CodelessOps · Projects · Estate review
Run 42 tests on your Anaplan models. Get the five that matter first.
Upload your models' line items, actions and modules. The review runs the same 42 tests on every model, then picks the most impactful things you can do and sets them out as an action plan, with the evidence behind each one. Free, open source, no AI, and it runs locally when the exports can't leave.
01 · The tests
42 defined tests in six areas. The same on every model.
- 1
Performance and size 8 tests
Where calculation time and cells go, and the formula patterns Anaplan documents as expensive.
- 2
Usage and leftovers 5 tests
What nothing appears to read: candidates to ask about, never verdicts.
- 3
Dependencies and change impact 5 tests
What depends on what, inside a model and between models.
- 4
Correctness and maintainability 17 tests
Totals that may be wrong, repeated logic, and formulas the next builder has to decode.
- 5
Imports, exports and processes 4 tests
What loads the model, what has stopped running, what sits outside every process.
- 6
How far to trust the analysis 3 tests
Tests the report runs on itself, so every finding says how strong its evidence is.
The full list says what each test looks for and which export it needs. The report shows the result of every one on your exports: found, clear, or not run and why. A test that could not run is never reported as clear. A hit is a candidate to review, not a verdict.
02 · The action plan
From everything found to the five things to do first.
- Every test that found something offers a candidate action.
- Candidates are ranked: evidence first, a named set of objects before an open-ended review, then how much is at stake.
- What is at stake is measured where it can be (share of calculation effort, cells) and fixed by the test where cells say nothing: a total that may be wrong ranks high however small the line item; upkeep ranks low.
- A concrete change comes before an investigation. The first five that meet the bar are the plan; everything else is listed under it with the reason.
On the built-in example (four models, 426 line items) 35 of the 42 tests find something. That becomes 45 candidate actions, 10 meet the bar, and the plan opens on five: a lookup repeated on every cell, a percentage whose totals are added up, a twelve-branch IF chain, one line item that breaks its neighbours' pattern, and a module nothing reads. The order is deterministic and explained in the report; it is a hypothesis about what is worth doing first, not a verdict.
03 · What the report contains
Three views of the same exports.
Action plan
The five most impactful things to do, each with the line items involved, why, the steps, a completion check and the test that found it. The other candidates follow, with the reason they are not on the plan.
Change impact
Pick a module or line item and follow its dependencies either way, with the paths and a clickable diagram. Download the change review as JSON.
Evidence
Every test with its result (found, clear, or not run and why), then the searchable findings, formulas, model inventory and dependency coverage. Add statuses and notes, export a register as CSV.
In the example, one formula, Forecast Opex, reaches 27 line items across 7 modules, eight links deep, and ends in the board pack EBITDA line. The explorer shows that reach before anyone changes it.
04 · Run it on your own models
Three exports from the model. One is enough to start.
- Line Items exportrequired
- Modules exportoptional, adds size context
- Actions exportoptional, adds import and process checks
Hosted
Upload at anaplan-estate.codelessops.com. Nothing is stored after the report is built.
Local
Clone klameer/anaplan-estate and run the same analysis on your machine. The formula parser is anaplan-grammar, also MIT.
This estate review is an independent creation, not affiliated with, endorsed by or supported by Anaplan, Inc.; Anaplan is their trademark. If you download or use it, you do so at your own risk.
05 · Public record
What changed, and what it found.
Walkthroughs, releases, posts and worked examples, each dated. Newest first.
- LinkedInMy estate review tool now runs 42 tests on your Anaplan models ↗A 36 second walkthrough of the upload, the 42 tests, the five-action plan and the change impact view
- Releaseanaplan-estate v0.2.0, 42 numbered tests and a plan of the five most impactful actions ↗Every test listed with its result; seven new tests; checked on eleven real exports before release
- PostAnaplan Estate: what it finds and how to use it
- PageAnaplan Estate gets its own page on this siteWhat the report contains, the checks, how to run it hosted or local
- Releaseanaplan-estate v0.1.0 public, hosted at anaplan-estate.codelessops.com ↗Action plan with every candidate shown, change impact explorer, evidence tab. MIT.
- Releaseanaplan-grammar v0.1.0 public ↗13,214 of 13,214 reference formulas parsed; dependency graph 0.998 precision, 0.994 recall vs Referenced By
06 · Common questions
Asked by the first architects who ran it.
Does my model data leave my machine?
The hosted tool does not store uploads. If exports can't leave your network at all, run the same analysis locally from the repository; nothing is sent anywhere.
How does it read formulas?
A parser for Anaplan's formula language, published separately as anaplan-grammar. On the reference corpus it parsed 13,214 of 13,214 formulas and matched Anaplan's own Referenced By on dependency edges at 0.998 precision and 0.994 recall.
What does a finding mean?
That a defined test matched something in the exports. It is an observation or a candidate to review, never a verdict: every finding carries its evidence strength, what the exports cannot tell, and the reasons the current design may be right.
Why five actions?
Because a plan you can act on this month is more useful than a list of everything. Every other candidate is still there, ranked, with the reason it is not on the plan; run the tool locally with --top to change the number.
I was expecting a test but it is not on the list. What can I do?
Tell me what you would check on a model by hand and what a hit would look like, and we can build it together. Message me on LinkedIn or open an issue on GitHub; the tests are numbered, so a new one takes the next number in its area. If you can run Python, the rules live in one file (lint.py) and a pull request is welcome.
Is this a replacement for Referenced By?
No. Referenced By answers one hop at a time. This follows the whole chain, groups what it finds into candidate actions, and shows the evidence. The architect still decides.
Does it use AI?
No. Every finding is a defined rule over the exports you supply. The same inputs give the same report every time.
Is it free?
Yes. MIT licensed, hosted or local. I built it to be useful to model builders and I want to know what it misses.
07 · Tell me what it missed
Run it on a model you know well.
The most useful thing you can send me is a finding it got wrong, a test you want that is not on the list, or a recommendation you would reject and why. That is how the tests get better.
Part of my projects: tools built in the open around finance systems and Anaplan.
