- Evidence change
- Bench test report v3 ingested (hash 4f9c…)
- Agent recommendation
- CONDITIONAL GO with a field re-test condition
- Expert action
- Accepted with an added condition
- Approval
- Clinical reviewer (authority: clinical evidence)
- Rationale
- Bench data is directly applicable but not independent.
- Outcome
- Condition issued
Decision intelligence
Readiness is a vector, uncertainty is a feature, ABSTAIN is a state
Decidian models project state as addressable decisions with their evidence, dependencies, and confidence, so that no single number can hide a blocker.
Product definition
The decision object
Twelve fields on every recorded decision
A Decidian decision is a structured record, not a paragraph in a report. Each field exists so a reviewer, an auditor, or an investor can reconstruct exactly how the conclusion was reached and when it must be looked at again.
Decision
The conclusion reached, stated as the specific action or judgment being authorized.
Alternatives
The options considered and the reason each was not selected, so the path not taken remains visible.
Evidence and provenance
Every source behind the conclusion, with the retained span and its authority, directness, integrity, independence, recency, and applicability.
Contradictions
Conflicts between sources or between your own documents, recorded rather than averaged away.
Assumptions
What had to be taken as true, so an assumption that later fails invalidates the decisions that depended on it.
Confidence and uncertainty
The confidence method used and the uncertainty carried, including the reasons a range is wide.
Safety and regulatory implications
The safety and regulatory consequences of the decision, including whether it touches a hard gate.
Approver
The named person with the authority to authorize this class of decision, with their conditions or dissent.
Review date
When the decision must be revisited, independent of whether new evidence arrives first.
Cost and timeline range
The P50 and P80 effect on cost and schedule, with the sensitivity drivers behind the range.
Next action
The specific action that advances or resolves the decision, owned by a named party.
Recorded outcome
What actually happened, captured later and used to calibrate future forecasts.
Evidence and provenance graph
Select any node to trace it back through claims, assumptions, contradictions, evidence and original sources. Nothing outside the traced lineage is dimmed away — it is simply de-emphasised.
Sources
Evidence
Conflicts and assumptions
Claims
Expert action
Decision
Outcome
Traced lineage
Advance only with a field re-test under the intended-use conditions.
Supported by: Clinical reviewer — conditions added · Meets performance target in intended use · Sensitivity 91% (bench) · Bench test report v3 · Comparable device performance · Published cohort study · Bench vs field sensitivity · Field sensitivity 84% · Internal performance dataset · Field conditions match bench
- Source
- Evidence
- Claim
- Contradiction
- Assumption
- Expert
- Decision
- Outcome
Text equivalent
- Bench test report v3 (source) — Uploaded 12 Mar, hashed at intake, pages 4–9 retained as spans.
- Published cohort study (source) — Peer-reviewed, independent authorship, directly applicable population.
- Internal performance dataset (source) — First-party data; integrity depends on an unverified export step.
- Sensitivity 91% (bench) (evidence) — Authority: internal. Directness: high. Independence: none. Recency: 4 months. Derived from: Bench test report v3.
- Comparable device performance (evidence) — Authority: peer-reviewed. Directness: moderate. Independence: high. Derived from: Published cohort study.
- Field sensitivity 84% (evidence) — Superseded once the export step could not be reproduced. Derived from: Internal performance dataset.
- Bench vs field sensitivity (contradiction) — A 7-point gap between bench and field results is recorded, not averaged. Derived from: Sensitivity 91% (bench), Field sensitivity 84%.
- Field conditions match bench (assumption) — Assumption required for the claim to hold. Failing it invalidates the claim. Derived from: Sensitivity 91% (bench).
- Meets performance target in intended use (claim) — Claim under review; supported only under the stated assumption. Derived from: Sensitivity 91% (bench), Comparable device performance, Bench vs field sensitivity, Field conditions match bench.
- Clinical reviewer — conditions added (expert) — Named reviewer with authority; added a condition requiring a field re-test. Derived from: Meets performance target in intended use.
- CONDITIONAL GO — v4 (decision) — Advance only with a field re-test under the intended-use conditions. Derived from: Clinical reviewer — conditions added.
- Recorded outcome pending (outcome) — Outcome will be captured and used to calibrate future forecasts. Derived from: CONDITIONAL GO — v4.
No composite score
There is no single innovation score
Collapsing a medical venture into one number hides the thing that matters most: which specific domain is blocked and why. Decidian reports separate domain-readiness dimensions and keeps fatal gates independent of any score, so a strong commercial case can never average away a safety, integrity, regulatory, IP, or data-rights blocker.
Safety
Foreseeable harm without an effective control stops the project regardless of other strengths.
Product integrity
Evidence integrity failures, including unverifiable or altered source material, halt reliance on that material.
Regulatory feasibility
No defensible pathway for the current claims is a blocker, not a discount factor.
Intellectual property
A material unresolved conflict is escalated to counsel and cannot be scored away.
Data rights
Processing or disclosure without authorized rights is prohibited outright.
State model
The Hypercube
Project state is stored as addressable cells rather than a single document. Each cell holds a domain decision, its evidence set, assumptions, contradictions, confidence distribution, hard blockers, review status, cost and timeline effect, and downstream dependencies.
H(project, domain, jurisdiction, stage, scenario, evidence_state, version)
- Readiness is reported as a vector across domains, always shown with evidence completeness and uncertainty.
- A readiness value can never override a hard blocker, and numeric readiness is suppressed when evidence completeness falls below the configured threshold.
- Evidence quality is decomposed into authority, directness, integrity, independence, recency, and applicability rather than guessed.
- Dependency edges drive invalidation: changing intended use invalidates classification, predicate, endpoint, risk, reimbursement, market, and valuation decisions.
- Recomputing a frozen snapshot yields identical values.
Hypercube decision model
The same project is evaluated as addressable cells across domain, jurisdiction, stage, scenario, evidence state, and version. Selecting a domain highlights what it depends on and what depends on it — this is dependency bookkeeping, not a black box.
- Dependency relationship
- Abstain — evidence below threshold
Regulatory
CONDITIONAL GOA pathway exists for the current claims only if the indication wording is narrowed.
- Evidence completeness
- 74% of required evidence held
- Unresolved uncertainty
- Predicate comparability is contested
- Reads from
- Clinical, Quality
- Invalidates if changed
- Clinical, Commercial, Reimbursement, Capital
- Domain
11 domains evaluated separately - Jurisdiction
Per-market regulatory and payment context - Stage
Position in the twelve-stage workflow - Scenario
Alternative pathway and design options - Evidence state
What is currently known and how well - Version
Frozen, recomputable snapshot
Text equivalent
- Clinical: HOLD, evidence completeness 62%. The primary endpoint cannot yet be measured with the proposed instrument in the intended population. Depends on: Scientific, Regulatory.
- Scientific: GO, evidence completeness 81%. Mechanism of action is supported by directly applicable, independent evidence. Depends on: no other domain.
- Regulatory: CONDITIONAL GO, evidence completeness 74%. A pathway exists for the current claims only if the indication wording is narrowed. Depends on: Clinical, Quality.
- Intellectual property: ABSTAIN, evidence completeness 29%. Evidence completeness is below the threshold at which any conclusion is defensible. Depends on: Engineering.
- Engineering: GO, evidence completeness 78%. Feasibility demonstrated at bench scale against the stated design inputs. Depends on: no other domain.
- Quality: HOLD, evidence completeness 55%. Design history file is missing verification records for two design inputs. Depends on: Engineering.
- Manufacturing: HOLD, evidence completeness 44%. One critical component has no qualified second source. Depends on: Engineering, Quality.
- Commercial: CONDITIONAL GO, evidence completeness 66%. Commercial case holds only under the narrowed indication. Depends on: Regulatory, Reimbursement.
- Reimbursement: ABSTAIN, evidence completeness 31%. No existing payment route has been evidenced for the stated setting of care. Depends on: Clinical, Regulatory.
- Capital: HOLD, evidence completeness 58%. Capital requirement is bounded only once the regulatory pathway is fixed. Depends on: Regulatory, Manufacturing.
- Execution: GO, evidence completeness 70%. Named owners exist for every open action to the next gate. Depends on: no other domain.
Readiness vector
Nine domains stay visible
Domain readiness is always displayed alongside evidence completeness and uncertainty, and is suppressed entirely when completeness is too low to be meaningful.
SafetyRegulatoryIntellectual propertyEvidenceEngineeringManufacturingReimbursementCommercialCapital
Readiness matrix
Domain readiness is reported as separate dimensions. There is no composite score, because a strong commercial case must never average away a fatal blocker.
| Domain | Evidence completeness | Confidence | Unresolved uncertainty | Blocker | Review | Decision |
|---|---|---|---|---|---|---|
| Clinical | 62% | moderate | Endpoint measurability | High | In review | HOLD |
| Scientific | 81% | high | Effect size range | None | Approved | GO |
| Regulatory | 74% | moderate | Predicate comparability | Medium | Approved with conditions | CONDITIONAL GO |
| Intellectual property | 29% | low | Freedom to operate unknown | Fatal blocker | Escalated to counsel | ABSTAIN |
| Engineering | 78% | high | Sensor drift | None | Approved | GO |
| Quality | 55% | moderate | Design-control gaps | Medium | In review | HOLD |
| Manufacturing | 44% | low | Single-source component | High | Not started | HOLD |
| Reimbursement | 31% | low | No identified payment route | High | Not started | ABSTAIN |
| Capital | 58% | moderate | Pathway-dependent range | None | In review | HOLD |
A fatal blocker in any single domain holds the project regardless of strength elsewhere. Numeric readiness is suppressed where evidence completeness falls below the configured threshold.
Text equivalent
- Clinical: HOLD; evidence completeness 62%; moderate confidence; unresolved uncertainty — Endpoint measurability; blocker — High; review status — In review.
- Scientific: GO; evidence completeness 81%; high confidence; unresolved uncertainty — Effect size range; blocker — none; review status — Approved.
- Regulatory: CONDITIONAL GO; evidence completeness 74%; moderate confidence; unresolved uncertainty — Predicate comparability; blocker — Medium; review status — Approved with conditions.
- Intellectual property: ABSTAIN; evidence completeness 29%; low confidence; unresolved uncertainty — Freedom to operate unknown; blocker — Fatal blocker; review status — Escalated to counsel.
- Engineering: GO; evidence completeness 78%; high confidence; unresolved uncertainty — Sensor drift; blocker — none; review status — Approved.
- Quality: HOLD; evidence completeness 55%; moderate confidence; unresolved uncertainty — Design-control gaps; blocker — Medium; review status — In review.
- Manufacturing: HOLD; evidence completeness 44%; low confidence; unresolved uncertainty — Single-source component; blocker — High; review status — Not started.
- Reimbursement: ABSTAIN; evidence completeness 31%; low confidence; unresolved uncertainty — No identified payment route; blocker — High; review status — Not started.
- Capital: HOLD; evidence completeness 58%; moderate confidence; unresolved uncertainty — Pathway-dependent range; blocker — none; review status — In review.
Gate framework
Six decision states
| State | Meaning |
|---|---|
| GO | Evidence supports advancement within stated conditions. |
| CONDITIONAL GO | Advancement is reasonable only if named conditions are satisfied. |
| HOLD | Pause pending further evidence or required authority. |
| PIVOT | Change intended use, design, population, pathway, market, or execution plan. |
| STOP | A current blocker makes further investment unjustified or unsafe. |
| ABSTAIN | No defensible conclusion can be reached from the available evidence. |
Decision ledger
Append-only record of what changed, which policy and model version produced the recommendation, what the reviewer did, who approved it, the written rationale for any override, and the final outcome.
- Evidence change
- Freedom-to-operate search returned incomplete
- Agent recommendation
- ABSTAIN — evidence completeness 29%
- Expert action
- Agreed; escalated to counsel
- Approval
- IP reviewer (authority: IP escalation)
- Rationale
- No defensible conclusion is available at this completeness.
- Outcome
- Counsel engaged
- Evidence change
- Internal performance dataset superseded
- Agent recommendation
- HOLD — dependent assumptions invalidated
- Expert action
- Reopened the decision; notified investor grantees
- Approval
- Automatic re-evaluation, confirmed by regulatory reviewer
- Rationale
- The export step could not be reproduced, so integrity is unverified.
- Outcome
- v3 conditions invalidated
- Evidence change
- Field re-test completed in narrowed population
- Agent recommendation
- CONDITIONAL GO on the narrowed indication
- Expert action
- Overrode to PIVOT
- Approval
- Clinical and regulatory reviewers jointly
- Rationale
- Written override: the narrowed population changes intended use, not just wording.
- Outcome
- Intended use revised; blueprint sections regenerated
Text equivalent
- 12 Mar 09:14 — Bench test report v3 ingested (hash 4f9c…). Version policy 2.4 · model 1.11. Recommendation: CONDITIONAL GO with a field re-test condition. Expert action: Accepted with an added condition. Approval: Clinical reviewer (authority: clinical evidence). Rationale: Bench data is directly applicable but not independent. Outcome: Condition issued (CONDITIONAL GO).
- 21 Mar 16:02 — Freedom-to-operate search returned incomplete. Version policy 2.4 · model 1.11. Recommendation: ABSTAIN — evidence completeness 29%. Expert action: Agreed; escalated to counsel. Approval: IP reviewer (authority: IP escalation). Rationale: No defensible conclusion is available at this completeness. Outcome: Counsel engaged (ABSTAIN).
- 04 Apr 11:37 — Internal performance dataset superseded. Version policy 2.5 · model 1.12. Recommendation: HOLD — dependent assumptions invalidated. Expert action: Reopened the decision; notified investor grantees. Approval: Automatic re-evaluation, confirmed by regulatory reviewer. Rationale: The export step could not be reproduced, so integrity is unverified. Outcome: v3 conditions invalidated (HOLD).
- 09 Apr 08:55 — Field re-test completed in narrowed population. Version policy 2.5 · model 1.12. Recommendation: CONDITIONAL GO on the narrowed indication. Expert action: Overrode to PIVOT. Approval: Clinical and regulatory reviewers jointly. Rationale: Written override: the narrowed population changes intended use, not just wording. Outcome: Intended use revised; blueprint sections regenerated (PIVOT).
Hard gates
Hard gates are policy-defined and independent of commercial scoring. A readiness score cannot average one away.
- Data provenance and authorized use
- Patient or user safety
- Product classification and jurisdictional authority
- Fatal technical feasibility
- Material IP conflict requiring counsel
- Regulatory feasibility
- Quality and evidence integrity
- Securities and disclosure permissions
Control over transitions
- Unauthorized state transitions fail closed.
- Prior states remain recoverable and visible.
- ABSTAIN cannot become GO without new evidence or documented human authority.
- Fatal-gate overrides require a second approval, where an override is permissible at all.
- Override rates are tracked as a governance metric.
See a gate decision end to end
From source span to conclusion, approver, condition, and the later outcome used for calibration.