A clinical engine that cannot drift.
Ten-category scoring, an evidence-tiered medication cross-reference, and a dose layer that reads every number from one locked card library. Every rule in the core cites its source. Every ruling is enforced by a test. Every report is held until an automated reviewer clears it.
Eight layers. One direction of authority.
The scoring brain is locked and never edited. Policy, safety and owner rulings live in wrappers around it, so the rules can evolve without touching the core. Select a layer.
When two sources disagree, this order decides
Watch one reference plan move through every layer.
Reference profile "Marcus" is one of the twelve locked 90-Day test plans: Advanced level, metabolic focus, HbA1c 5.9, fasting glucose 108, GLP-1 preference "let the report suggest". These are the engine's real outputs.
Every medication class and condition, against every compound.
17 medication classes and 22 conditions, crossed with 26 compounds. Every cell is filled and gated. Brand names resolve to generics, and negated mentions such as "stopped" or "former" are ignored. Covers all 73 intake safety boxes and all 14 medication-picker classes.
- Antipsychotic → Semax tier B
- Seizure history → Cerebrolysin tier B
- Seizure history → Semax tier B
- Heart disease → PT-141 tier B
Substitution: when Semax is withheld for an antipsychotic and the client has a brain signal, Cerebrolysin takes the slot, unless it is withheld too.
What a verdict is allowed to rest on
A withhold requires human evidence. Anything weaker can only put a card in front of the prescriber. The gate refuses a withhold keyed on cancer.
Randomized trial or systematic review
can withholdDrug label or human cohort
can withholdMechanism
card onlyAnimal data
card onlyLibrary caution text
card onlyA removed compound is never silently dropped. It stays on the page with what it is and the documented reason, with no dose, no schedule and no slot on the chart. A client who is never told could source it elsewhere without knowing why it was a problem."Explained, not scheduled," the clinician-founder's executive command, applied to every contraindication on every report surface
Nothing ships unless every gate passes.
The system is built to prove, on every change, that nothing moved that was not meant to move.
Locked reference plans
Metabolic, perimenopause, first-timer, pancreatitis, libido-only, hormone-led, brain fog, explicit GLP-1 pick, skin and gut, active cancer, Moderate at ceiling, visceral fat. Each rebuilds byte for byte.
verify_90Day_locked_set.pyContent-locked files
Every locked sentence must survive. A locked report changes only through a refreeze script that proves only the approved lines changed. Anything else aborts.
verify_content_lock.py · refreeze_*.pyRulings, each with a test
Every owner ruling is registered with the tests that enforce it. A ruling is removed only by a logged reversal.
OWNER_RULINGS.json · verify_owner_rulings.pyParity canaries
Nine profiles across four products, compared against golden copies to prove the live site matches the package exactly.
verify_live_parity.py --checkTests passing
Every fix ships with a test that fails before it and passes after. Nightly pressure tests add thousands of fuzzed intakes, hostile input and every dose recomputed.
pytest · run_audit.pyLogged corrections
Every change since the first report: what changed, why, and the proof. Fingerprints pin the package; the core is verified on every run.
Owner_Corrections_Log.md · verify_package.pySeven agents. None can touch a dose.
Every report is held at generation. Only the reviewer can release it. The fixer repairs display issues with tests and every gate, and never touches a dose, a clinical rule or the locked core. No agent contacts a customer, and no agent trusts another agent's word.
- Reviewer
Runs 19 checks on every held report. The only path to release.
- Fixer
Re-verifies findings, fixes the display layer, proves it with tests and gates, regenerates.
- Chief of staff
Reads the agents' mail, finishes stuck work, reports only when something happened.
- Nightly auditor
Pressure-tests the whole engine and reports only what is new.
- Evidence watch
New studies, FDA and compounding changes, graded A to E. Every library change is the founder's decision.
- Competitive intel
Who else sells this, what they charge, where the gaps are.
- Daily brief
One e-mail: decisions needed, what was fixed overnight, what could not be verified.
One card library. One formula. No exceptions without a ruling.
Units on a U-100 syringe = dose (mcg) × diluent (mL) ÷ (vial (mg) × 10)
Every card was checked against this formula. Nine report surfaces must agree on every vial value, and a reconciliation gate checks them. The only departures are logged, owner-approved roundings, such as GHK-Cu printed at 10 units where the exact figure is 12. GLP-1s are always research vials with titration ladders, each step held at least four weeks.
Check any card
units
mL
uses per vial
Reference values the engine prints for clinician review. Nothing here chooses a dose for a person.
Eight months, corrected one card at a time.
Designed and directed by a clinician-founder with more than 30 years of peptide experience, who teaches peptide therapeutics to other clinicians.
First reports
Corrected card by card. The no-fabrication rule is written the day an invented dose is caught.
The model
GLP-1 by preference only. Staged 90-day plan. Content locks. 21 and older only.
The true engine
Ten-category scoring, compound ceilings, locked reference plans, side-effect library sign-off.
Live generators
Every product on the live path, the intake echo page, the parity canary, lab upload.
Launch
Live payments. Then hold-and-review, the agents, the rulings registry and Library 30.
Ask it anything. Out loud is fine.
Answers come only from the engine's own documentation, pitched at expert level. If something isn't covered, it says so.
Press the microphone and speak, or type. Press M anywhere to talk.
Ask from your own phone with the same access code