{"schema_version":1,"experiment_id":"eoa_inverse_innovation_exp12_substrate_denial72_20260805","cell_id":"authority_mentor_relationship_anchoring__computer_science","arm":"ORDINARY_MAX","candidate_id":"authority_mentor_relationship_anchoring__computer_science__ORDINARY_MAX","proposal_index":1,"version":0,"title":"Incident Stewardship Anchor for New On-Call Commanders","problem":"In a distributed-software team, engineers preparing to command production incidents may pass technical and runbook checks yet fail to enact shared stewardship norms when telemetry conflicts and speed, reversibility, customer harm, data integrity, and escalation compete. Documents can name procedures without transmitting which obligations take precedence, how uncertainty should be disclosed, or when a respected responder deliberately stops rather than acts. The target problem is non-internalized professional judgment, not inability to debug software.","actors":["New or prospective on-call incident commanders","Vetted veteran incident commanders serving as mentors","Incident-response program owner","Service owner or engineering manager","Security and privacy reviewer controlling incident artifacts","Independent secondary mentor or ombudsperson","Service users exposed to incident consequences"],"observable_state":"Across controlled ambiguous incident replays or deidentified postmortem timelines, a novice can identify technically possible actions but repeatedly chooses a high-blast-radius action without stating a hypothesis or rollback, conceals uncertainty, delays escalation past a predefined boundary, or cannot distinguish official policy and stewardship invariants from a veteran's personal style. Decisions, timing, stated rationales, challenges, and escalation requests are directly recordable.","consequence":"If transferred to live incidents, these patterns can enlarge a failure, delay containment, obscure risk from collaborators, or leave the team dependent on a few veteran commanders. Conversely, treating every ambiguity as a reason to defer can prevent novices from ever assuming independent command.","affected_objective":"Safe and timely service restoration by incident commanders who can exercise independent, reviewable judgment under uncertainty without increasing production risk or dependence on one expert.","intervention":"Offer each volunteer novice a six-session Incident Stewardship Anchor with one vetted veteran who is respected for incident conduct but is outside the novice's performance-reporting line. A written relationship charter defines consent, confidentiality, reassignment, complaint routes, and an explicit payload: minimize customer harm, prefer reversible moves, state evidence and uncertainty, preserve truthful records, escalate at agreed boundaries, and separate learning from blame. In approved reconstructed incidents, the novice first observes the mentor make and explain decisions, then debriefs what was invariant, policy-specific, personal, or contested. Roles progressively reverse so the novice proposes and leads simulated actions while the mentor gives corrective feedback and fades prompts. A second veteran reviews selected cases and disagreements. Completion does not itself grant production access or incident-command authority.","structural_mapping":[{"archetype_element":"legitimate_mentor_anchor","domain_realization":"A veteran incident commander selected for demonstrated technical competence, care under pressure, ethical conduct, explanatory ability, and peer-recognized stewardship rather than rank or charisma alone."},{"archetype_element":"mentee_readiness_and_consent_boundary","domain_realization":"Participants first demonstrate basic service and runbook knowledge, enter voluntarily, receive the influence charter, and may decline or request reassignment without effects on employment evaluation or on-call eligibility."},{"archetype_element":"relational_safety_container","domain_realization":"The mentor is not the novice's manager; debriefs normalize uncertainty and respectful disagreement, use approved artifacts, and have predefined confidentiality exceptions for safety or misconduct."},{"archetype_element":"cultural_norm_and_value_payload","domain_realization":"The payload is explicitly limited to production-stewardship norms: harm minimization, reversibility, evidence discipline, uncertainty disclosure, timely escalation, truthful coordination, and blameless learning."},{"archetype_element":"modeled_practice_and_judgment_window","domain_realization":"Reconstructed incident timelines pause at authentic decision points so the novice can observe how the mentor interprets weak signals, competing obligations, emotional pressure, rollback options, and escalation thresholds."},{"archetype_element":"dialogic_interpretation_loop","domain_realization":"After each pivotal decision, mentor and novice identify the reason, rejected alternatives, exceptions, and whether the guidance is an invariant, written policy, local convention, personal style, or contested judgment."},{"archetype_element":"mentor_selection_and_matching_criteria","domain_realization":"The incident-program owner screens mentors for conduct, facilitation skill, service-context fit, availability, conflicts of interest, and willingness to have their guidance questioned and independently reviewed."},{"archetype_element":"secondary_reference_anchor","domain_realization":"A second veteran, the written incident policy, and the service's safety constraints provide comparison points when mentor advice may be idiosyncratic or inconsistent."},{"archetype_element":"progressive_autonomy_release","domain_realization":"The sequence moves from observation to proposing options, leading a reconstructed incident, and completing an unfamiliar scenario without prompts; any live authority remains governed by the existing qualification process."},{"archetype_element":"autonomy_and_exit_safeguard","domain_realization":"The novice may challenge guidance, use an independent complaint channel, change mentors, pause participation, or leave without a loyalty test, punitive identity loss, or automatic qualification penalty."}],"mechanism_mapping":[{"mechanism_slug":"guided_shadowing_with_debrief","role":"The mentor works through a reconstructed incident while exposing choices and tacit cues, followed immediately by a debrief connecting each choice to the bounded stewardship payload.","counterfactual_removal":"Without observation plus debrief, participants receive declared rules but cannot inspect how those rules are interpreted when evidence and obligations conflict; the intervention becomes ordinary instruction."},{"mechanism_slug":"joint_practice_with_corrective_feedback","role":"The novice leads later replays while the mentor corrects reasoning at reversible choice points, requests an articulated rationale, and gradually removes prompts.","counterfactual_removal":"Without joint practice, the novice may recognize or imitate the mentor's decisions without demonstrating that the norms can guide the novice's own action."},{"mechanism_slug":"reflective_norm_dialogue","role":"Each debrief requires the novice to question at least one decision and classify guidance as invariant, formal policy, local convention, mentor style, or contested practice.","counterfactual_removal":"Without explicit interpretation and permission to challenge, relational authority can produce compliant copying rather than reflective internalization."},{"mechanism_slug":"mentor_rotation_or_second_opinion_channel","role":"A second veteran reviews selected cases and disputed interpretations while the primary anchor remains stable.","counterfactual_removal":"Without a plural reference, improvement could be overfitting to one mentor, and biased or unsafe guidance could acquire undeserved authority."}],"causal_chain":["Ambiguous incidents require prioritizing stewardship obligations that cannot be reduced to selecting a technically valid command.","A stable, legitimate mentor repeatedly makes those priorities observable in realistic decision contexts and gives them credible professional meaning.","A bounded, non-evaluative relationship is hypothesized to lower the social cost of admitting confusion, uncertainty, or disagreement.","Debrief dialogue converts witnessed behavior into explicit distinctions among invariant norms, formal policy, local convention, and personal style.","Role-reversed practice lets the novice enact those distinctions and receive timely correction while choices remain simulated and reversible.","A secondary anchor challenges mentor-specific interpretations and prevents one relationship from becoming the whole professional culture.","Fading prompts exposes whether the novice can transfer the norms to an unfamiliar incident and can reject inappropriate mentor advice.","If that transfer reaches live work under existing authorization controls, more explicit uncertainty, reversible action, and timely escalation could support safer independent incident command."],"baseline":"Use an equal-resource comparison rather than a documentation-only straw baseline: participants receive the same stewardship charter, runbooks, reconstructed cases, session time, expert demonstrations, corrective feedback, and existing technical guardrails, but experts rotate between sessions and no durable mentor dyad is formed. Existing access controls, escalation rules, dual approvals, and incident-command policies remain active in both conditions.","nearest_rivals":["Rotating-expert cognitive apprenticeship using think-aloud demonstrations and deliberate practice without a stable authority relationship or identity-forming payload","Scenario-based incident-command certification with explicit rubrics, facilitator feedback, and repeated tabletop exercises","Executable runbooks, decision-support tooling, automated rollback, blast-radius limits, and dual authorization that constrain unsafe choices directly","A team-level community of practice in which peers review incidents collectively without a primary mentor anchor","Formal decision logs, postmortems, governance rules, and escalation matrices that make expectations more explicit"],"remaining_contrastive_claim":"The bounded claim to test is that, holding cases, explicit content, practice time, demonstrations, and feedback constant, a stable and legitimate mentor relationship contributes something distinctive to reflective internalization of production-stewardship norms in unfamiliar ambiguous incidents. The claim fails if rotating experts produce the same independent transfer, willingness to disclose uncertainty and challenge authority, and separation of norm from personal style, or if tooling and missing explicit knowledge explain the observed failures.","authority_safety":{"decision_authority":"The incident-response program owner and service owner may jointly authorize an offline training pilot; the security or privacy owner controls use of incident artifacts. During any live incident, authority remains exclusively with the commander designated under existing policy. A mentor gains no operational, access-control, employment, or policy authority by serving in the program.","authorized_first_step":"Authorize only a four-week offline mechanism probe using consenting volunteers, approved synthetic or deidentified replays, existing devices, and no connection to production systems. Mentors may pause a simulation and give feedback but may not issue production instructions or qualification decisions.","excluded_actions":["Granting, expanding, or revoking production permissions because of pilot participation or scores","Allowing participants to command or modify a live incident as part of the pilot","Using individual observations for compensation, promotion, discipline, performance review, or on-call qualification","Compelling participation, penalizing refusal, or pairing a novice with a direct evaluator","Sharing customer data, secrets, exploitable incident details, or unapproved recordings","Treating mentor guidance as an override of written policy, security controls, or the designated incident commander","Replacing runbooks, automated safeguards, independent oversight, or formal complaint channels with the mentor relationship"],"halt_rollback":"Immediately pause a pairing for reported fear, retaliation, confidentiality breach, mentor-policy conflict, pressure to obey, or inability to question. Offer reassignment or withdrawal, revoke access to pilot artifacts, delete or deidentify pilot-specific participant notes under the approved retention policy, investigate through the independent channel, and revert the participant to the ordinary training baseline. The offline design leaves no production state to roll back."},"negative_tests":{"strongest_counterevidence":"With cases, time, mentor quality, content, and feedback matched, novices taught by rotating experts transfer the stewardship norms to unfamiliar scenarios as well as those with stable mentors, while reporting equal or greater safety and agency. Equivalent counterevidence would be incident analysis showing that the target errors disappear when missing observability, authority, or guardrails are repaired.","problem_falsifier":"Before mentor exposure, blinded assessment shows novices already explain and apply the stewardship distinctions in unfamiliar cases after reading the explicit charter, or the observed failures trace primarily to missing technical information, overload, conflicting incentives, unclear decision rights, or defective tooling rather than a tacit enculturation gap.","intervention_falsifier":"Participants merely reproduce their mentor's preferred moves, cannot explain governing invariants, fail to transfer to a novel case, become less willing to question authority, or cannot safely exit. The intervention is also falsified for this setting if any apparent difference vanishes under the equal-content rotating-expert comparison.","risks":["Mentor bias, outdated practice, or personal style may be mistaken for professional norms","Authority and career dependence may make nominal consent or disagreement unsafe","A stable dyad may create favoritism, unequal access, dependency, or a new expert bottleneck","Participants may gain false confidence and seek production authority prematurely","Incident artifacts may expose customer information, secrets, vulnerabilities, or blame-sensitive history","Replaying stressful failures may cause distress or revive interpersonal conflict","Mentor attention, charisma, or facilitation quality may confound the relational mechanism","Simulated behavior may not transfer to time-pressured live incidents","Mentor workload may degrade ordinary incident readiness","Overemphasis on inherited culture may preserve exclusionary or technically obsolete practices"]},"next_evidence_step":"Run one bounded four-week offline probe with eight consenting novice commanders and four screened mentors. Randomly assign four novices to a stable mentor and four to a sequence drawn from the same mentor pool, changing mentor each week; keep four reconstructed cases, session duration, written content, demonstrations, and feedback format identical. Before exposure and after the fourth session, give all participants an unfamiliar incident replay. Blinded assessors use a predeclared rubric for hypothesis and uncertainty disclosure, reversibility and rollback, escalation timing, distinction among norm, policy, and style, and ability to challenge unsafe advice. Collect participant-reported safety, agency, and reassignment pressure. Stop at eight participants and four weeks, make no production-access decisions, and treat the result only as a problem-and-mechanism probe rather than an effectiveness estimate.","prior_art_status":"UNSEARCHED","diversity_from_prior_proposals":"Cross-proposal diversity was not assessed because no other candidates or experiments were inspected. Within the supplied inputs, this candidate selects the narrow substrate of ambiguous production-incident stewardship and isolates stable relational legitimacy from otherwise matched expert demonstration and feedback.","revision_record":{"parent_version":null,"progress_targets_addressed":["Initial complete candidate specification","Preservation of mentor legitimacy, explicit cultural payload, dialogic interpretation, plural reference, and autonomy release","Serious comparison against an equal-content rotating-expert baseline","Operational authority boundaries, safeguards, falsifiers, and a bounded evidence step"],"conceptual_changes":["Initial formulation selected tacit production-stewardship judgment in incident command rather than general programming mentorship or technical debugging instruction.","Defined reflective internalization through a stable, trusted, bounded relationship as the hypothesized active ingredient."],"operational_changes":["Specified a six-session mentor workflow with explicit payload, role reversal, secondary review, and progressive simulated autonomy.","Separated mentor influence from performance evaluation, production authority, access control, and formal qualification."],"evidence_changes":["Specified an offline randomized mechanism probe with matched content and mentor pool, an unfamiliar transfer case, blinded assessment, and agency measures.","Added distinct problem, intervention, and strongest-counterevidence tests."],"claim_changes":["Restricted the proposal to a falsifiable contrastive mechanism claim.","Made no claim about novelty, prevalence, demand, or effect size; prior art remains unsearched."]}}