OmniRayn · Foundry Classroom · UGA BCMB 4030L

BioProcess Technology
Capstone — Cordycepin

The Foundry predicted a molecule from a genome. Your class ferments the organism and decides — by measurement — whether the prediction holds.

CAS 73-03-0 PubChem CID 6303 3′-deoxyadenosine · MW 251.24 Cluster Cns1–4
GENOTYPE → CHEMOTYPE
Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
01
OmniRayn/BioProcess Capstone · BCMB 4030L
Genotype → chemotype

The Foundry predicted a molecule. Your job is to test it.

The Foundry read the C. militaris genome, called the Cns1–4 cluster, and predicted cordycepin (3′-deoxyadenosine) as its product. That prediction is computational. This capstone closes the loop — you ferment, isolate, and decide by measurement whether it holds.

What the Foundry actually did

  • Fungal-BGC caller scanned the C. militaris genome and identified the Cns1–4 cluster (the cordycepin/pentostatin locus)
  • Predicted the product as cordycepin — a known, public, off-patent nucleoside: CAS 73-03-0, PubChem CID 6303
  • A genome→compound biosynthetic prediction — no molecule was made or measured by the Foundry
  • Exchange dossier: dos-cordycepin-genome-discovery

What YOU do — close the loop

  • Ferment C. militaris (shake-flask → benchtop bioreactor), then isolate and purify the candidate
  • Confirm or refute identity by HPLC-UV (~260 nm) against an authentic cordycepin standard, plus MS
  • Report identity, purity, and titer with a CI against a release spec you did not set
  • The wet lab is the honesty gate — the referee call is yours, not the algorithm's

Why this is genuinely hard

  • Cordycepin must be resolved from near-identical adenosine and co-produced pentostatin — a real separation problem
  • Filamentous morphology drives non-Newtonian, shear-thinning broth and kLa/OTR collapse on scale-up
  • Fermentation must be justified on the numbers vs a viable chemical-synthesis route (COGS, productivity, DSP recovery)
  • Cordycepin is a bioactive nucleoside analog with documented cytotoxicity — a live safety/QC teaching point

The honest advertisement for the Foundry

  • The Foundry's claim is falsifiable — and this course is built to try to falsify it, in public, with data
  • A confirmed prediction shows genome→compound discovery works on a real, testable case
  • A refuted prediction is equally valuable and equally graded — it maps the platform's limits honestly
  • Scope: this course characterizes a MOLECULE (identity, purity, titer) — zero efficacy or disease claims
The Foundry output is a COMPUTATIONAL prediction, pending wet-lab. Keep the Foundry PREDICTION and your MEASUREMENT visually and verbally separate at all times — never present a student HPLC trace as if the Foundry validated it. Any bioactivity or indication is out of scope, stamped “computational hypothesis, pending independent wet-lab efficacy studies — out of scope for this course.” Confirming and refuting the prediction are both A-grade outcomes; you are the referee, not the fan club.
Student deliverable

A one-page “prediction-vs-plan” brief: restate the Foundry's Cns1–4 → cordycepin prediction (Exchange listing cited), state it explicitly as computational/pending-wet-lab, and define the falsifiable test each team will run — the specific identity, purity, and titer criteria by which they will confirm or refute that cordycepin is present, keeping predicted claim and planned measurement in separate columns.

Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
02
OmniRayn/BioProcess Capstone · BCMB 4030L
Data access

You get the prediction and the public genome — not the engine.

Nothing here is gated behind the Foundry. The genome is public, the prediction is a shareable hypothesis, and you can re-run the computational call yourself — so the referee stance holds at every step, on fully reproducible data.

The genome — public

  • The C. militaris genome is on NCBI (assembly CM01; Zheng et al., Genome Biology 2011) — download it yourself
  • The cordycepin/pentostatin Cns1–4 cluster is published (Xia et al., Cell Chem Biol 2017)
  • Open reference data — no special access required

The Foundry prediction — the hypothesis

  • You receive the BGC-prediction dossier: Cns1–4 → cordycepin (Exchange: dos-cordycepin-genome-discovery)
  • A computational call, stamped pending wet-lab — the thing you are here to test
  • No molecule was made or measured by the Foundry

Re-call it yourself — antiSMASH

  • Run the free, open antiSMASH tool on the public genome to independently call the cluster
  • Compare your call to the Foundry's — you are the referee on the computational step, not just the wet lab
  • Agreement or disagreement with the Foundry is itself a result to report

The firewall — outputs, not the engine

  • You consume the Foundry's outputs (a prediction) — never its engine, BGC caller, or trained models
  • The honest advertisement: a powerful, testable capability shown on public, reproducible data
  • Public genome + public tool + shareable hypothesis = reproducible by any lab
Student deliverable

A one-page data-provenance note: the exact NCBI accession used, your antiSMASH cluster call side-by-side with the Foundry's Cns1–4 prediction, and an explicit statement of what is public (genome, compound, prediction) versus proprietary (the Foundry engine).

Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
03
OmniRayn/BioProcess Capstone · BCMB 4030L
Upstream process design

Upstream: strain, media, morphology, and a DoE that actually moves titer.

The Foundry predicts the Cns1–4 cluster encodes the machinery to make cordycepin. Before confirming that molecule downstream, you must grow the organism — and every upstream choice sets a ceiling on how much cordycepin the cells can secrete. This is where you engineer that ceiling.

Strain handling & aseptic technique (the invisible failure mode)

  • C. militaris is a slow-growing ascomycete; one contaminant outgrows it and yields a broth that looks fermented but has zero cordycepin — aseptic technique is a titer decision
  • Work from a defined, low-passage seed train; repeated subculturing degenerates the strain and can silence Cns1–4 output
  • Standardize inoculum age and density — inconsistent inoculum is the biggest source of run-to-run titer scatter
  • Log passage number, seed morphology, contamination checks — an unlogged event is uninterpretable later

Media: C:N ratio, carbon source, adenosine precursor

  • Cordycepin is 3′-deoxyadenosine — a purine analog — so it draws on the adenosine precursor pool; feeding adenosine is rational but risks the neighbor pool you must later separate
  • C:N ratio governs the growth-vs-secondary-metabolism trade-off; N-limitation often pushes idiophase production
  • Light and DO are documented modulators — control and record them even when not DoE factors
  • In ink: precursor feeding is a hypothesis to test; any co-eluting neighbor it creates becomes a downstream cost — upstream and downstream are coupled

The pellet-vs-dispersed decision (morphology governs everything)

  • Submerged fungi grow as compact pellets or dispersed mycelia — a controllable outcome and the master variable of the whole process
  • Dispersed mycelia give more surface area / productivity but make broth non-Newtonian and shear-thinning — which collapses OTR on scale-up
  • Pellets keep viscosity low and OTR easier, but core mass-transfer limits starve interior cells and cap productivity
  • The honest hinge: the morphology you pick upstream is the one you live with downstream (filtration, OTR, shear) — justify it against whole-process consequence

DoE: fractional-factorial screen → RSM optimization

  • Stage 1 — fractional-factorial / Plackett-Burman screens many factors cheaply to find the vital few, controlling for aliasing
  • Stage 2 — RSM (CCD or Box-Behnken) on survivors maps curvature; replicated center points give honest pure-error and lack-of-fit
  • The response is a MEASURED cordycepin titer (from downstream HPLC) with a CI — not a Foundry-predicted number
  • A statistically flat surface (no significant factor) is a legitimate, gradeable result — real information, not a failure
Keep the two claims physically separate: the Foundry's genome→cordycepin prediction is a COMPUTATIONAL hypothesis, pending wet-lab. Your DoE optimizes and your HPLC later confirms the actual MOLECULE (identity, purity, titer) — zero efficacy or disease claims, and any reported bioactivity is out of scope. A higher titer proves you grew more of a compound; it does not prove the compound is cordycepin until the downstream identity gate says so.
Student deliverable

An upstream process design package: (1) a documented seed-train and aseptic SOP with a justified inoculum spec; (2) a defined production medium with mechanistic rationale for C:N ratio, carbon source, and the adenosine-feeding decision; (3) a declared target morphology argued against its downstream OTR/filtration/shear consequences; and (4) a two-stage DoE — fractional-factorial screen then RSM model — with ANOVA, fitted response surface, and predicted optimum, response = measured cordycepin titer with CI, plus a one-paragraph honesty statement separating Foundry prediction from student measurement.

Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
04
OmniRayn/BioProcess Capstone · BCMB 4030L
Scale-up cliff

Bioreactor scale-up: where the flask lies to you.

Moving from a 250 mL flask to the 2–5 L benchtop reactor, filamentous morphology turns the broth non-Newtonian, oxygen transfer collapses in the poorly-mixed core, and titer usually drops. That drop is the physics of the system — and characterizing it is the point.

The oxygen-transfer collapse (the intrinsic hard problem)

  • OTR = kLa × (C* − C_L): as mycelia grow, apparent viscosity climbs and kLa falls, so supply cannot meet mycelial demand (OUR)
  • Shear-thinning broth (power-law, flow index n<1) shears thin near the impeller but stays viscous in bulk — creating stagnant, O₂-starved dead zones
  • Measure kLa directly (dynamic gassing-out) in water, then in real broth at low and high biomass — watch it drop an order of magnitude
  • DO sagging below setpoint despite full agitation/sparge = the OTR ceiling made visible; log DO gradients top vs bottom

The shear-vs-OTR tradeoff (no free lunch)

  • More RPM raises kLa and shrinks pellets — but higher tip speed shears hyphae and can wreck the biomass you need
  • Impeller choice: high-shear Rushton (radial, Np≈5) vs axial-flow pitched-blade/hydrofoil (Np≈1–1.5) for bulk mixing at lower shear
  • Power draw P = Np·ρ·N³·D⁵; compute P/V and gassed power, reconcile against mixing time and kLa
  • Pellet vs dispersed morphology is the hidden control variable — it sets viscosity, kLa, internal mass-transfer, and filterability at once

Fed-batch & precursor-feed design

  • Batch runs out of carbon/nitrogen and hits DO/viscosity limits; fed-batch glucose keeps growth in a controllable, O₂-matched regime
  • Adenosine precursor feeding is the pathway lever — design feed timing/rate as a testable hypothesis, let HPLC referee it
  • Feed strategy is a DoE variable (feed rate, start time, C:N) with volumetric productivity g/L/h as the response — links to techno-economics
  • Control-loop tuning (pH; DO via cascade RPM→sparge→O₂ enrichment; temp) is where feeding lives or dies operationally

Feel the cliff — and quantify it

  • Report titer, volumetric productivity (g/L/h), and specific productivity side-by-side for flask vs reactor; expect the reactor to look worse per-liter first
  • Attribute the gap: OTR-limited, shear-damaged morphology, mixing-time-limited, or feed-limited? Evidence, not vibes
  • Mixing time (t_m) rising with viscosity is a diagnostic you measure (tracer/pH-pulse) and tie to the DO gradient
  • The engineering deliverable is a defensible scale-up account — a well-characterized failure to scale is a passing result
The flask-to-reactor titer drop is EXPECTED and gradeable. You are not graded on beating the flask — you are graded on measuring kLa/OTR/DO honestly and correctly diagnosing WHY the number moved. A pretty titer with no transport-phenomena evidence loses to an ugly titer that is fully explained. Any bioactivity of the molecule is out of scope — this section produces process data (kLa, OTR, DO, titer, g/L/h), not efficacy or disease claims.
Student deliverable

A scale-up engineering report: measured kLa (water vs low- and high-biomass broth), logged DO gradients and mixing time, a documented impeller/agitation/sparge and fed-batch/precursor-feed strategy with rationale, and a flask-vs-reactor comparison of titer and volumetric productivity (g/L/h) with a mechanistic diagnosis (OTR-limited / shear-damaged / mixing-limited / feed-limited) of the observed change — refutation of a hoped-for improvement graded equally with confirmation.

Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
05
OmniRayn/BioProcess Capstone · BCMB 4030L
Unit ops & resolution

Downstream: recovery and the separation problem.

Cordycepin (target)
3′-deoxyadenosine · C₁₀H₁₃N₅O₃ · MW 251.24
The predicted product. Elutes slightly later than adenosine on C18 (missing 3′-OH → less polar).
Adenosine (near-identical)
C₁₀H₁₃N₅O₄ · MW 267.24 · Δ = one O
Differs by a single oxygen; shares the ~260 nm chromophore. UV alone cannot tell them apart.
Pentostatin (co-product)
2′-deoxycoformycin · potent ADA inhibitor
Co-produced by Cns1–4. A real handling/QC hazard and a third peak to resolve and quantify.

Biomass separation: the filamentous cake fights you

  • Pellets/dispersed mycelia yield a compressible, near-impermeable cake — dead-end filtration blinds fast (flux collapse, rising ΔP)
  • Choose and defend a route: depth/rotary-vacuum + filter aid, basket centrifugation, or crossflow MF — scored on flux, retention vs cake loss, clogging
  • Decision fork you own: cordycepin is largely EXTRACELLULAR but a fraction stays cell-associated — assay both broth and lysate, decide if intracellular recovery is worth it
  • Cake washing (displacement wash) recovers entrained product — quantify wash-recovery vs dilution penalty

Extraction, clarification, capture

  • Primary capture by adsorption onto polymeric resin or activated carbon, then step-gradient desorption — report dynamic binding capacity and breakthrough
  • Polish by prep RP-HPLC (C18, aqueous/low-MeOH) and/or crystallization from water/ethanol — cheap but rejects impurity poorly if mother liquor is uncontrolled
  • Lyophilize the final aqueous fraction; record water activity / residual solvent (a nucleoside cake holds water)
  • Every step logs an aliquot for HPLC so recovery is measured per unit op, never inferred at the end

The resolution stress test: cordycepin vs adenosine vs pentostatin

  • Cordycepin (MW 251.24) differs from adenosine (267.24) by ONE oxygen; both absorb at ~260 nm — a UV peak alone can't tell them apart, and co-elution silently inflates apparent purity
  • Method development is the graded skill: tune phase, %organic, pH/ion-pairing to baseline resolution (report Rs, plate count, tailing); confirm with authentic standard + MS/DAD, not retention alone
  • Pentostatin (2′-deoxycoformycin) is co-produced by Cns1–4 — a potent ADA inhibitor, a real hazard and a third peak to resolve and quantify
  • Deliverable resolution table: per impurity report Rs, spike-recovery, and pass/fail vs the imposed spec — a fail here is legitimate and gradeable

The mass-balance ledger (yield accounting)

  • One ledger tracks cordycepin mass (mg) and % step-recovery across filtration → capture → polish → crystallization → lyophilization
  • Name and quantify losses: entrained in wet cake, incomplete desorption, mother-liquor rejection, mass lost to co-elution when neighbors are trimmed off
  • Feeds techno-economics: step-recovery × titer sets g/L delivered → COGS vs the chemical-synthesis route — justify or refute fermentation on the numbers
  • Consistency check: (product out + measured losses) reconciles to product in per step; an unbalanced ledger flags an assay error to debug, not a result to report
Keep the two claims physically separate: the Foundry PREDICTED cordycepin as the Cns1–4 product (computational, pending wet-lab); the student HPLC trace is the independent MEASUREMENT that confirms or refutes it. Never present a student chromatogram as if the Foundry validated it. This section characterizes a molecule (identity, purity, titer) and makes zero efficacy or disease claims; cordycepin is a bioactive, cytotoxic nucleoside analog — handled as a real safety/QC teaching point. Refuting the prediction (spec fail, wrong peak, no cordycepin) is an A-grade outcome.
Student deliverable

A downstream process report: (1) a per-unit-op mass-balance ledger with cordycepin mg and % step-recovery and an overall DSP recovery that reconciles losses to specific steps; (2) a validated resolution table showing baseline separation (Rs, plate count, tailing) of cordycepin from adenosine and pentostatin, identity confirmed by authentic standard + MS/DAD spectral match and spike-recovery; and (3) a pass/fail call of the final isolated solid against the imposed release spec (identity confirmed, ≥95% purity, titer with CI) — a documented refutation counts as full credit.

Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
06
OmniRayn/BioProcess Capstone · BCMB 4030L
Genotype → chemotype gate

Analytics & QC: prove it, or fail spec.

Foundry — PREDICTION

Cns1–4 → cordycepin

A computational genome→compound call. No molecule made or measured. Stamped pending wet-lab. This box never contains a student chromatogram.

≠
Student — MEASUREMENT

Your HPLC-UV trace

The independent honesty gate: identity, purity, titer±CI vs a release spec. Confirms or refutes the prediction — and is never shown as Foundry validation.

Build the method (don't borrow a number you can't defend)

  • RP-C18, aqueous methanol/ACN, UV/DAD at ~260 nm (adenine chromophore, λ-max ~259–260 nm); no ion-pairing needed
  • Critical pair = cordycepin vs adenosine: cordycepin lacks the 3′-OH → slightly less polar → elutes LATER on C18; resolve to baseline, Rs ≥ 1.5
  • Pentostatin (2′-deoxycoformycin), the co-produced ADA inhibitor, is a third peak to account for, not ignore
  • System suitability first: %RSD of replicate injections, tailing factor, retention reproducibility — before you trust any sample

Partial validation (ICH Q2 flavor, undergrad scope)

  • Linearity: ≥5 authentic-standard levels bracketing expected titer; report R² AND a residual plot (curved residual = wrong range or overload)
  • LOD/LOQ from signal-to-noise or the σ/slope method — state which and why
  • Accuracy by spike-recovery into blank/low broth matrix; report % recovery and acknowledge matrix effects
  • Specificity: prove your cordycepin peak is not adenosine or pentostatin hiding underneath — peak purity via DAD

Identity: co-elution is necessary but NOT sufficient

  • Co-elution with authentic standard = same retention time — a strong but defeatable claim (adenosine can be close)
  • DAD spectral overlay helps, but cordycepin and adenosine share the adenine chromophore — UV alone will NOT cleanly separate them; say so in ink
  • Honest identity = orthogonal confirmation: co-elution + spectral match + LC-MS mass (C₁₀H₁₃N₅O₃, [M+H]⁺ ~252.1) when available
  • HPLC-UV only? Your claim is “consistent with,” not “confirmed” — the grade rewards the correct hedge

The release spec you must hit (given, not negotiated)

  • Identity confirmed by the standard above; purity ≥ 95% (area %); titer as mean ± 95% CI from replicate ferment/analysis
  • A result that misses spec is a real, gradeable outcome — a well-argued FAIL beats a hand-waved PASS
  • You are the referee on the Foundry's prediction, not its fan club — refuting “cordycepin present at spec” is an A-grade result if analytics are sound
  • Never present your student trace as Foundry validation — prediction and measurement live in separate boxes, always
The Foundry output is a COMPUTATIONAL prediction, pending wet-lab — your chromatogram is the honesty gate on it, and the two must never be shown as one. You are characterizing a MOLECULE (identity, purity, titer) and making zero efficacy or disease claims; any bioactivity is a computational hypothesis, pending independent wet-lab efficacy studies, out of scope. Cordycepin is a bioactive nucleoside analog with documented cytotoxicity — handle it as a genuine safety/QC concern, not a clean-tox assumption.
Student deliverable

A partial method-validation report + Certificate of Analysis: HPLC-UV method with a linearity curve (≥5 levels, R² with residual plot), LOD/LOQ, spike-recovery (%), resolution (Rs) for the cordycepin/adenosine/pentostatin critical pair, plus an identity determination (co-elution + DAD spectral match, LC-MS m/z 252.1 [M+H]⁺ if available) and a PASS/FAIL verdict against the release spec with titer as mean ± 95% CI. A rigorous, correctly-argued FAIL earns full marks.

Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
07
OmniRayn/BioProcess Capstone · BCMB 4030L
Techno-economics & the verdict

Prove it on the numbers: productivity, recovery, cost.

The molecule is not the deliverable — the defensible accounting is. Cordycepin CAN be chemically synthesized, so fermentation is not assumed; students justify or refute it on the arithmetic they generated. The Foundry predicted the cluster; the students own the verdict.

Volumetric productivity + batch vs fed-batch

  • Report titer (g/L, with a bootstrapped/t-based 95% CI from replicate HPLC), duration, and Qp = titer / EFT (g/L/h) — not just peak titer
  • Compare batch vs adenosine-fed-batch on the SAME axes: final titer, Qp, specific yield on carbon/precursor; feeding often raises titer but can depress Qp if it stretches EFT
  • Reconcile magnitudes: filamentous cordycepin titers are typically mg/L to low g/L over days-to-weeks → small Qp; a suspiciously high Qp is a flag to re-audit calibration, not a win
  • Tie the ceiling back to the intrinsic hard problem: kLa/OTR collapse in shear-thinning broth caps OUR and therefore Qp on scale-up

Stepwise DSP recovery ledger

  • Mass-balance ledger with measured cordycepin mass (mg) and step recovery (%) at EVERY unit op: separation, extraction, clarification, chromatography, lyophilization
  • Overall recovery = product of step yields; show the multiplicative erosion (five 85% steps ≈ 44% overall) and identify the recovery-limiting step
  • Close the balance: in + generated = out + losses within a stated tolerance; an unclosed ledger (>~10–15% unaccounted) is a finding to explain, not hide
  • Purity is gated: recovery without the release-spec purity attached is incomplete — pulling pentostatin/adenosine forward inflates apparent mass

COGS vs the chemical-synthesis route

  • Cordycepin is chemically synthesizable — fermentation is NOT assumed; assemble order-of-magnitude COGS: media + precursor + utilities (aeration/agitation power for OTR) + DSP consumables + labor/time, per gram of spec-grade product
  • Contrast against a bounded synthesis cost driver (protected-nucleoside starting materials, step count, chromatographic purification); state assumptions and cost basis
  • Deliver a defensible VERDICT: dominant cost driver identified, sensitivity to titer/Qp and DSP recovery shown, break-even titer estimated
  • Keep it honest: a techno-economic teaching estimate at bench scale, NOT a manufacturing quote; no efficacy, indication, or market claims

Deliverables & the refutation-rewarding rubric

  • Six artifacts: (1) PFD, (2) mass-balance ledger, (3) validated HPLC method + release report (identity, linearity, LOD/LOQ, spike-recovery, purity, titer±CI vs a spec students did NOT set), (4) techno-economic memo, (5) genotype→chemotype confirmation statement, (6) troubleshooting/DoE log
  • The confirmation statement keeps Foundry PREDICTION and student MEASUREMENT visually and verbally separate; the computational call is stamped “pending wet-lab,” and a student trace is never Foundry validation
  • Rubric weights RIGOR OF EVIDENCE, not the direction of the result: a well-argued “cordycepin NOT confirmed / spec NOT met” with sound error analysis earns the same top band as a clean confirmation
  • Grade on: method-validation quality, mass-balance closure, correct CI/uncertainty handling, honest spec pass/fail call, and a COGS verdict that follows from the students' own numbers
A confirmation with sloppy calibration is a fail; a rigorous refutation is an A. Students are the referee on the Foundry's computational prediction, not its fan club — and the spec (identity, ≥95% purity, titer with CI) is one they must hit but did not set, so failing it honestly is a legitimate, gradeable outcome. No efficacy or disease claims appear anywhere; the deliverable is a characterized molecule and defensible accounting.
Student deliverable

A techno-economic decision memo (volumetric productivity g/L/h, stepwise DSP recovery ledger with closed mass balance, batch-vs-fed-batch comparison, and fermentation-vs-chemical-synthesis COGS verdict), bundled with the validated HPLC release report (titer ± 95% CI, purity vs an un-set release spec) and a genotype→chemotype confirmation statement that keeps Foundry prediction and student measurement strictly separate — graded on evidentiary rigor, with refutation rewarded equally to confirmation.

Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
08
OmniRayn/BioProcess Capstone · BCMB 4030L
Deliverables · Grading
Deliverables & grading

A rubric that pays for “no” exactly as much as “yes.”

The six artifacts

Process flow diagram · mass-balance ledger · validated HPLC method + release report (identity, linearity, LOD/LOQ, spike-recovery, purity, titer±CI) · techno-economic memo · genotype→chemotype confirmation statement · troubleshooting/DoE log.

Prediction ≠ measurement

The confirmation statement keeps the Foundry PREDICTION and the student MEASUREMENT in separate columns. The computational call is stamped “pending wet-lab.” A student HPLC trace is never presented as Foundry validation.

Refutation = confirmation

Graded on defensibility, not the direction of the result. A clean confirm, a rigorous refute, or a well-characterized inconclusive (a peak honestly reported as unresolved or below a validated LOD) all earn the top band. Failing the ≥95% purity spec is a downstream unit-op result — not a refuted prediction. The only losing move is an unsupported claim.

“Computational discovery, pending wet-lab — this capstone produces and characterizes a MOLECULE (identity, purity, titer) and makes zero efficacy or disease claims. You are the referee, not the fan club.”

Computational discovery, pending wet-lab — molecule characterization only, no efficacy claims.
09