The science is already done
Trials, prices, patient counts and claims data sit in spreadsheets and source files.
Pramana (working name) by VedAGI, US workflow
For teams evaluating AMCP Section 4 signed figures, Pramana (working name) starts with the budget-impact table. It recalculates figures from source, labels estimates and locks figures until a named person signs.
A reviewer can verify the receipt in a browser. This is a production candidate, not a claim of broad production availability or AMCP endorsement. The figure-record version chain is being added and is not live today.
The name Pramana (working name) comes from Sanskrit, where it means proof: a means of valid knowledge.
Example numbers only. This checks drafted text going out through Pramana (working name). Checking slides and decks against the number comes later and is being built. Pramana (working name) blocks the pack and people fix it. It never quietly edits a number.
The idea
Teams already build the model and already use AI to write the pack. The number then hides inside the writing, a person signs it anyway, and the other side cannot check it.
Pramana (working name) works each number out again from the source and marks guesses as guesses. It refuses to send until a person signs, and it gives the other side a receipt they can open in a browser. Same pack. Same model. The leak is closed.
of US prescription claims in 2025 ran through just three PBMs (pharmacy benefit managers, the companies that run drug benefits for health plans)1
lower cost per QALY (quality-adjusted life year, a standard measure of health gained) in industry-funded studies than in independent ones2
average sales per day for a prescription drug, the sales each day of launch delay pushes back. Tufts Center for the Study of Drug Development, 2024, across 645 drugs launched since 20003
The path
A company has a drug, biologic or device and wants health plans to cover it. There is no single national reviewer, so the same numbers travel to many readers.
Trials, prices, patient counts and claims data sit in spreadsheets and source files.
Usually Excel or a specialist tool: a budget impact model and a cost-effectiveness model. It should be the one true source.
An AMCP (Academy of Managed Care Pharmacy) dossier, the standard document sent to health plans, 8, plus a pre-approval information exchange (PIE) deck, payer slides, a budget calculator and, for drugs picked for Medicare price talks, data for CMS (the Centers for Medicare & Medicaid Services).
Often by an AI assistant now. One team estimated AI cuts first-draft time by about 40% 4. Faster writing means more numbers to check.
MLR (medical, legal and regulatory) reviewers approve it and a named person signs. They trust the numbers match the model.
To PBMs, health plans and P&T (pharmacy and therapeutics) committees, often when they ask for it. Copies live on in slides and field decks.
P&T committees vote. PBMs negotiate. ICER (the Institute for Clinical and Economic Review, an independent group) may publish its own model. CMS (the Centers for Medicare & Medicaid Services) reviews data the company certified.
Canadian public coverage runs through a few national bodies. The US has many separate payers, and three PBMs processed 80% of prescription claims in 2025 1. One number, copied into many documents, is read by all of them.
The gap
The weak joint is not whether AI can write. It is that whoever writes each sentence can copy, round, or quietly change the number inside it. In the US that happens across many documents at once.
One number lives in five or more documents. A hand edit or an AI rewrite changes one copy and not the others.
An assumption about uptake reads like hard data because nothing marked it as an estimate.
MLR reviewers check claims and wording. They cannot re-run every number from source before signing.
When a P&T member, ICER or the FDA asks to see the working, teams dig through emails and old spreadsheets.
Economic information sent to health plans must rest on competent and reliable scientific evidence, and if the FDA (the US Food and Drug Administration) asks, the company must provide the primary data and analysis. A June 2026 revised draft brings devices in 5.
For Medicare drug price negotiation, the company certifies its data are complete and accurate, and must send corrections if they are not 6.
ICER builds its own models and shares them with drug makers under a transparency program, so your numbers get compared with theirs 7.
The industry already has models, writers and signers. It does not have a checkpoint that says: this number was worked out, labeled, signed, and can be checked by a stranger. That checkpoint is the gap. Pramana (working name) is only that checkpoint.
The checkpoint
Keep the model. Keep the writer. Keep MLR review. Pramana (working name) sits between “the words are drafted” and “the pack is signed and sent”, and it checks the numbers in the drafted text.
Each number is worked out again, in code, from the company's own sources. The AI may suggest a formula. It never writes the value.
If a number is an assumption, it is marked as one in the workbench (the on-screen work area), in the exported table, and in drafted text going out through Pramana (working name). There, it cannot pass as a measured fact.
Numbers in drafted text going out through Pramana (working name) must match a locked number, or the send is blocked. Tying each dossier row and, later, each slide to the number itself is being built.
What was worked out, from which sources, with which formula, signed by whom. Payers open it in a browser, with no VedAGI account. The same record can help answer an FDA or CMS request.
One number
Follow a single number, the year-one cost per member per month for a commercial health plan, from the company's files to a P&T reviewer's screen.
Pramana (working name) reads the company's own files: trial data, prices, patient counts based on claims, and the model. Nothing new is invented.
The AI can propose how the number should be worked out. Pramana (working name) runs that formula itself, in code, on the real inputs.
Numbers in the drafted text are checked against the locked value. Any mismatch is flagged. Tying each dossier row and, later, each slide to the number itself is being built.
Worked out from measured sources, it becomes a locked number. Built on an assumption, it stays a labeled estimate that everyone can see.
MLR (medical, legal and regulatory) reviewers approve worked-out numbers, not paragraphs. Pramana (working name) records who signed and when.
Value, unit, formula, sources, time and signer, in one page any payer can open.
Locked. Anyone can recheck this number.
Example receipt. Values are illustrative.
Being built
Dossiers are updated over time, for example after a price change or a new assumption. Today each Pramana (working name) release gets its own receipt, and the record ends there. Pramana (working name) is adding a history of each number across versions: a chain of locked versions, not a single lock. None of this is live yet.
A revision or clarification sent through Pramana (working name) will be a new version of a number it already holds, locked on its own by a named signer. Nothing earlier in the chain will be overwritten. The receipt for version two, or any later version, will point to the receipt and the number version before it.
If a pack going out through Pramana (working name) is marked as a revision or clarification, it must link back to a number Pramana (working name) already holds. If it does not, Pramana (working name) will refuse to send it.
When a plan changes the inputs to see its own numbers, that scenario will stay a separate, labeled item. It is never a version in your chain and never becomes your signed version. This is a product rule of Pramana (working name), not an AMCP requirement.
A health plan that opens the receipt, with no account, will see the chain, or see plainly that it is missing.
Your model stays the calculator. Anyone can still email a new spreadsheet outside Pramana (working name), and plans can still run their own scenarios. What it is built to refuse is narrower: a revision or clarification sent through Pramana (working name) that does not link to a number it already holds. Receipts already issued stay exactly as issued. The chain will link them, not change them.
Value
Pramana (working name) does not make a product look better. It saves the hours spent chasing numbers and protects the launch from avoidable slips. Anything we estimated is marked as an estimate, the same rule Pramana (working name) applies to its own numbers.
How we got this: About 40 key numbers, each repeated in about 5 places (dossier text, PIE deck, payer slides, budget calculator, executive summary), makes 200 spots. At about 10 minutes each by hand, that is about 33 hours per review round, times 3 rounds: about 100 hours. Today Pramana (working name) checks drafted text going out through it, which here is the dossier text and the executive summary: 80 spots, about 13 hours per round, about 40 hours over 3 rounds. The PIE deck, payer slides and calculator (120 spots, about 60 hours) are still checked by hand, because slide checks come later. The time to review what Pramana (working name) flags is not subtracted, so the saving is an upper bound. Pramana (working name) also re-runs and locks the 40 numbers themselves. That time is not counted, because this method has no assumption for it. FDA expects payer information to be updated when it goes out of date, so this repeats with every refresh 5.
~$500K
average drug sales per day of delay 3
Source: researchers at the Tufts Center for the Study of Drug Development (Smith, DiMasi and Getz), published in Therapeutic Innovation & Regulatory Science in May 2024. They measured average daily sales of prescription drugs and biologics across 645 drugs launched since 2000, as the sales missed for each day of delay. This is an average across many drugs. It is sales, not profit, and one plan's decision is only a slice of it. Many P&T committees meet on a set schedule, often quarterly, so the 90-day case is an assumption. The point is scale.
| Question | Locked before it leaves | Corrected after someone else finds it |
|---|---|---|
| When the error is found | Inside the company, before MLR sign-off | After payers, ICER or FDA have already seen it |
| Who finds it | Pramana (working name) flags it automatically | A P&T reviewer, a PBM analyst, or a regulator |
| What it costs | A few hours to fix a flagged number | Delayed coverage, tougher rebate talks, stricter access rules |
| What you can say | “Here is the receipt for the numbers we released.” | “We will look into it and get back to you.” |
We will measure, not guess. The first engagement locks one live results table, checks the drafted text around it, and records the real hours before and after. Those measured numbers replace every estimate here.
Who it helps
Each person who touches the number gets something different.
They approve worked-out numbers, not paragraphs. If a number cannot be locked, it does not go out.
They keep the model they trust and stop chasing numbers through the drafted dossier text and executive summary.
A wrong number in front of a PBM or P&T committee can delay coverage, weaken rebate talks, or lead to tighter access rules.
They can check the receipt without buying anything. Trust moves from “we like this company” to “we can re-run this.”
A record of sources, formula, value, signer and release that can help answer an FDA or CMS request. It is not a compliance check, and neither agency endorses Pramana (working name).
Limits
Pramana (working name) does not:
If the model is wrong, the lock will faithfully lock the wrong math. The benefit is a trustworthy release, not a better economic answer.
Start
The US AMCP (Academy of Managed Care Pharmacy) format is the production candidate for Pramana (working name). Lock the budget impact table in one live AMCP dossier, and check the numbers in the drafted text that goes with it. You keep your model and your MLR process. We measure the hours saved, and those measured numbers replace every estimate on this page.
Time and money figures marked as estimates are illustrative and will be replaced by measured results from the first engagements. Nothing on this page is legal or regulatory advice. Looking for the other market? Read the Canada edition (in progress).
If you are about to send a table whose numbers you cannot re-run, talk to us.
VedAGI. Intelligence for Humanity.