Equity research infrastructure

The model and the note are the same thing.

Live financial models of managed-care companies, and the research written from them — built so that every figure in a note can be traced back to the assumption that produced it, and to the filing that produced the assumption.

How it works

2 companies modelled end to end, each from its own filings

68 research surfaces, from the three statements to state-level Medicaid

10yr of filed segment history, per segment and sub-segment

3 note formats across the earnings cycle — preview, print, call

Three things, joined up

Most research is a model in one place and a document in another, reconciled by hand and by memory. These are one system, which is what makes the rest of it possible.

A model you can interrogate

Full three-statement forecast driven by segment-level assumptions, on the company’s own reporting calendar — annual, quarterly, or a mix. Change one driver and every statement, scenario and valuation moves with it, in your own copy, visible to nobody else until it is published.

Research built from the model

A preview before the print, a review of the result against sealed consensus, and a note on what the call added — written from the model’s own numbers rather than retyped from it, then scheduled and sent to a named list.

A record of what changed

Every published version keeps what it replaced, field by field, with the reason it moved. A number from six months ago still explains itself, and so does the person who changed it.

What’s inside

The model is the spine. Most of the platform is the work that has to exist underneath a forecast before it means anything — the industry it sits in, the history it came from, and the evidence for each assumption.

The model

Three statements on a mixed annual and quarterly axis
Segment forecasts built from members, price and cost
Scenarios — base, bull, bear, and named cases of your own
Results vs consensus, sealed before the print
EPS power, uses of cash, credit and the debt schedule
Excel export that computes what the app computes

Valuation

Historical multiples against the company’s own record
Projected value from growth and multiple, side by side
Implied return — what today’s price has to believe
Point-in-time consensus, reconstructed from dated guidance
Portfolio backtest over many start dates, against a null
Rating and price target, each axis argued and dated

The company

Company history from inception, with what the market paid
Segment history, ten years, segment by segment
Risks sized in EPS on drivers that actually move
Target margins, each graded by what it is sourced to
Management, proxy and acquisitions from the filings
Catalyst calendar and quarterly commentary

The industry

Medicare Advantage — rates, stars, benefits, SNPs
Medicaid — expansion, state programmes, monthly enrolment
Commercial — employer, exchange, and the uninsured
Pharmacy and PBM economics
Medical cost trend and industry margins by line
Peer scorecard and every peer’s quarterly print

How a note actually gets out

The part that usually lives in somebody’s inbox and memory. Writing, reviewing, scheduling and sending are one path, and every step of it is refusable.

Built from the model

Pick the company and the format. The note is generated from the current model — its figures are the model’s figures, not a transcription of them.

Audited before it moves

The finished note is checked against the reported figures it was built from, and against its own text. Rehearsal wording, an unverified claim or a stale figure stops it.

Addressed deliberately

The company’s standing distribution list, or a set of people chosen for this note. The list is resolved when it sends, so an unsubscribe in between takes effect.

Scheduled, and cancellable

Queued for a time you choose and visible until it fires — what is waiting, when it goes, and exactly who receives it. Nothing sends without passing every check again.

What the machinery is actually for

Nearly all of it exists to make one kind of mistake impossible: a figure reaching a reader without anyone able to say where it came from.

Consensus is sealed first The benchmark is hashed before the company reports, so a beat is measured against what was expected beforehand — not against a number revised afterwards.
Nothing is inferred A figure that was not disclosed is recorded as absent and says so. A null with an explanation is the correct answer; an interpolation presented as a fact is not.
Sourcing is a field, not a footnote A margin target the company stated and one the analyst assumed are different objects, and the platform will not let them share a column without saying which is which.
A note is audited before it goes The finished note is checked against the reported figures it was built from. If those changed after it was written, it does not send.
Sending is off until switched on A container that is merely running cannot email an investment committee. Live sending is enabled by hand, after a dry run has been read.
Drafts are private until published Anyone can work on their own version of the model. Only the owner of that model can make one the official build, and doing so is recorded.
Access is granted one surface at a time There is deliberately no bundle to hand out. Each person’s reach is set surface by surface, and every change to it is written to the audit log.
Every change has an author Who changed what, when, and what it replaced — kept for good, and readable by everyone who reads the numbers.

Built for the second company, not just the first

A platform written around one company will quietly serve that company’s answers under every other name. Its reporting calendar becomes the calendar; its segments become the segments; its peer group, its research and its executives end up on another company’s page, and nothing errors.

So the time axis, the segment tree, the revenue build, the driver registry, the research and the risk register are all declared per company rather than assumed — and the tests assert that the second company behaves, not merely that the first still does. A surface no one has researched for a company is hidden rather than filled in with somebody else’s.

Adding a company is a sequence of declarations, each one checkable on its own. It is the least glamorous work in the system and the reason a number can be trusted once there is more than one company in it.

Access is by invitation

The platform is private. If you have an account, sign in with the address it was issued to.

That page is on the platform.