Audit FX-4471 · 12 September 2026
Halcyon Mobility, Series A Operating Model
halcyon_mobility_model_v7_FINAL.xlsx · 11 sheets · 6,284 populated cells
Integrity score
Not ready to send
3 severe issues to clear
Severity
Where the 39 points went
Two thirds of the deduction sits in internal consistency, meaning statements that do not agree with each other.
Do not send this model yet
Three severe defects break the model's internal consistency. The balance sheet stops balancing in Month 37 and the drift compounds to $1.24M by Month 60, so anyone who opens this and checks the tie-out will find it in under a minute. The underlying structure is sound and the three fixes are contained, roughly 70 minutes of work in total. Fix those, re-run the audit, and this model scores 82.
All three auditors independently reached the same conclusion on every severe finding.
Claude Opus 5
Anthropic
GPT‑6 Astra
OpenAI
Gemini 3.1 Pro
Google DeepMind
Agreement
How much the three models agreed.
This is the part a single AI tool cannot give you. When three models from three different labs independently flag the same defect, it is almost certainly real. When only one raises it, treat it as a judgement call worth your attention rather than a confirmed error.
Four problems were confirmed by every model i, including all three severe ones. Those four are not opinions. Fix them first.
Where they disagreed
Churn applied to the gross subscriber base
Claude and GPT‑6 both called this a modelling error that understates Years 4 and 5. Gemini read the same formula as an intentional conservative assumption and did not flag it. The two that flagged it gave the stronger argument, but this one is your call.
Sign convention flips between sheets
Only Claude raised this. It does not change any number today, but it makes the model harder for a third party to read, and it is the kind of thing that breaks later when someone extends a row.
Findings
Every issue, with the cell, the cause, and the formula that fixes it.
Ordered by severity, then by how many auditors agreed. Open any finding for the full reasoning and the replacement formula.
Remediation
Fix these in this order.
Sequenced so that each step unblocks the next. Times are for someone who knows the model.
Clearing the three severe findings takes about 70 minutes and moves this model out of the danger zone. The six moderate findings are another three hours and would take it to 97.
What holds up
These parts are already right.
An audit that only lists problems is not telling you the whole picture. These held up under all three reviews, so leave them alone.
By sheet
Where the problems live.
Two of eleven sheets came back clean. The damage is concentrated in the three statements.
Next step
Want a human to look at it?
Three AI auditors will find what is mechanically wrong. They will not tell you whether your Series A story holds together, or how this model will read to whoever is on the other side of the table.
Al Anany
Founder, Albusi GmbH, Zurich
15+ years in finance and strategy · 400+ models built · YC and 500 Global companies
I review these personally. Not a template, not a checklist. You get a recorded walkthrough of your model where I talk through what I would change and why, plus the reasoning an investor will apply when they open it.
Re-audit
$19$9
Fixed the three severe findings? Run all three auditors again and watch the score move. Half price for 30 days after your audit.
Human review
$500
A recorded video walkthrough of your model from me, plus written notes. Turnaround 3 business days.
Rebuild
$2,450 from
I rebuild the model from your assumptions. Fully unlocked Excel, ready to send. 5 to 7 days.