Anthony Riccardo, an inmate of the Centraba Correctional Center in Illinois, needed a new cellmate after his former cellmate complained about being housed with him. Normally that pairing would have been made by Centralia’s placement office, but when the evening of May 28, 1997, arrived and some inmates remained unassigned after the regular placement officers had left, the task fell to Lt. Larry Rausch, who was serving the second shift. Rausch matched Riccardo with Juan Garcia, a pairing that should have lasted only until the placement officers on the day shift could review matters. Two days later Garcia sexually assaulted Riccardo, who sued Rausch under 42 U.S.C. § 1983. A jury concluded that Rausch had subjected Riccardo to cruel and unusual punishment and awarded $1.5 million in compensatory damages. The district court entered judgment on this verdict, and Rausch appeals.
His lead-off argument is that Riccardo failed to use his administrative remedies. If so, then 42 U.S.C. § 1997e(a), part of the Prison Litigation Reform Act, forecloses this suit even though Riccardo challenges a discrete incident and wants a form of relief — money damages — that the administrative process in Illinois does not provide. See Porter v. Nussle, 534 U.S. 516, 122 S.Ct. 983, 152 L.Ed.2d 12 (2002); Booth v. Churner, 532 U.S. 731, 121 S.Ct. 1819, 149 L.Ed.2d 958 (2001). Riccardo did file an administrative grievance, but Rausch contends it was too late (in February 1998, while Illinois sets a limit of six months) and asked the state to prosecute Garcia rather than do anything about Rausch and the classification system.
Prisoners must follow state rules about the time and content of grievances. See Pozo v. McCaughtry, 286 F.3d 1022 *513(7th Cir.2002); Strong v. David, 297 F.3d 646 (7th Cir.2002). Failure to do this means failure to use (and thus to exhaust) available remedies. Yet the state’s administrative apparatus did not reject Riccar-do’s grievance as untimely; it accepted and denied the grievance on the merits. At the time of these events, Illinois permitted a filing after six months when the prisoner had good cause, see 20 Ill. Admin. Code § 504.810 (1997). The official handling the grievance must have found good cause; anyway, we held in Pozo that, when a state treats a filing as timely and resolves it on the merits, the federal judiciary will not second-guess that action, for the grievance has served its function of alerting the state and inviting corrective action. 286 F.3d at 1025.
As for the content of this grievance: true enough, its main objective was to have Garcia prosecuted. (Rieeardo deemed inadequate Garcia’s punishment within the prison system.) But it also at least hinted at problems in prison administration. Rieeardo wrote: “[T]he administration don’t [sic] do there [sic] job. [A sexual assault] should’ve never [sic] happen again.” This language is ambiguous. There are two principal ways to reduce the number of sexual assaults in prison: better steps ex ante to separate potential aggressors from potential victims; and harsher penalties ex post in order to deter future assaults. Rieeardo did not distinguish between the two, and a prison administration receiving such a grievance should have considered both. Illinois has not adopted any rule governing the level of detail required of prisoners’ grievances. “When the administrative rulebook is silent, a grievance suffices if it alerts the prison to the nature of the wrong for which redress is sought. As in a notice-pleading system, the grievant need not lay out the facts, articulate legal theories, or demand particular relief. All the grievance need do is object intelligibly to some asserted shortcoming.” Strong, 297 F.3d at 650. The document that Rieeardo filed is at the border of intelligibility; it is hard to imagine much less that a prisoner could do and still alert the prison; yet this grievance did complain that Garcia had committed a rape and that “the administration don’t do there job.” A generous construction of this grievance would have induced the prison to consider the possibility that the guards could have prevented this assault. So we conclude that Rieeardo exhausted the administrative process. If Illinois wants grievances to be more detailed, it must adopt appropriate regulations and inform prisoners what is required of them. Rie-cardo is entitled to a decision on the merits of his constitutional claim.
Because Rieeardo prevailed at trial, we recap the facts in the light most favorable to his position. Rieeardo was anally raped by his cellmate at Cook County Jail, shortly after his conviction for aggravated assault. When he arrived at Centraba Correctional Center in November 1996, he told the prison psychologist that he did not feel safe. After a stint in segregation for violating prison rules, Rieeardo declined to return to the general population. He told guards that a cellmate in the segregation unit had stolen some of his property and objected to spending another day with that cellmate. He believed that the responsible inmate belonged to the Latin Kings gang and that the Latin Kings may have been preparing to kill him — though he did not say why he believed this. But Centraba allows inmates to veto housing with persons they declare to be enemies, so the prison found Rieeardo a new cellmate. When, after a few days, that cellmate objected to spending more time with Riccar-do, another pairing was required. (The segregation unit was too crowded to permit Rieeardo a cell of his own.)
*514During the afternoon of May 28, Garcia had offered to help Riccardo retrieve his stolen property. Riccardo took this as an ill omen rather than as a genuine offer of assistance and told Lt. Alemond that he feared for his life if celled with Garcia. Although Alemond said that he would “take care of it,” he did nothing — he did not either find a cellmate for Riccardo or alert Lt. Rausch, Alemond’s replacement on the next shift. About 9:30 that evening, Rausch brought Garcia to Riccardo’s cell and told him that Garcia was his new cellmate. Before the cells were locked for the night, Riccardo sought out Rausch in private and told him that he believed that the Latin Kings had a “hit” out on him, and that he feared for his life if celled with Garcia. Rausch replied that there was no place else to put Garcia (or Riccardo) that evening, and that he could not refuse housing while in segregation. Rausch then brought Riccardo and Garcia back together and asked each, in turn, if he had a problem with the other. Riccardo shook his head in the negative. Rausch took that as agreement to the assignment. That was Riccardo’s last contact with Rausch. As we have mentioned, nothing untoward happened that evening or the next, but during the evening of May 30 Garcia compelled Riccardo to perform oral sex. The record does not suggest that this assault had any connection to the Latin Kings. During the time between assignment and assault, Riccardo did not ask for a different cellmate (though he did file two grievances on May 29 about other matters). Circumstances brought out at trial suggest that other guards should have recognized on May 30 that problems had developed between Riccardo and Garcia; their failure to intervene may be culpable but cannot be attributed to Rausch, whose liability depends exclusively on his actions the evening of May 28.
Rausch did not assault Riccardo and is not vicariously liable for Garcia’s crime. Like other guards, however, Rausch was required to refrain from placing Riccardo in harm’s way gratuitously. The qualification “gratuitously” is important, because prisons are dangerous places. Inmates get there by violent acts, and many prisoners have a propensity to commit more. Guards cannot turn away persons committed by the courts; nor do individual guards have any control over crowding and other systemic circumstances. All that can be expected is that guards act responsibly under the circumstances that confront them. See McGill v. Duckworth, 944 F.2d 344 (7th Cir.1991). A guard may be responsible without being credulous. Some prisoners are manipulative and cry “wolf’ in an effort to have a cell to themselves or choose a favored cellmate. Other prisoners perceive specters in every shadow, even though their fears are unsupported. (There is, for example, no reason to think that the Latin Kings ever had it in for Riccardo. He did not belong to a rival gang, and there is no history of violent or overtly hostile encounters between Riccar-do and any gang member.) Guards therefore must discriminate between serious risks of harm and feigned or imagined ones, which is not an easy task given the brief time and scant information available to make each of the many decisions that fill every day’s work.
The eighth amendment does not demand that guards perform this task flawlessly. It does not even hold them to the negligence standard. Liability is possible, instead, only when a guard is deliberately indifferent to a substantial risk of serious harm. See Helling v. McKinney, 509 U.S. 25, 113 S.Ct. 2475, 125 L.Ed.2d 22 (1993); Wilson v. Seiter, 501 U.S. 294, 111 S.Ct. 2321, 115 L.Ed.2d 271 (1991); Estelle v. Gamble, 429 U.S. 97, 97 S.Ct. 285, 50 *515L.Ed.2d 251 (1976). “Deliberate indifference” means subjective awareness. See Farmer v. Brennan, 511 U.S. 825, 114 S.Ct. 1970, 128 L.Ed.2d 811 (1994). It is not enough, the Court held in Farmer, that the guard ought to have recognized the risk. Instead, “the official must both be aware of facts from which the inference could be drawn that a substantial risk of serious harm exists, and he must also draw the inference.” Id. at 837, 114 S.Ct. 1970.
Rausch contends that Riccardo did not face a “substantial risk of serious harm” from Garcia on the evening of May 28, 1997; that, if such a risk was present, Rausch did not appreciate its existence; and that at all events he is entitled to qualified immunity because reasonable officers would not necessarily have understood that the law clearly required Riccardo and Garcia to be in separate cells. The first two matters (the objective and subjective components of the eighth amendment) are for the jury in the first instance, with appellate review limited to the question whether any reasonable juror could have found that the requisite level of risk existed, and that Rausch knew it. Immunity, however, is a matter of law for the court, to be decided without deference to the jury’s resolution — and preferably before the case goes to the jury. See Saucier v. Katz, 533 U.S. 194, 202, 121 S.Ct. 2151, 150 L.Ed.2d 272 (2001); Anderson v. Creighton, 483 U.S. 635, 107 S.Ct. 3034, 97 L.Ed.2d 523 (1987). The district court brushed aside Rausch’s invocation of immunity, writing that a guard cannot benefit from immunity if the action taken was not a reasonable response to a risk actually foreseen. That approach, which merges immunity and the merits, is incompatible with Saucier and its predecessors. See 533 U.S. at 203-06, 121 S.Ct. 2151. Immunity protects officials who act at the “hazy border” (id. at 206, 121 S.Ct. 2151) between the lawful and the forbidden. That Rausch may have overstepped the line does not mean that every reasonable officer would have been bound to know that Rausch acted improperly. We need not pursue the immunity defense, however, because Saucier calls on appellate courts to address the merits first, see id. at 201, 121 S.Ct. 2151, and Rausch is entitled to prevail outright: no reasonable juror could have concluded, on this record, that Rausch actually recognized that placing Garcia and Riccardo together exposed Ric-cardo to substantial risk.
Now it might seem that Rausch had to appreciate the risk, because (a) Riccardo claimed to fear for his life if celled with any member of the Latin Kings, and (b) Garcia in fact harmed Riccardo. One problem with relying on how things turned out to show knowledge of risk beforehand is that Garcia did not act for the Latin Kings; he told Riccardo that he was fulfilling a personal fantasy, and Riccardo believed this explanation. Thus the risk that Riccardo professed to fear (a “hit”) did not come to pass. Even under the law of negligence, this is an important distinction. If a school district entrusts a bus to a driver with a bad drinking record, and the tipsy driver runs the bus off the road while speeding, the school district is liable; but if instead there is an accident for which the driver is not at fault (a tree falls on the bus), or the driver collapses of a heart attack while on the road, the district is not liable, because that was not the type of risk created or increased by the negligent conduct even though hiring this particular driver was in the causal chain. See, e.g., Berry v. Sugar Notch Borough, 191 Pa. 345, 43 A. 240 (1899). The risk from which Riccardo sought protection was not realized; for all this record shows, the (objectively evaluated) risk to Riccardo of sharing a cell with Garcia was no greater than *516the risk of sharing a cell with any other prisoner.
As for Rausch’s subjective assessment: though Riccardo initially asserted mortal fear, when later asked whether he “had a problem” with Garcia he shook his head to give a negative answer. Rausch then had to decide which statement to believe. Ric-cardo argues, and the jury evidently concluded, that Rausch should have believed the first statement, communicated in private, rather than the second, communicated in Garcia’s presence. A rational jury could have thought that guards should give priority to statements made in private. (Rausch testified that, if Riccardo had claimed to “have a problem” with Garcia, they would have been separated; but Ric-cardo might have feared the consequences in a later encounter in the prison’s general population.) Still, what Rausch should have believed is not the right question; we need to know what he did believe. No reasonable jury could have found, in light of Riccardo’s denial of “a problem” with Garcia and Rausch’s decision to act accordingly, that Rausch subjectively appreciated that his action would expose Riccardo to a substantial risk of serious harm.
As we have already explained, prisoners may object to potential cellmates in an effort to manipulate assignments, or out of ignorance; thus although a protest may demonstrate risk it does not necessarily do so. The Constitution does not oblige guards to believe whatever inmates say. How does a reasonable guard separate fact from fiction? Rausch knew when making the assignment at least two things beyond Riccardo’s contradictory assertions. First, Rausch knew that Garcia was himself in segregation for protection from the Latin Kings (or at least a subset of them). Perhaps Garcia was manipulating the system himself, falsely asserting to fear the Latin Kings so that he could serve as their assassin; but at least at first cut Garcia could not be deemed a gang enforcer (and, as we learned ex post, his attack on Riccar-do was neither a “hit” nor gang related). Second, Rausch knew that Garcia had a clean record in prison. He had not been disciplined for acts of violence (let alone for sexual assault). That makes it reasonable for Rausch to have deemed Riccardo’s initial protestation unjustified. It is not as if Rausch housed Riccardo with a known sexual predator.
Riccardo might have responded to these facts by showing that there is a strong correlation between prisoners’ professions of fear and actual violence. How many murders (or homosexual assaults) occur in Centraba (or the Illinois prison system) per hundred inmate-years of custody? How many violent events were preceded by requests for protection? How many requests for protection were dishonored, yet nothing untoward happened? Data along these lines would have enabled a jury (and the court) to evaluate actual risks. If violence is common at Centraba, and inmates have good track records in identifying potential aggressors, then guards who do not have their heads in the sand must actually (that is, subjectively) understand the risk an inmate faces when a protest is disregarded. But if violence is rare, or if there is poor correlation between inmates’ alarums and subsequent violence, then Riccardo’s initial protest would not have provided Rausch with actual knowledge of an impending assault. The record does not contain any evidence along these lines. At oral argument Ric-cardo’s counsel expressed dismay at the idea that inmates’ professions of fear should be put to an empirical test. As counsel saw things, prisoners are unerringly accurate in appreciating the risks they face and invariably truthful in dealing with the staff. That seems to us unlikely; *517and if it is so it must be proved to be so and was not.
Rausch also was entitled to believe that his assignment of Garcia and Riccardo to share a cell would last for one night only. During the next day shift the placement office, armed with better information, was supposed to make a fresh evaluation and, if appropriate, a new assignment. Apparently that did not happen; the record does not show why. (Maybe it did happen and the staff approved Rausch’s action.) If Rausch knew that the staff charged with this responsibility routinely failed to carry it out, then he might have .been obliged to take additional precautions (such as separate interviews of Garcia and Riccardo to probe these issues more deeply) before making an assignment. Rausch himself testified that separate interviews would have been better practice, but the Constitution does not enforce all “better practices”; this is one respect in which the eighth amendment standard differs from the negligence standard. But Riccardo does not contend, and the record does not demonstrate, that disregard of the classification system at Centraba was so common that Rausch was bound to know that his assignment would last indefinitely. Nor was Rausch bound to foresee that, if Ric-cardo was in genuine fear, he would neglect to complain the next day, when he readily could have done so. (Recall that Riccardo filed two grievances on May 29 about other subjects.) Riccardo testified that he was too terrified to protest and was put off by Rausch’s assertion that prisoners in segregation can’t refuse assignments; yet grievances are confidential (so Garcia would not have known), and prisoners often appeal over the head of a guard who has told them that something can’t or won’t be done. Riccardo had already objected to, and obtained the removal of, at least one cellmate assigned to him in segregation. At all events, the question on the table is what Rausch knew (or deliberately avoided learning) on May 28; and there is no evidence that Rausch subjectively believed that Riccardo would fail to use his opportunity to seek further review the next day.
Illinois is free, if it wishes, to give prisoners veto power over the identity of their cellmates. But the eighth amendment does not do so of its own force, and prisoners cannot use'the Constitution to achieve this control indirectly by making unsubstantiated assertions. The constitutional question is not what Riccardo (initially) said, but what Rausch actually believed. This record does not permit a reasonable jury to find that Rausch knew or deliberately disregarded the fact that his actions subjected Riccardo to a substantial risk of serious harm, so the judgment is
REVERSED.