks establish consistency and provenance; they do not establish that a recommendation will work.
Rules outA figure that appears twice with two values, or a claim with nothing behind it.
- 06
Two cold readings
By readers that have seen none of the work, followed by Julian Skeels's sign-off.
Rules outA page a board member could not follow on its own.
- 07
Two sign-offs
On the options taken forward, and on the document before it is released. Every review is directed and signed off by Julian Skeels, Evidmere's founder: thirty years in investment banking, and more than ten advising boards and investors on strategy.
Rules outAnything reaching a board that the director responsible for the review has not approved.
- 08
The decision document
About ten days after the information arrives.
Rules outA board waiting for the full review before it has a first answer.
What stands between the first answer and the full review
48 specialised components, which we group under four headings for this page: researchers 10, modellers 8, writers 5, checkers and readers 16, with 9 parts that run and control the work.
72 automatic checks in seven groups: specification 6, numbers 17, writing 23, rendering 11, reading 1 and release 13, and one group reported but not counted.
A count above zero refuses the build; nothing downstream reads it; every new attempt is read again by fresh readers.
On the method's test company: two fresh readers on each of eight attempts, with findings from 142 at the first attempt to 48 at the last.
- 09
The full review
By the time on the reviews page, with the workbook.
Rules outA figure the reader cannot recompute.
Nothing is released until every automatic check passes and both cold readings are complete. The checks establish consistency and provenance; they do not establish that a recommendation will work.