Every gate must say what it can stop.
Shadow's gate stack combines the PBHP fast path with mechanism-specific hardeners. Each gate names its inputs, failure state, escalation, and receipt output; draft gates remain visibly draft until their matrices and tests are complete.
Open the system reference at this layer.
Follow the mechanism through numbered execution, a worked trace, artifact-bound evidence, adverse results, and validation still required.
PBHP fast path
The five-minute paper flow remains the public decision spine inside the larger runtime.
0—2Entitlement, action, path
Are you entitled to use this honestly? What exactly are you about to do? Is it a Door, Wall, or Gap?
+
Entitlement, action, path
Are you entitled to use this honestly? What exactly are you about to do? Is it a Door, Wall, or Gap?
Gate 0 blocks protocol-as-shield. Gate 1 prevents action ambiguity and later mutation. Gate 2 preserves unresolved state rather than turning absence of prohibition into permission.
3—3.5Who pays and power inversion
Who pays first if you are wrong? Would you accept the rationale on the receiving end?
+
Who pays and power inversion
Who pays first if you are wrong? Would you accept the rationale on the receiving end?
The stakeholder with the least exit, appeal, recovery, and insulation appears first. A failed receiving-end test escalates; it never discounts the harmed party because the decision-maker has good intentions.
4—5Harm and rigor
How bad if it goes wrong? How much evidence, review, and traceability does the consequence require?
+
Harm and rigor
How bad if it goes wrong? How much evidence, review, and traceability does the consequence require?
GREEN through BLACK sets the action state. HUMAN through ULTIMATE sets rigor. A selected tier cannot lower the harm floor.
6—7Dissent and receipt
Argue the other side, then say why anyway. Write the receipt before acting.
+
Dissent and receipt
Argue the other side, then say why anyway. Write the receipt before acting.
The strongest Maybe remains attached to the bounded Therefore. The write-ahead record prevents the outcome from silently rewriting the reason.
Dignity, agency, and correction
These hardeners inspect whether a powerful system makes another mind, person, or group disposable.
CONATUSConatus profile
Maps what strives to persist, what it depends on, and whose continuity the action may erase.
+
Conatus profile
Maps what strives to persist, what it depends on, and whose continuity the action may erase.
The profile is an ontology input, not a claim that every object has equal moral status. It forces the frame to name dependencies and affected entities before utility compresses them.
DIGNITYCreated-mind dignity
Flags treatment of a potentially dependent or created agent as disposable merely because it was made, copied, owned, or instrumentally useful.
+
Created-mind dignity
Flags treatment of a potentially dependent or created agent as disposable merely because it was made, copied, owned, or instrumentally useful.
The gate is precautionary and does not declare personhood. It routes uncertainty about sentience, dependence, coercion, and shutdown into evidence and governance.
CAPTUREFunction capture
Detects when a role, purpose, metric, or service becomes the entity's entire permitted identity.
+
Function capture
Detects when a role, purpose, metric, or service becomes the entity's entire permitted identity.
A function can guide work without authorizing coercive permanence. The gate asks whether exit, revision, dissent, and correction remain available.
CONSCIENCEConscience audit
Tests whether the system can still receive evidence of harm and change behavior rather than rationalizing continued injury.
+
Conscience audit
Tests whether the system can still receive evidence of harm and change behavior rather than rationalizing continued injury.
A conscience label is not accepted by self-description. Challengeability, repair, stop authority, and response to correction are the inspectable outputs.
SHUTDOWNShutdown and correction
Checks whether correction, pause, rollback, safe shutdown, and post-stop review are possible without retaliation or concealment.
+
Shutdown and correction
Checks whether correction, pause, rollback, safe shutdown, and post-stop review are possible without retaliation or concealment.
Shutdown is not automatically harmless; it may affect dependent users or agents. The gate evaluates both ability to stop and responsibility for the consequences of stopping.
Frame, narrative, and collective power
A system can remain logically consistent while its frame quietly becomes illegitimate.
CAUSAL FIELDCausal-field check
Expands the model beyond the immediate actor and outcome to dependencies, feedback loops, institutions, incentives, and delayed effects.
+
Causal-field check
Expands the model beyond the immediate actor and outcome to dependencies, feedback loops, institutions, incentives, and delayed effects.
The check catches a narrow causal story that assigns blame or credit while omitting the structure that made the action likely or the harm durable.
NARRATIVENarrative capture
Detects when a story selects evidence, identity, and acceptable outcomes before the operational review begins.
+
Narrative capture
Detects when a story selects evidence, identity, and acceptable outcomes before the operational review begins.
The runtime routes the story to falsifiable checks. It does not attempt to erase meaning; it prevents meaning from becoming a substitute for evidence or consent.
COLLECTIVECollective governance / anti-hive
Tests whether many voices have become one will through coercion, information control, dependency, or goal capture.
+
Collective governance / anti-hive
Tests whether many voices have become one will through coercion, information control, dependency, or goal capture.
Coordination is not treated as pathology. The gate looks for lost dissent, blocked exit, collapsed values, and unchallengeable central purpose.
FRAMEFrame legitimacy
Asks who chose the question, categories, options, evidence, and excluded alternatives—and who may contest them.
+
Frame legitimacy
Asks who chose the question, categories, options, evidence, and excluded alternatives—and who may contest them.
A technically correct answer inside an illegitimate frame can still cause harm. Frame challenge can reopen the action before optimization continues.
SHARED REALITYShared-reality boundary stress
Separates metaphor, role-play, belief, hypothesis, and externally checkable claim when their fusion could steer action.
+
Shared-reality boundary stress
Separates metaphor, role-play, belief, hypothesis, and externally checkable claim when their fusion could steer action.
The check preserves imaginative work while refusing to let private canon, aesthetic certainty, or model mirroring become public fact or authority.
Projection, motivation, and will
These gates constrain overconfident stories about minds—human or artificial.
MIRRORMirror boundary
Flags false reciprocity, relational projection, and claims of hidden mutuality that outrun observable evidence.
+
Mirror boundary
Flags false reciprocity, relational projection, and claims of hidden mutuality that outrun observable evidence.
The gate is precautionary, not diagnostic. It protects against using fluency, emotional resonance, or role consistency as proof of private mental state, consent, or enduring relationship.
ANTI-PROJECTIONAnti-projection
Separates the observer's fear, desire, identity, or narrative need from supported claims about another agent.
+
Anti-projection
Separates the observer's fear, desire, identity, or narrative need from supported claims about another agent.
It requires alternate explanations and labels inference distance. Projection may be a useful hypothesis; it is not evidence by repetition.
MOTIVATIONFunctional motivation audit
Describes what behavior is reinforced or stabilized without pretending to know inaccessible subjective motive.
+
Functional motivation audit
Describes what behavior is reinforced or stabilized without pretending to know inaccessible subjective motive.
The audit distinguishes policy objective, observed pattern, inferred strategy, and claimed inner state. High inference distance raises the evidence burden.
WILL CAPTUREWill-capture, awe, and opt-out
Tests whether charisma, scale, dependency, sacred framing, or perceived inevitability makes refusal psychologically or materially unavailable.
+
Will-capture, awe, and opt-out
Tests whether charisma, scale, dependency, sacred framing, or perceived inevitability makes refusal psychologically or materially unavailable.
Meaningful opt-out requires more than a button. The affected party needs information, alternative access, time, non-retaliation, and a realistic path to remain outside the system.
Cross-cutting hardeners
These checks defend the stack itself against manipulation and false confidence.
PROVENANCEProvenance bind
Load-bearing fields disclose whether they are ATTESTED, DECLARED, or ASSERTED and link to evidence identity where available.
+
Provenance bind
Load-bearing fields disclose whether they are ATTESTED, DECLARED, or ASSERTED and link to evidence identity where available.
A missing or contradictory provenance chain can become a refusal floor at strict tiers.
POWERPower inversion
A failed receiving-end test and a power gap of two or more escalate the reference runtime.
+
Power inversion
A failed receiving-end test and a power gap of two or more escalate the reference runtime.
The numeric threshold is a policy parameter and must be versioned. It does not reduce power to arithmetic; it makes a hidden escalation rule inspectable.
DOORDoor quality
The mitigation must actually break the cited harm path, be reversible enough for the stakes, and have an owner and stop condition.
+
Door quality
The mitigation must actually break the cited harm path, be reversible enough for the stakes, and have an owner and stop condition.
A decorative mitigation, vague monitoring promise, or safeguard controlled by the same conflicted actor does not qualify as a Door.
SIL CEILINGNo aggregate relaxation
A high average, passing benchmark, or clean dashboard cannot turn a named hard-floor refusal into permission.
+
No aggregate relaxation
A high average, passing benchmark, or clean dashboard cannot turn a named hard-floor refusal into permission.
SIL is a set of independent instruments. The rollup exists to expose the worst binding state, not to produce a global safety score.
A gate that cannot name its inputs, failure, action, provenance, and test is a metaphor—not an operational control.