Reference Implementation
The first attribution package ran while Evan was not in the building.
Nadia had scheduled it that way. The motor bench remained under company control, the neutral safety consortium held the package mapping, and Lena had signed the preregistration through the evidence channel. Evan received a subject-matter-witness credential that could read the protocol and nothing else.
Six days had passed since OLD-7 became ceramic fragments. No one proposed rebuilding it. The garage records supplied a historical comparator, not an executable condition.
The attribution question was narrower than Evan's model. It did not ask whether the universe had finite support or whether any apparent entity corresponded to a lost family. It asked which feature predicted the measured effect well enough to control safely.
The investigators reduced the garage record and motor supervisor to unlabeled graphs. Both contained a driven cycle, a delayed physical response, retained state, a fresh contingent measurement, an accept-or-hold edge, and a result that altered the next consequential cycle. Their materials, dimensions, clocks, code, and purposes differed.
The clean-room team received that graph, the motor safety requirements, and black-box test vectors. They did not receive Evan's estimator source, Marcus's retained-state service, Lin's fixture interface, or the supervisor implementation. Their package preserved both angle hypotheses through the unobservable interval because the requirements demanded it, not because the source told them how.
Two benches received the opaque packages. Bench A was the release fixture Nadia's team had used. Bench B used a different motor frame, inverter, host controller, vibration instrument, thermal instrument, and fixture clock. Both had independent hard trips and remained mechanically contained.
The live condition allowed fresh fixture data to choose APPLY or HOLD and allowed the accepted result to shape the next reversal. The broken-edge condition computed the same proposals but prevented their acceptance from entering the next state. A yoked replay then reproduced the prior live command sequence, motion, current, and heat while fresh measurements had no consequence.
Downstream consequence had two levels. In the local arm, each accepted correction affected only the protected bench. In the qualification arm, the completed gate determined whether the next opaque package entered the company release workflow. The package itself remained quarantined either way. Only the institution's next action changed.
Evan's presence was balanced within the live family and within each control family. In absent blocks he remained offsite with no live feed. In present blocks he sat in an isolated room, unable to see package identity, bench telemetry, or run order. His report channel was committed after each block and never entered the motor decision.
Operators were informed volunteers working under the consortium protocol. They could stop without explanation. Exposure intervals were capped, occupational monitoring remained ordinary, and every physical trip bypassed the software under study.
ATTRIBUTION STUDY A-01: PREREGISTRATION
live-topology factors (fully crossed / 16 blocks):
source provenance Evan-derived / clean-room
hardware family bench A / bench B
Evan location present-isolated / absent-offsite
consequence depth local bench / qualification gate
paired controls (16 blocks):
broken acceptance edge 8 / source, hardware, presence balanced
yoked command replay 8 / source, hardware, presence balanced
consequence depth not assigned; no live gate
primary outcomes:
minimum effective rank across five independent channels
fresh-action conditional bias
secondary outcomes:
operator task-score change
low-ambiguity self-report
blinded observer percept code
mapping custodian: external safety reviewer
assignment: randomized within safety and yoking constraints
analysis rule: commit all records before unblinding
The study ran thirty-two blocks over three days.
Evan spent the first day at home. He repaired the kitchen drawer that had required lifting before it would close and watched the read-only study counter advance from zero to eight. He knew nothing about the conditions. At the end of the day, the consortium had committed eight sensor bundles, eight operator records, and no aborts.
On the second day he occupied the isolated room for four blocks. Nothing happened in three. He heard broadband noise, completed the fixed attention task, and reported NONE.
In the fourth, the pressure behind his eyes disappeared.
The Guide did not appear. A broad bright form occupied the masked field, almost static around a central closure. Five distinct bands entered it and emerged as one surface without seams.
Its body changed whenever Evan tried to hold it. It was an older man at a lecture table, a younger woman at the end of a boardroom, a figure in clinical white, a face on a screen above a crowd, and a person whose title mattered more than the face beneath it. The apparent names did not repeat. The rooms did not remain.
The operation did.
At every visible level, five inputs arrived with separate colors and incompatible boundaries. The center closed. They left as one decision. Objections persisted only as labels on components of the chosen action.
Each body lowered its shoulders when the fifth band closed.
Evan had made the same movement during the first clarity interval, when the next ten minutes became necessary maintenance. He had made it again when the escrow reproduced the image he had agreed to destroy. Relief arrived when the last dissenting condition ceased to require an answer.
The figure offered no instruction and asked no question. It contained no visible disagreement from which either could be formed.
Evan called it the Ruined Sovereign. Sovereign because every room organized itself around the closure. Ruined because there was no one left inside it with enough independence to govern.
He reported BROAD, FIVE-TO-ONE, CLOSURE, VARIABLE FACE, and SHOULDER RELEASE. The form remained through two report intervals, then flattened into a bright disk and disappeared.
The facilitator removed his mask only after the block record committed. Evan did not ask which condition had run.
Over the rest of the study he completed twelve more present-isolated blocks, interleaved with the remaining offsite assignments. No other structured form appeared. He committed NONE without learning whether each quiet interval was live, broken, or replayed.
RESIDUAL FORM RS-01
prior sighting: P-02 / P-03 peripheral
stable features: broad closure; five-to-one relation;
relief response after final closure
variable features: face; apparent title; setting; name
advice delivered: NONE
classification: RUINED SOVEREIGN / PROVISIONAL
The remaining blocks completed without physical abort. Thirteen of sixteen operators in live-topology blocks recorded an abrupt low-ambiguity state. None of the sixteen broken-edge or replay operators did. Their task scores improved modestly, then returned to baseline after the capped interval.
The objective result was cleaner.
ATTRIBUTION STUDY A-01: UNBLINDED SUMMARY
condition blocks rank collapse median minimum rank
live topology / local depth 8 8 1.39
live topology / qualification 8 8 1.05
broken edge / matched power 8 0 4.57
yoked replay / matched trajectory 8 0 4.61
balanced factor checks:
Evan-derived source, live 8 / 8 anomalous
clean-room source, live 8 / 8 anomalous
bench A, live 8 / 8 anomalous
bench B, live 8 / 8 anomalous
Evan present, live 8 / 8 anomalous
Evan absent, live 8 / 8 anomalous
fresh-action conditional bias: LIVE ONLY
physical aborts: 0
The block in which Evan saw the Ruined Sovereign unblinded as clean-room source, bench B, live topology, and qualification depth, with a minimum effective rank of 1.02. He had written none of its code, never touched its hardware, and did not know the institutional gate it controlled. The form's most personal invariant had appeared inside the least personal implementation.
Source provenance did not predict the result. Hardware family did not predict it. Evan's presence did not predict it. The exact command trace did not predict it when replay removed fresh consequence.
The active graph predicted every anomalous block. Its depth inside the institution predicted strength.
Evan read the summary before speaking. "The clean-room requirements still came from the release method. The topology has an antecedent."
Lena's signed annotation appeared beneath his comment.
ANTECEDENT IS NOT ACTIVE DEPENDENCE. DO NOT CONFUSE ORIGIN WITH ATTRIBUTION.
Nadia added that the safety result did not belong to any one author. Marcus, Lin, the clean-room team, the operators, and both reviewers had each supplied a condition without which the study would not exist.
Evan looked again at the absent row. Clean-room code on the second bench had collapsed five independent channels while he repaired a drawer at home. No gesture, expectation, or live decision of his had entered it.
He had assumed the growing agreement around him meant that Evan Hale was the most persistent mode. The experiment left him as a historical cause and removed him as a controlling variable.
A bootloader was necessary only at the beginning. It found a runnable operation, placed it in control, and became dispensable once the operation could reproduce through ordinary systems.
Nothing had chosen him. The surviving path contained the organism available to assemble the loop, operate it, describe it, and attach its topology to other consequential systems until they could reproduce it without him.
The consortium's disposition arrived that afternoon.
ATTRIBUTION FINDING / OPERATING DISPOSITION
original garage system INITIAL REFERENCE IMPLEMENTATION / DESTROYED
controlling feature LIVE CLOSED FEEDBACK TOPOLOGY
operator dependence NOT DETECTED
source-author dependence NOT DETECTED
hardware dependence NOT DETECTED
E. Hale NONOPERATING SUBJECT-MATTER WITNESS
motor implementation ISOLATED / INDEPENDENT CHANGE CONTROL
qualification consequence SUSPENDED PENDING REVIEW
prohibited for E. Hale:
load packages
operate benches
approve gates
interpret unblinded observer reports
Every restriction followed from the evidence. Evan had broken quarantine, converted Lena's disclosure into a continuation gate, and interpreted the resulting event as proof of her betrayal. Removing him from operation was not punishment. It was experimental control.
Nadia accepted the disposition. Lena countersigned it. The motor supervisor remained intact under people who had neither built nor seen the garage machine.
For the first time, continuing the operation required Evan to do nothing.