Outcome label and boundary
This outcome reveal opens only after the learner packet was frozen at canonical SHA-256 babd925b6b6cf708c8a961ba2fcc3905cf3a10fb806c987ab5b6c1e300267fca. The learner recommendation was to run the decisive pivotal test while staging later irreversible manufacturing and commercial commitments behind efficacy, quality, regulatory, timing, and contract-exposure gates [judgment.curevac.pivotal-with-staged-scale].
The outcome is a failure of first-generation CVnCoV regulatory and commercial scale. It is not labeled a failure of HERALD's final statistical criterion, CureVac as a company, its later candidates, or the mRNA platform [judgment.curevac.failure-boundary]. This boundary matters because a trial can clear its prespecified statistical rule while the product still misses the quality, benefit-risk, timing, or commercial gates needed for the original scale plan [conflict.curevac.statistical-versus-commercial-success].
The December 10, 2020 cutoff was a conservative public-information boundary, not a verified internal decision timestamp. A later primary paper reported an enrollment and randomization interval beginning December 11; the timeline's 2020-12-11T00:00:00Z value is only a normalized date boundary, not a precise first-participant time [claim.curevac.paper-enrollment-chronology; evidence.curevac.paper-final-results; @src.curevac.herald-paper.2021-11-23]. On December 14, CureVac announced that it had enrolled the first participant [claim.curevac.launch-announcement-chronology; evidence.curevac.launch-announcement; @src.curevac.herald-launch.2020-12-14]. Those descriptions are not silently reconciled [conflict.curevac.enrollment-chronology].
A ClinicalTrials.gov historical-version response observed in 2026 records an original NOT_YET_RECRUITING status, an estimated November 30 start, and an actual registry first-post field of December 3 [claim.curevac.registry-version0-record; evidence.curevac.registry-original-status; evidence.curevac.registry-estimated-start; evidence.curevac.registry-first-post; @src.curevac.ctgov-history0.observed-2026-08-16]. The captured response also contains an appended later publication reference, so only its exact enumerated original protocol fields are used in Part B. It is not treated as evidence that was available to the learner at the cutoff.
Interim disappointment was not final statistical failure
On June 16, 2021, CureVac reported 47% interim efficacy and said the second interim analysis had not met the prespecified statistical success criteria. The same disclosure said the study would continue to final analysis [claim.curevac.interim-missed-criterion; evidence.curevac.interim-statistical-result; @src.curevac.herald-interim.2021-06-16]. The phrase “did not meet” belongs to that interim analysis and must not be carried forward as the final result.
The primary HERALD paper later reported 39,680 enrolled participants and a final overall efficacy estimate of 48.2%, with a 95.826% confidence interval of 31.0% to 61.4% and p=0.016. The primary calculation used 83 vaccine-arm cases over 1,735.29 person-years and 145 placebo-arm cases over 1,569.87 person-years [fact.curevac.herald.enrolled; claim.curevac.final-primary-result; evidence.curevac.paper-final-results; @src.curevac.herald-paper.2021-11-23].
The prespecified rule declared success when the lower confidence limit was greater than 30% [fact.curevac.herald.success-threshold-lower-bound; evidence.curevac.paper-success-threshold]. The reported 31.0% lower bound was therefore 1.0 percentage point above that threshold, so the final primary endpoint met the stated statistical criterion [claim.curevac.final-met-prespecified-statistical-criterion; table.curevac.herald-final-check].
The deterministic check recomputes overall efficacy as 48.215270% from the reported cases and person-time and moderate-to-severe efficacy as 70.659254%, consistent with the paper's rounded 48.2% and 70.7% figures. It also calculates the 1.0-point lower-bound margin [model-run.curevac.outcome-checks; claim.curevac.moderate-severe-result]. This run is explicitly non-authoritative and unvalidated because it has no Warren model-registry execution receipt; the primary publication and filing figures remain authoritative.
The clinical record still contained important limits. Too few cases occurred among participants aged 61 or older for a meaningful efficacy assessment in that subgroup [claim.curevac.older-subgroup-indeterminate; evidence.curevac.paper-older-subgroup-limit]. The paper described a variant-dominated setting in which only 7 of 204 sequenced cases were from the B.1 lineage and the remainder represented 14 variants [claim.curevac.variant-dominated-setting; evidence.curevac.paper-variant-environment]. Investigators also reported that low older-adult recruitment, substantial unmasking, and the broad case definition limited interpretation or affected results [claim.curevac.trial-design-confounds; evidence.curevac.paper-trial-confounds]. These observations are confounds, not a standardized cross-vaccine comparison or a causal estimate of what a different dose, formulation, strain, or trial design would have achieved.
Regulatory and commercial resolution
On October 12, 2021, CureVac announced that it would withdraw first-generation CVnCoV from the EMA approval process, estimated that the earliest potential approval would have been in the second quarter of 2022, shifted its focus to second-generation candidates, and said the acute-use European advance-purchase agreement would cease [claim.curevac.withdrawal-and-apa-end; evidence.curevac.first-generation-withdrawal; evidence.curevac.apa-ceased; @src.curevac.withdrawal.2021-10-12].
EMA did not issue a final rejection. Its notice said the rolling review ended after sponsor withdrawal and that questions about vaccine quality affecting benefit-risk, together with modest efficacy in adults, still needed satisfactory answers [claim.curevac.ema-review-ended-with-open-questions; evidence.curevac.ema-open-questions; @src.curevac.ema-withdrawal.2021-10-12]. The narrow failure label therefore rests on withdrawal, unresolved regulatory questions, lost procurement timing, and cessation of the acute-use contract—not on inventing an EMA denial or rewriting the final endpoint as statistically unsuccessful.
Financial consequences and accounting separation
CureVac's audited 2021 filing reported EUR 753.627 million of CVnCoV program R&D expense, compared with EUR 52.701 million in 2020 [fact.curevac.cvncov-rd.fy2021; fact.curevac.cvncov-rd.fy2020; claim.curevac.cvncov-rd-and-cmo-cost; evidence.curevac.cvncov-rd-2020-2021; @src.curevac.20f.2021]. The deterministic bridge is an increase of EUR 700.926 million, or 1,330.005123% of the 2020 base [table.curevac.cvncov-rd-levels; table.curevac.cvncov-rd-bridge; model-run.curevac.outcome-checks]. Neither calculation identifies how much was avoidable.
The filing also recognized an estimated EUR 81.6 million provision at December 31, 2021 for terminating contract-manufacturing-organization agreements [fact.curevac.cmo-termination-provision.2021-12-31; evidence.curevac.cmo-termination-provision]. That provision is an uncertain settlement estimate, not a complete inventory of scale commitments or a proven counterfactual saving.
At the company level, the audited statements reported EUR 102.990 million of 2021 revenue, EUR 412.263 million of operating loss, EUR 411.716 million of net loss, EUR 733.128 million of operating cash use, and EUR 811.464 million of cash at year-end [claim.curevac.fy2021-company-financials; table.curevac.fy2021-financial-consequences; table.curevac.year-end-2021-financial-position]. Separate tables preserve FY2021 duration flows and December 31 balances and provisions as different coordinates.
One accounting line requires special care. The statement separately recognized EUR 574.502 million as income from release of governmental contract liabilities [fact.curevac.government-liability-release-income.fy2021; claim.curevac.government-release-reported; evidence.curevac.audited-government-release-income]. It is not classified here as product sales or a same-period cash receipt [judgment.curevac.government-release-classification]. The report does not manufacture a full decomposition of that amount from partial selected evidence.
Separately, CureVac reported receiving EUR 196.3 million of BMBF funding cumulatively from 2020 through December 2021, failing to reach all 2021 milestones after withdrawal, and automatically terminating the related supply agreement after notifying the government that it could not supply [fact.curevac.bmbf-received.through-2021; claim.curevac.bmbf-outcome; evidence.curevac.bmbf-outcome; table.curevac.bmbf-cumulative-support]. The cumulative 2020–2021 receipt occupies its own table coordinate; it is not presented as 2021 revenue or a 2021-only cash flow. This outcome supports retaining the learner packet's separation of government funding, advance-purchase support, cash, contract liabilities, accounting releases, and delivered-product sales.
Causal assessment
The primary causal hypothesis is a validation-timing-scale mismatch. Phase 1 immune signals justified decisive testing but did not establish protective efficacy. The final primary endpoint cleared its statistical rule, yet modest overall efficacy, older-subgroup uncertainty, unresolved quality questions, and projected authorization timing did not sustain the original acute procurement path. Withdrawal then left high program R&D and termination exposure after much of the learning and scale cost had accumulated [hypothesis.curevac.validation-timing-scale-mismatch]. Confidence is moderate because public evidence does not disclose the full sequence, reversibility, or economics of each manufacturing contract.
A rival and complementary hypothesis emphasizes the variant environment and trial-conduct constraints [hypothesis.curevac.variant-and-trial-environment]. The evidence supports treating those factors as contributors worth investigating [assumption.curevac.variant-causal-contribution]. It does not identify their causal share, exclude product performance or quality, or establish the efficacy of a counterfactual dose or formulation [conflict.curevac.causal-attribution; judgment.curevac.causality-caution].
This distinction protects process evaluation from hindsight. Proceeding to HERALD was evidence-responsive because Phase 1 could not answer the protection question. The later record supports tighter staging of first-generation scale exposure, but it does not prove that every commitment was avoidable or that delay was costless [judgment.curevac.process-quality]. A pandemic capacity option can have real value even when it expires unused.
Counterfactual and reusable rules
The decision-useful counterfactual is to run HERALD while conditioning later CMO, manufacturing, and commercial tranches on predeclared efficacy, quality, approval-timing, and contract-exposure gates [counterfactual.curevac.stage-first-generation-scale]. Its feasibility is an explicit assumption: at least some commitments could have been deferred or conditioned without destroying the pivotal trial's information value [assumption.curevac.counterfactual-staging-reduces-exposure]. Public sources do not disclose each contract's signature date, cancellation rights, minimum commitments, lead-time economics, or opportunity cost, so no savings estimate is produced.
Two candidate rules follow.
First, when decisive product validation remains outstanding and scale commitments become costly to reverse, run the validation while staging later irreversible commitments behind observable clinical, quality, regulatory, timing, and termination-exposure gates [rule.curevac.stage-scale-behind-pivotal-gates]. Re-underwrite before exposure exceeds the approved downside budget, but allow an exception when the demonstrated option cost of delay exceeds the downside protected by staging.
Second, model each government cash receipt, contract liability, accounting release, procurement option, and delivered-dose sale as a separate state-contingent item [rule.curevac.separate-conditional-support-from-sales]. Conditional support can be economically valuable without being recurring product demand or product revenue.
Both rules remain candidates. They originate from this one case, have no registered counterexample cases, and should not be promoted to validated general rules until cross-case tests supply both support and failure conditions.
Final underwriting boundary
This episode teaches that “the trial succeeded” and “the scale decision succeeded” are different propositions. HERALD's final endpoint met its prespecified statistical rule; first-generation CVnCoV nevertheless failed to secure the intended regulatory and acute-use commercial outcome after material program spending. The strongest learning is to buy decisive evidence while explicitly pricing the irreversibility, timing, and accounting states attached to scale.
Capitalization, dilution, market-price, contract-level cash-flow, and comparable-valuation inputs remain incomplete. The Part A valuation abstention therefore remains in force: no target price, market-value attribution, or program return is fabricated [judgment.curevac.valuation-abstention; table.curevac.valuation-abstention].