Jiawan Harbor's New Shore project enters its second year. The works proceed quarter by quarter, and the dissent has not gone away: night-shift workers say the temporary passages are accumulating fatigue, fishers report that a stretch of inshore current has changed, and shopkeepers oppose keeping the old pier closed. All three voices are filed under "objections to the repair plan," yet they are not the same kind of material. The first reports repeated consequences for bodies and procedures, the second raises an environmental anomaly not yet confirmed, the third expresses waiting costs and an ordering of values.
If the system files all of them into a single opinion column, the voices appear to have arrived, while the feedback has in fact been flattened at the point of entry. If, in addition, every objection automatically suspends the works, every action loses its window of time. The question is not how to bring opinions to eventual agreement, but whether dissent can function as a signal within the recovery process: what object it points to, how long it takes to reach the position capable of changing the path, under what consequences it escalates, and how it decays when new observations no longer support it.
Dissent needs to be registered, reviewed, and retained within institutions, but the existence of a procedure does not yet mean the system is able to respond. Here the analytical clock places the three objections on one and the same recovery timeline, to observe the delay between signals and system action. Disagreement is not automatically correct because it has entered the feedback chain, nor can the system treat it as noise simply because construction is still under way.
The First Curve: Night-Shift Dizziness That Appears Every Day
In the first week after the temporary passages opened, two members of the night shift reported dizziness from the lighting. The site supervisor checked the illumination levels; the readings were within standard range, so the incident was recorded as individual discomfort. In the second week three more reported similar symptoms, but they belonged to different contractors, and the records were never aggregated. Only in the fourth week, when a worker was injured falling at a turn, did management first see the signals that had until then been scattered.
The failure of this feedback chain requires no one to suppress opinion. The classification is separated by employer, the safety forms count only accidents, and the dizziness has not yet reached the threshold for stopping work; each local unit handles things by the rules, yet the repeated pattern cannot form. The delay is not slow transmission of messages but the fact that the observations never reached the position able to compare different crews and change the passage design.
Two reports do not suffice to prove that the lighting causes harm. The dizziness may be related to sleep, weather, personal health, or unfamiliarity with the route. The function of an early signal is not to complete a causal determination but to change the next observation: record the specific times and turning positions, compare different lamps and shifts, and adjust a short segment of the route when the risk is controllable. To demand that the first reporter prove the general mechanism is to transfer the system's investigative capacity onto whoever bears the consequences earliest.
After the fourth week, the harbor reassembled the low-intensity reports that had preceded the accident by location and time slot, and found the problem concentrated at one angle of reflection on rainy nights. After the lamp positions were adjusted, new reports fell markedly. This outcome raises the credibility of the lighting explanation without proving that all the dizziness shares one cause. What the feedback completes is a loop from vague discomfort to limited correction, not the manufacture of a uniform truth for every experience.
The severe judgment is this: a system can keep demanding stronger evidence until injury supplies the missing strength. A "reliable conclusion" obtained this way uses avoidable losses as its sampling cost. The value of early feedback lies precisely in letting the system, before irreversible consequences appear, obtain more material through smaller actions.
The Second Curve: Three Current Anomalies That Did Not Repeat
Within one month, fishers three times reported anomalies in the inshore backflow, worrying that the new breakwater had altered the tidal regime. The engineering models showed no such trend, and measurements over the following two months did not repeat it. If lived experience were taken directly as proof of causation, the works could stall repeatedly over any non-reproducible phenomenon; if only the observation positions presupposed by the model were accepted, changes outside the model would never have a chance to enter.
This signal requires a temporal structure different from the first. The dizziness occurred on passages that people traverse daily and could be compared quickly; seasonal currents demand longer observation, and a missed tidal period means waiting for the next round. Feedback delay sometimes comes from institutions, and sometimes from the object itself. Requiring all questions to reach conclusions within the same deadline condemns slow objects either to early dismissal or to a permanent maximum alert.
The harbor neither wrote the three anomalies down as "confirmed risk" nor deleted them. It adjusted the observation positions for the next season and retained the wind direction, tidal level, and imagery recorded at the time. If the next cycle still shows no repetition, the signal's strength continues to decline; if it reappears under similar conditions, it enters a higher level of engineering review. That a signal may decay does not mean the reporter has been proven untrustworthy. The judgment concerns the strength of this claim under the current material; it does not turn a single non-replication into a personality file.
Conversely, the model must also state what it cannot see. The existing sensors were placed for the safety of the main channel and need not be able to answer questions of inshore livelihood and ecology. That the model shows no anomaly shows only that the measured objects show no anomaly. To read the uncovered areas as safe is to manufacture a blank in reality out of the boundaries of observation.
The system needs to permit the intermediate state of "not escalating now, but continuing to watch next season." It has none of the drama that travels well, yet it is the ordinary condition of processual cognition: conclusions change their strength with the material, neither feigning certainty to preserve authority nor refusing, as a display of openness, to form any provisional judgment.
The Third Curve: The Shopkeepers' Objection Is Not a Fault Alarm
The old pier remains closed, and the income of nearby shopkeepers is falling. They demand an earlier opening. This objection is not about whether equipment is faulty but about who bears the wait for recovery. If management required the shopkeepers to produce a safety model proving that the pier should open, it would wrongly impose the evidentiary standard of an engineering question on those harmed by the arrangement; if the shopkeepers' losses automatically decided the opening, risk would be shifted onto workers and users.
Value conflicts can also enter feedback, but the object they change is different. The shopkeepers' material shows that waiting costs are accumulating, forcing the plan to compare compensation, phased opening, and continued closure in terms of their consequences; it does not directly alter the engineering judgment of whether the breakwater is safe. Disguising a value objection as a technical signal sets both sides to contesting whose "facts" are truer, while the ordering of priorities is never faced.
Chapter 17 already treated value conflicts in the order of recovery, and Chapter 18 wrote intertemporal costs back onto their bearers. This chapter adds only the temporal dimension: that shopkeepers can wait a month does not mean they can wait a year; that workers can accept a temporary risk does not mean they can work under exceptional conditions indefinitely. Feedback must not enter only at the first decision; capacity to bear changes over time, and old trade-offs lapse accordingly.
The works keep the old pier closed while launching a reversible interim arrangement for trade. This satisfies no one among the shopkeepers, nor does it convert the objection into engineering consent. What the system gains is feedback of another kind: if the substitute arrangement cannot actually be used by anyone, waiting costs will continue to accumulate along the original path. Action and opposition can proceed simultaneously, provided the consequences the opposition points to can still redirect resources, rather than being heard only in meetings.
When Signals Multiply, What the System Loses First
Once repair enters its contested phase, the platform receives each day a flood of duplicate complaints, forwarded screenshots, and mutually conflicting risk judgments. Handlers begin to rely on automatic merging by keyword, and genuinely new anomalies may be folded into old threads. More information does not necessarily let the system see more clearly; when verification capacity becomes the shared bottleneck, any party can exhaust it with low-cost repetition, so that high-harm signals never reach the front of the queue.
Ranking cannot rest on judging who is sincere. Motives cannot be read directly, and the position of a dissenter offers no automatic factual advantage. What the system can compare is whether the object is definite, whether the sources are independent, whether the consequences may be irreversible, and whether the window of time is closing. Repeated signals may be merged in counting, but the new conditions must be retained; a hundred copies of one text are not a hundred independent observations, and a location not previously covered cannot be ignored because it has only a single report.
Filtering is equally open to abuse. Managers may label all unfavorable material as flooding, and opponents may call every merger suppression. Both charges must return to the path of the information: what was merged, on what basis, whether new material can re-enter detached from the old thread, and what consequences a wrong filter caused. If the measures that prevent overload also delete the entrance for counterexamples, what the system gains is quiet, not a capacity to recover.
A feedback system also changes reporting behavior. People who are publicly questioned after a single report stop reporting; if every report immediately triggers a work stoppage, crews may likewise stay silent to avoid bearing the delay. High-risk signals need protection and rapid response, while ordinary deviations need a low-cost entrance for trial and error. If all voices travel a single high-consequence channel, small problems dare not appear, and large ones are drowned in a mass of low-intensity information.
How Signals Escalate, and How They Fall Quiet
A system that keeps raising alarms without ever affecting action is only manufacturing visibility. A system in which every alarm reroutes the main path will be pulled along by noise. Feedback needs tiers of escalation proportionate to consequences: early anomalies change observation, repeated anomalies trigger local trials, strong evidence combined with high harm drives a switch of path; where irreversible risk is involved, protection may come first and attribution be left to stronger material.
Signals also need to decay. When new observations persistently withhold support, when the conditions of formation have changed, when the corresponding path has been retired, the alarm's strength should fall. The fall does not erase history; it prevents past dangers from permanently occupying present attention. Without a decay mechanism, the system carries more and more alarms that cannot be closed and in the end can only ignore them wholesale; without historical traces, the same anomaly starts from zero when it reappears.
When to escalate and when to let decay cannot be settled by a single formula. The reversibility of harm, the independence of the signal's sources, feedback delay, and the cost of action jointly bear on the judgment. Value choices remain. The analytical clock's role is not to eliminate choice but to return choice to time: what is lost by not acting now, what material a small preliminary action would obtain, and who bears the waiting.
Repair, too, must permit ending. That absolute safety cannot be proven does not mean an investigation can never close; that unanimous consensus has not formed does not mean the works can never proceed. The honest form of an ending states the scope of the current conclusion and the conditions for reopening it. What matters is whether, when a similar signal next arrives, the system can still pick up the earlier timeline, rather than misrecording a closed round of investigation as a conclusion never to be reopened.
Repair Does Not Need Quiet; It Needs Noise That Can Return
The three curves end differently: the dizziness reports drove local modification, the current anomalies decayed for now and await the next season, and the shopkeepers' objection left the engineering safety judgment unchanged while altering the arrangement of waiting costs. Had they all been recorded at the outset merely as "opposition to the repair," none of these three outcomes would have appeared.
Disagreement is valuable to stability not because opposition is right by nature, but because only where things depart from expectation can the system learn the boundaries of its models. Nor is more noise always better. Repetition that cannot identify object, source, and time occupies processing capacity; a judgment that has been falsified yet refuses adjustment turns from feedback into an identity commitment. Returnability means the signal can go back to the object it points to, and the object's new state is likewise allowed to change the signal's strength.
Whether a recovering system still has feedback loops short enough determines whether it can change its actions before error crosses a critical point. Flattening opposition makes surface recovery faster while deferring the real cost; waiting for all opposition to disappear lets repair miss its window. Between the two lies not a muddled compromise, but letting action advance under conditions that remain open to refutation.
The next chapter discusses how transitional losses are shared. Feedback tells the system where damage is occurring, but it does not automatically decide who should be compensated first and who can go on waiting. That still requires a public value judgment. As of this chapter, one thing only can be affirmed: a recovery plan that can run only in quiet has already lost the capacity to be the first to know where it is wrong; only a system that lets noise return to its object, change observation, and lower its alarms as evidence fades still has a chance to correct course before collapse.