Where the dates come from, and how the value is built.

We search SEC EDGAR for 8-K and 6-K filings that mention phrases such as "PDUFA date", "topline" or "advisory committee". A language model reads each match for dated events, and a date is kept only if the passage it quotes appears word for word in the filing.

Most FDA decision (PDUFA) dates come from these filings, because the company is usually the one that discloses them. A new date publishes on its own only when two independent documents agree, a government source states it, or the company states it in its own SEC filing. Anything else waits for a person to review it.

Phase 2 and Phase 3 trials that are recruiting, active or completed, with the registry's primary completion date.

While that date is still expected, the row reads "Registry estimate", meaning a trial’s planned primary completion date on ClinicalTrials.gov: when data collection for the main result is expected to end, not when results are announced. Results usually follow within 0–3 months.

Once the registry records the date as actual, data collection has ended but the results are not out, and the trial is listed as completed, with results pending. A trial whose results are already posted on the registry is not listed.

Scheduled advisory committee meetings, read from the FDA's own calendar, with the day the FDA published.

Each event is rated low, medium or high by a fixed rule, not by a language model. The event type sets the starting level: an FDA decision, an advisory committee or a Phase 3 result starts high; a Phase 2 result or a response to a complete response letter starts medium; a Phase 1 result and other events start low.

The company's pipeline then moves it one level at most: up when the programme is the company's most advanced, or the company has only one or two programmes in the clinic; down when the company has three or more programmes at Phase 3 or later and this one is not the most advanced.

A language model with web search researches the company, starting with its SEC filings and press releases. Each statement it keeps is tied to the pages it came from, and only to pages the search actually retrieved; a statement it cannot tie to a source is left out.

A second pass turns that research into model inputs: one line per drug and indication, each with its phase, launch year, peak sales and probability of success, and the reasoning behind them. Cash, debt and share count are then taken from the company's SEC filings wherever the filings report them, not from the model. The result is saved as a research snapshot with its date.

Used for drug developers with no product on sale. By default, each line's revenue ramps up to peak sales over five years, holds for five, then falls 30% a year for five years. Revenue is multiplied by the company's share of the drug, an operating margin and tax, then weighted by the probability of success. A product already on sale counts in full.

Development spending before launch is not weighted by probability, because the trial is paid for however it reads out. The lines are discounted and added up, corporate costs come off, disclosed financing adds its cash and its new shares, and net cash is added. Dividing by the share count gives our estimated value per share.

The research sets the discount rate, and is instructed to keep it between 12% and 15% for a clinical-stage company. Bear, base and bull are separate sets of peak sales and probabilities, not a confidence interval.

For a company with an approved product and meaningful product revenue, a discounted cash flow model of its operating cash flows applies instead. A company whose stage the research cannot establish gets no headline value rather than a forced one. Comparable company multiples appear only as a cross-check.

Each catalyst is joined to the drug lines its outcome decides. An approval sets their probability of success to 100%. A failure sets it to zero, which removes their revenue. A delay moves their launch back one year.

Each lever edits the same model as the headline value, so a catalyst's result and our estimate always come from one set of numbers.

The risk-adjusted present value of the lines the event decides, as a share of the model's total equity value. It says how much of our estimate this one event decides.

At the probability you choose, the expected value per share is the failure value plus that probability times the gap between the approval and failure values.

The breakeven probability is the one at which that expected value equals the share price. When the price sits outside the range between the failure and approval values, there is no breakeven.

Cash and short-term investments from the latest filing, divided by the quarterly cash burn from operating cash flow, give the number of quarters the cash lasts. The catalyst card sets that against the event date.