Verify the molecule,
not the label.

Independent analysis of grey market peptide seizures found that 30% of samples contained the wrong molecule entirely (Janvier et al., 2018). 40% of vendors fail their stated purity thresholds across 6,100 samples tested by Finnrick Analytics. 65% of injectable preparations exceed endotoxin safety limits. The FDA issued 50+ warning letters to GLP-1 compounders in a single month in September 2025. Fourteen of nineteen Category-1 peptides were reinstated for 503A pharmacies in February 2026.

Existing identity layers stop at sequence-name lookup. They confirm the peptide name exists. They do not confirm the peptide is in the vial. Titer treats that gap as the entire problem.

30%

Grey market peptides · wrong molecule · Janvier et al. 2018

40%

Vendors fail stated purity · Finnrick Analytics · 6,100 samples

65%

Injectables exceed endotoxin limits · Rohrbough 2020

14 / 19

Cat-1 peptides reinstated · 503A · Feb 2026

One paste, five stages, single dossier. Cache hits short-circuit any stage that has already resolved for an identical canonical sequence.

1. § 01
   Canonicalize
   parse · normalize · sha-256
2. § 02
   Identity
   UniProt PS → BLAST → grade
3. § 03
   Structure
   AlphaFold v6 / ESMFold / PEP-FOLD4
4. § 04
   Batch + COA
   vendor bond · lab COA · MS hash
5. § 05
   Receipt
   commit · reveal · Arweave

The dossier ships as a typed JSON object plus a Markdown rendering. Every claim that survives to the dossier carries a citation that resolves to a real public record: UniProt accession, AlphaFold model, ChEMBL ligand, OpenTargets evidence row, PubMed PMID. If a citation does not resolve, the claim does not ship.

Titer ships with a curated table of canonical research peptides. An exact backbone match short-circuits to grade A without an external API call. Modified analogs (D-amino acids, amidation, conjugates, DAC, fatty-acid spacers) share the backbone and surface the full modification list separately.

Sequences are research-only. No medical claim is made or implied. Modifications surface in the dossier alongside the canonical core.

Every lookup generates a tamper-evident commit receipt before the bio APIs run. The commit is a blinded hash of the sequence, a random salt, and the request timestamp. The salt is revealed alongside the result, binding the output to the commitment. Phase 2 anchors this to Solana mainnet so the binding is permanent and verifiable without Titer.

// commit (lands at confirmed, ~400ms)
titer:commit:v1:<sha256(sequence || salt || requested_at)>:<requested_at>

// reveal (lands after dossier upload)
titer:reveal:v1:<commit_tx_sig>:<salt>:<result_cid>:<result_hash>

On-chain anchoring costs ~$0.0015 per lookup in Solana fees. Ships in phase 2. The commit receipt is already generated on every lookup today.

The lab uploads the raw spectrum (mzML / mzXML). Titer computes a content-stable hash by binning peaks at 0.01 Da and rounding intensity, then SHA-256. The hash is bound to the batch on chain via a lab-signed memo. Anyone with the raw spectrum can re-hash and verify.

titer:ms:v1:<batch_id>:<sha256>:<lab_id>:<acquired_at>

input:   peaks[] = [{mz, intensity}, ...]
binning: bin = round(mz / 0.01)
norm:    intensity / max_intensity, threshold 0.001
hash:    sha256(sorted "bin:intensity" entries)

Vendors stake $TITER to list. Bond size scales with cumulative batch GMV and is multiplied by the vendor’s historical incident rate (capped at 4×). Filed disputes lock the bond pending arbitration. Slashes pay out to the impacted buyers from the same pool.

5%

Bond floor · cumulative batch GMV

1× → 4×

Incident-rate multiplier · ramps with off-spec history

2%

Default per-vial insurance premium · routed to claims pool

≤ 0.5×

Quality-score discount on insurance · cohort-fed

Every reasoning task (novelty scoring, mechanism synthesis, evidence grading, claim verification) runs through gpt-5-2 via kie.ai with high reasoning effort. The Red Team and Synthesizer fail hard if the model is unreachable or the response cannot be parsed into the expected shape. Templated stubs are never written to the ledger.

upstream → real public APIs                  (PubMed, UniProt, AlphaFold ...)
reason   → kie.ai · gpt-5-2 · effort: high
critique → red team rejects unsupported claims
verify   → cited-claim agent re-checks each citation against the source
ship     → only on verified citations + structured output parse

Buyers log N-of-1 outcomes against a verified batch ID. Logs are pseudonymous (wallet pubkey only) and processed in the browser by default. Biomarker uploads stay in encrypted IndexedDB unless the user opts in to aggregate publication.

N ≥ 1000

Cohort grade A · controlled signal

N ≥ 200

Cohort grade B · preliminary

N ≥ 50

Cohort grade C · directional

N < 50

Cohort grade D · insufficient evidence

• →Not a vendor. Titer does not sell, broker, or resell peptides. There is no checkout path on this site.
• →Not a recommendation engine. The vendor table shows bond, COA freshness, and dispute history, never a 'best-of' rail.
• →Not a medical service. All output carries a research-only disclaimer. No dose, no protocol, no treatment language is generated.
• →Not a closed garden. Public bio APIs are public, on-chain receipts are verifiable without Titer's servers, and the source is open.