In the

    United States Court of Appeals
                 For the Seventh Circuit
                     ____________________
No. 18‐2214
TEREZ COOK,
                                                Petitioner‐Appellant,
                                 v.

BRIAN FOSTER, Warden,
                                               Respondent‐Appellee.
                     ____________________

         Appeal from the United States District Court for the
                   Eastern District of Wisconsin.
          No. 13‐CV‐989 — Nancy Joseph, Magistrate Judge.
                     ____________________

   ARGUED OCTOBER 3, 2019 — DECIDED JANUARY 29, 2020
               ____________________

   Before WOOD, Chief Judge, and BARRETT and SCUDDER,
Circuit Judges.
    WOOD, Chief Judge. Federal courts do not lightly grant pe‐
titions for a writ of habeas corpus brought by state prisoners.
As the Supreme Court put it in Harrington v. Richter, 

(2011), if the “standard [for relief] is difficult to meet, that
is because it was meant to be.” 

Nonetheless, “diffi‐
cult” does not mean “impossible,” as the Court reaffirmed in
Richter: “The writ of habeas corpus stands as a safeguard
2                                                   No. 18‐2214

against imprisonment of those held in violation of the law.”

Our task in the present case is to decide whether
petitioner Terez Cook demonstrated that Wisconsin’s court
of appeals unreasonably assessed his contention that he did
not receive the effective assistance of counsel guaranteed by
the Sixth Amendment. See Strickland v. Washington, 

(1984). The district court thought that Cook’s showing
fell short, but we conclude that he is entitled to relief. We
therefore reverse.
                               I
    Cook’s criminal case arose out of a home invasion that
took place in Peshtigo, Wisconsin, in May 2005. The state
charged Cook and another man, John Egerson, with armed
robbery, armed burglary, false imprisonment, battery, theft,
and mistreatment of an animal causing death. The jury con‐
victed Cook on all counts, as a party to the crimes and as a
repeat offender. Throughout these proceedings Cook’s basic
contention has been that the state identified the wrong man
as Egerson’s accomplice.
    We begin with an overview of the trial, to provide a
framework for the particular ways in which Cook contends
that he received constitutionally ineffective assistance of
counsel. We then address each error individually, and finally
we consider whether, taken as a whole, they add up to a
Sixth Amendment violation.
    First, we provide the cast of key characters:
          Terez Cook: defendant accused of the home‐
           invasion crimes; possibly the same person as “BN”
           or “Rex”
No. 18‐2214                                                3

         John Egerson: co‐defendant, tried separately and
          convicted
         Ashley Sadowski: Egerson’s girlfriend and ac‐
          complice to the home‐invasion crimes
         Jessica Babic: Sadowski’s friend, and accomplice to
          the home‐invasion crimes
         David Hall: long‐time friend of Egerson, and also
          friends with Sadowski, Babic, and an ex‐boyfriend
          of the victims’ daughter; Cook alleges that Hall,
          not he, was Egerson’s accomplice.
         Stacy Thede: Cook’s girlfriend
         Jimmy and Margaret Harper: the victims
    The events underlying this case unfolded as follows.
Egerson and Sadowski believed that there was marijuana in
the Harpers’ garage, and they wanted to steal it. Around
midnight on May 22, 2005, Sadowski and Babic met up with
Egerson and Cook. At 2:30 a.m. or so Sadowski, Babic, Eger‐
son, and another man (Cook or “Rex,” according to the state;
Hall, according to Cook) went to Walmart; there, at Eger‐
son’s urging, Sadowski and Babic bought gloves, bandanas,
and duct tape. So equipped, Sadowski drove Egerson’s car
past the Harpers’ home; she tapped the brakes when she
reached the front of the house in order to signal to the men,
who were following in Sadowski’s car, which house to tar‐
get. The women then waited nearby for the men to do the
job.
   Around 4:00 a.m., Egerson called Sadowski and told her
that he had crashed her car after stealing cash and speakers
from the Harpers’ home. The women picked up Egerson and
4                                                    No. 18‐2214

his companion and drove to a hotel in Green Bay. About six
hours later, Egerson dropped Sadowski and Babic (but not
Cook, it seems, who disappears at that point from the state’s
story) at a gas station in Peshtigo. The women called Hall for
a ride to Babic’s house. There they were greeted by the po‐
lice, who arrested Hall and took the women into custody for
questioning. After initially denying any involvement in the
crime, Sadowski and Babic admitted their involvement and
named Egerson as one of the robbers.
    Putting together evidence from these interviews, along
with cell tower evidence, the state obtained an information
charging Cook and Egerson with the crimes. They were tried
separately, though by the same judge. In a trial that the pre‐
siding judge later characterized as unworthy of confidence,
the jury convicted Cook and the judge sentenced him to 40
years in prison and 18 years of extended supervision. His
conviction was affirmed on direct appeal in the Wisconsin
courts. At that point Cook (acting pro se) filed a petition for
postconviction relief pursuant to Wis. Stat. § 974.06; in it, he
alleged ineffective assistance of his trial counsel, Alf Langan,
and his appellate counsel, Milton Childs. The court appoint‐
ed postconviction counsel for Cook and ultimately held evi‐
dentiary hearings over the course of three days. It concluded
that Cook’s motion had to be granted because of the cumula‐
tive effect of trial counsel’s many missteps. In so ruling, the
court stressed the exceptional nature of the case:
       You know, I’ve been on the bench 20 years, and I
    can’t remember ever granting a new trial because of
    ineffective assistance of counsel. It may have hap‐
    pened, but I can’t as I sit here today recall. It’s a heavy
    decision. I understand that.
No. 18‐2214                                                   5

      ***
       I’ve given it a lot of thought. And the bottom line
   is that the deficiencies are so big that I would have to
   conclude if it had been tried correctly, that there’s a
   probability of a different result and that confidence of
   the Court has been shaken as to the results because of
   the deficient performance.
Doc. 62‐10 at 153–54.
    The state appealed from the trial court’s decision, and the
Wisconsin Court of Appeals reversed. Cook then filed a peti‐
tion for a writ of habeas corpus under 28 U.S.C. § 2254,
which a magistrate judge, proceeding by consent of the par‐
ties under 28 U.S.C. § 636(c), denied. Despite its conclusion
that the state appellate court, in resolving some of Cook’s
claims, had unreasonably applied Strickland or unreasonably
determined the facts, the district court thought that there
was enough to squeak by under the deferential standard that
applies to these cases. Because of the disagreement between
the state trial and appellate courts, however, it granted Cook
a certificate of appealability. See 28 U.S.C. § 2253(c).
                              II
    Under the Antiterrorism and Effective Death Penalty Act
of 1996 (AEDPA), a federal court is not authorized to issue a
writ of habeas corpus on a claim rejected by a state court on
the merits unless the state‐court decision was “contrary to,
or involved an unreasonable application of, clearly estab‐
lished Federal law, as determined by the Supreme Court,” or
was “based on an unreasonable determination of the facts.”
28 U.S.C. § 2254(d). A state court unreasonably applies fed‐
eral law if it correctly identifies the governing Supreme
6                                                    No. 18‐2214

Court precedent but unreasonably applies its holding to the
facts of the case. See Lockyer v. Andrade, 

, 75 (2003).
A state‐court decision involves an unreasonable determina‐
tion of the facts if the court finds that “the factual premise
was incorrect by clear and convincing evidence.” Miller‐El v.
Cockrell, 

, 340 (2003); see also Brumfield v. Cain,

, 2277 (2015) (facts may not be set aside if “rea‐
sonable minds reviewing the record might disagree about
the finding in question”).
    With these principles in mind, we turn to the specifics of
Cook’s case. At this stage of the proceedings, with the assis‐
tance of able recruited counsel, Cook has identified six re‐
spects in which Langan rendered ineffective assistance of
counsel. He has not raised any argument about the lawyer
who represented him in his direct appeal, and so we have
nothing to say about that. We address each specific conten‐
tion of defective performance separately. We then turn to the
question of prejudice, which we assess by evaluating the trial
as a whole, not one slip at a time.
                               III
                                A
    1. Failure to locate and produce Hall at trial
    Hall, as we noted, is the person who Cook insists was
Egerson’s accomplice for the robbery. Cook’s theory of de‐
fense was that he backed out of the robbery at the last mi‐
nute. He did not contest the fact that he was with Egerson,
Sadowski, and Babic on the morning of the crime. He main‐
tained, however, that he “decline[d] to go along” with the
break‐in, and Hall took his place. Calling Hall the “biggest
hole” in the state’s case, Langan argued that the police failed
No. 18‐2214                                                7

adequately to investigate his involvement. Despite arresting
Hall with the women the morning after the robbery, officers
did not test for his DNA on a headwrap and glove found in
Sadowski’s car. They did test for Cook’s DNA but did not get
a match on these items. Cook’s DNA was, however, found
on a cigarette in Sadowski’s car. Langan was also able to put
before the jury the following facts: Egerson and Hall had
known each other since childhood; Hall was friends with
Sadowski, Babic, and an ex‐boyfriend of the Harpers’ daugh‐
ter; Hall had stopped by Babic’s house twice on the night of
the robbery; and neither Sadowski nor Babic had told the
police about Hall’s visits. Babic and Sadowski both denied
that Hall was involved in the robbery.
    But the jury never saw Hall, because Langan did not
subpoena him to testify at Cook’s trial. Worse, he made no
effort to locate Hall. When the judge asked about this out‐
side the presence of the jury, Langan reported that Hall was
“in prison.” The judge then stated, “at least you know where
to find him, and you could get a writ to produce him,” but
Langan said that he had been unable to figure out which
prison Hall was in. (Wisconsin has a readily available inmate
locator system. See https://offender.doc.state.wi.us › lop.
Moreover, it turned out that Hall was not in prison, nor was
he in hiding; he was working in the open and reporting to a
probation officer.) The judge worried that “if a fair trial to
Mr. Cook hangs in the balance,” Hall’s presence may be nec‐
essary to allow the jury to “eyeball him and give some con‐
sideration to this notion, that maybe it was Egerson and Hall
and not Cook and Egerson.” Langan did introduce a photo
of Hall to the jury. Both Cook and Hall are African‐American
men of roughly the same age and build.
8                                                    No. 18‐2214

    Notwithstanding Hall’s centrality to Cook’s theory of de‐
fense and Langan’s lassitude in attempting to explore his
role in the events, the state appellate court found neither de‐
ficient performance nor prejudice in this respect. This is hard
to understand, as reasonable professional performance in‐
cludes the duty to investigate. Wiggins v. Smith, 

,
522–23 (2003); 

. The district court, in
contrast, concluded that the state appellate court’s “finding
that Cook failed to establish deficient performance as to
Langan’s failure to find Hall before trial is an unreasonable
application of Strickland.” We agree with the latter assess‐
ment. We add that counsel’s failure to investigate cannot be
cured by later discovered facts—in this instance, that Hall
for a time had absconded while on parole. See 

–27 (rejecting “strategic decision” for attorney’s
performance relied on by state court as a “post hoc rationali‐
zation”).
    Had the jury seen Hall, it is reasonable to conclude that
there would have been strong corroboration of some damn‐
ing evidence: both Sadowski and Babic were friends with
Hall; Hall had known Egerson since childhood; Hall stopped
by Babic’s home twice on the night of the robbery; neither
woman told the police about Hall’s visit; and Hall was ar‐
rested at Babic’s house the next day. It is true that Hall did
not blurt out a confession at the postconviction hearing, but
that does not mean that his presence before the jury would
have had no impact. The content of his testimony is relevant
instead to the question of prejudice.
    The state appellate court also opined that Hall’s presence
at trial would not have lessened the impact of the women’s
testimony identifying Cook as the second robber, particular‐
No. 18‐2214                                                  9

ly since Cook admitted that he was with them shortly before
the robbery. (Note that the jury never learned of the lenience
both women expected from the state, either—here as else‐
where, the errors Cook asserts are intertwined.) The stand‐
ard, however, is not whether Hall would have exonerated
Cook; it is whether the absence of Hall’s testimony under‐
mines our confidence in the jury’s verdict that Cook was the
one involved in the crime. This is troublesome, even though
the court elsewhere phrased the test properly. We conclude
that counsel’s failure even to try to obtain Hall for the trial
represented deficient performance.
   2. Failure to object to hearsay testimony, unsupported
      by a proper foundation, about cell phone records
    The jury heard that the police learned that around 4:00
a.m. on the day of the robbery, Egerson’s cellphone had re‐
ceived calls from a phone that they tracked first to Stacy
Thede, Cook’s girlfriend, and ultimately to Cook. This in‐
formation came in through the testimony of Detective Bald‐
win, who said that the records revealed that the calls hit cell
towers near the Harpers’ home. Thede told detectives that
she had bought the phone for Cook, but that a few days be‐
fore the robbery, he told her that he had lost it. The officers
took Cook into custody for questioning.
    Langan did not object to Baldwin’s testimony about the
phone records, even though the state did not introduce them
into evidence. His failure to object was serious, because this
testimony provided the critical link between Cook’s phone
and the robbery and thus enabled the police to locate Cook.
And it was worse than inadvertent: outside the presence of
the jury, the trial judge expressed concern that “so much”
hearsay about the phone records and tower hits was coming
10                                                    No. 18‐2214

in through the detective’s testimony. Langan responded that
he thought the state was going to introduce the telephone
records, which he had reviewed during discovery. The state
never did so.
    Langan’s disregard of the cellphone evidence does not
stand in isolation from other moving pieces. The state appel‐
late court saw the use of Cook’s phone during the crime as
evidence of his involvement. Yet, it concluded, other evi‐
dence proved a similar point and thus diluted the impact of
the missing records. The state court observed that a DNA
swab found on a cigarette from Sadowski’s car matched
Cook’s DNA, and it stated that “Cook eventually admitted
that he was in Peshtigo and was with Egerson, Sadowski, and
Babic at the Walmart and after the home invasion … .” But the
latter statement does not appear in the record.
    In its brief, the state asserted in several places (consistent‐
ly with the state appellate court) that Cook told the police
not only that he was with Babic, Sadowski, and Egerson be‐
fore the robbery, but also—and this is important—that Cook
“admitted that he was in the Walmart parking lot with the
others when Egerson directed the two women to purchase
gloves, bandanas and duct tape after 3 a.m. shortly before
the home invasion.” We found no record support for this
statement, and at oral argument we invited counsel to direct
our attention to whatever he could find. That supplemental
letter has not assuaged our concerns. The closest the state
was able to come was a statement from Baldwin reporting
that he believed that Cook said that he had been in the
Walmart parking lot, with no reference to time and no refer‐
ence to Egerson’s alleged instructions to the women. And
Baldwin admitted that Cook never said he went into the
No. 18‐2214                                                             11

Walmart. Thus, we find the state’s brief overstated perhaps
the most material facts in the case. This is concerning, for, if
true, the state’s characterization would mean that Cook ad‐
mitted to being with Egerson, Babic, and Sadowski at the
very location, at the very time, and for the very purpose of
purchasing the very supplies used to commit the robbery.
But the state’s characterization is inaccurate: the trial testi‐
mony supports no more than Cook perhaps admitting he
was at the Walmart at some unspecified time before the rob‐
bery. Nowhere did he admit the broader storyline the state
pressed in its brief.1
    This is a serious misapprehension of the facts: without
the Walmart parking lot evidence, the error with the cell‐
phone records becomes critical. Because of its mistake, the
state appellate court found adequate performance and said
nothing about prejudice. We, however, cannot agree with
respect to performance. Once again, Cook’s attorney failed to
undertake the necessary background work and thereby left a
glaring gap in the evidence. Langan knew that the state had
no witness from the phone company and would not be able
to establish a foundation if the court sustained an objection.
The records directly linked Cook (or, more accurately, his
cellphone) to the 4:00 a.m. getaway call. And, as the district
court noted, the argument Langan actually made—that the
phone had been stolen—pales in comparison to the evidence
the jury heard about the cell phone tower hits in the vicinity
of the robbery near the time of the robbery. The cellphone
evidence also corroborated Sadowski and Babic’s testimony

    1 The state shoulders a weighty obligation to play entirely straight
with facts that affect a person’s liberty. Too much is at stake for all in‐
volved to see what we did here from the state.
12                                                  No. 18‐2214

that they received Egerson’s call from Cook’s phone after the
robbery. Nothing explains Langan’s lack of effort to keep it
out. In this respect as well, his performance was deficient.
     3. Failure to bring out the de facto immunity the state
        had given to Sadowski and Babic in exchange for
        their testimony
    Everyone recognized that the credibility of Sadowski and
Babic was pivotal to the prosecution’s case. Once again,
Langan left essential work undone. And once again, it can‐
not be because it slipped his notice. The judge held a sidebar
and expressly asked whether law enforcement had promised
immunity or other concessions to Sadowski or Babic or both;
he described Langan’s failure to explore this issue as
“strange.” At the judge’s prompting, but without the jury
present, Langan finally questioned both women about a po‐
tential deal. Babic testified that she gave her statements to
law enforcement with the understanding that she would not
be charged, and that she thought she had immunity. None‐
theless, Langan dropped the matter there; he did not elicit
this testimony from Babic in front of the jury. He did briefly
question Sadowski about a possible immunity deal in front
of the jury but did not press the point. And when the time
came for jury instructions, Langan did not request anything
about accomplice witnesses or concessions granted to wit‐
nesses. He did, however, attempt to undermine Babic and
Sadowski’s credibility throughout the trial by pointing out
the inconsistencies in their testimony, their lies to police, and
their overall lack of credibility.
   The state appellate court concluded that this omission
did not reflect deficient performance. Counsel knew, from
the sidebar we described earlier, that Babic believed that she
No. 18‐2214                                                  13

had an immunity deal with the state in exchange for her tes‐
timony. Yet Langan left this unexplored in open court. He
never established that Babic had not been charged but could
be. Whether or not there was a formal immunity agreement,
that would have been powerful impeachment evidence. See
Youngblood v. West Virginia, 

, 869 (2006), citing
United States v. Bagley, 

, 676 (1985). Given the
centrality of Babic’s role in the state’s case, the failure to
bring out Babic’s motives for testifying fell below an objec‐
tively reasonable level of performance.
    Langan asked Sadowski only four questions on this
point, and her testimony was equivocal. Nonetheless, he
brought out the fact that she was at risk of being charged but
had not been. He had no explanation for the difference in his
questioning of the two. Indeed, he testified at the postconvic‐
tion hearing that he could not remember why he did not
question Babic in front of the jury and that perhaps he “for‐
got” to do so. See, e.g., Woolley v. Rednour, 

, 423–
24 (7th Cir. 2012) (finding “no strategic rationale” underly‐
ing counsel’s failure to secure expert testimony where coun‐
sel admitted his failure was an oversight).
    Both the state appellate court and the district court found
that Langan’s performance was not so flawed as to be consti‐
tutionally deficient in these respects. We consider this a close
call. Cook argues that walking the jury through Babic’s un‐
derstanding of her immunity would have better exposed her
motivations to protect herself by testifying the way the state
wanted. Perhaps so. On the other hand, the fact that Babic
was an accomplice to the crimes and thus was motivated to
cooperate with the state was developed through other testi‐
mony. Finally, Langan did point out inconsistencies within
14                                                   No. 18‐2214

Babic’s testimony and between Sadowski’s testimony and
hers and discussed their motives to lie in his closing argu‐
ment. Perhaps standing alone, this episode would not justify
a finding of constitutionally deficient performance. But it is
another negative factor when we assess counsel’s perfor‐
mance as a whole.
    Related to the de facto immunity point is Langan’s deci‐
sion not to request a jury instruction advising caution with
the reliability of the testimony of accomplices. At the post‐
conviction hearing, he explained that he did not do so be‐
cause he was satisfied with the general instruction on wit‐
ness credibility. The latter instruction tells the jury to consid‐
er bias, motives for falsifying testimony, and the witnesses’
interest in the result of the trial. Although the more specific
accomplice instruction contained a stronger admonition, the
state court reasonably concluded that Langan’s choice lay
within the broad bounds of acceptable performance. We thus
place no weight on that aspect of his work. The same is true
of the state appellate court’s assessment of the lack of a spe‐
cial instruction on immunity.
     4. Failure to object to Margaret Harper’s unanticipated
        in‐court identification of Cook
    Throughout the period leading up to the trial, neither
Jimmy nor Margaret Harper was able to describe the intrud‐
ers to police. Their daughter recalled only that both were Af‐
rican‐Americans. The police showed Margaret photo arrays
(all of which included Cook’s picture), but she did not iden‐
tify anyone as her attacker.
   At trial, however—and to Langan’s astonishment (but not
that of the prosecutor, who knew what was coming)—
No. 18‐2214                                                  15

Margaret took the witness stand and identified Cook, the on‐
ly black man in the courtroom, as her assailant. After elicit‐
ing testimony that Margaret saw her assailant’s eyes during
a physical struggle, the prosecutor asked if “anything [had]
happened since this incident that would help [her] identify
this person.” Margaret replied that she had seen Cook’s eyes
in court and told her family, “I will never forget those eyes if
I ever saw them again, and I just had a flashback when Mr.
Cook looked at me this morning.” Langan neither objected,
moved to exclude this testimony as unduly prejudicial under
Wis. Stat. § 904.03, nor moved for a mistrial. He did, howev‐
er, address her “flashback” in his closing argument. He
pointed out that it was dark that night; that events were un‐
folding quickly; that the attacker, with “bugged out” eyes,
was screaming at her; and that she had consistently failed to
recognize Cook’s picture in the photo arrays. Addressing the
obvious, he also commented that “given the racial makeup
of this area there’s a decent possibility that this trial is the
closest that she’s ever been to a black man since the night of
the robbery.” In light of all that, he urged the jury to find
that the state had not shown Cook’s guilt beyond a reasona‐
ble doubt.
    The state appellate court determined that counsel’s per‐
formance was not deficient, because the in‐court identifica‐
tion was admissible under the then‐controlling case of State
v. Marshall, 

(Wis. 1979). In Marshall, the Su‐
preme Court of Wisconsin held that the state exclusionary
rule applicable to unnecessarily suggestive identification
procedures did not apply to spontaneous identifications. 

Cook argued that Marshall did not apply because this
identification was worse than suggestive—it was unreliable.
The state appellate court thought that the state supreme
16                                                   No. 18‐2214

court embraced that distinction in May 2006, four months
after the conclusion of Cook’s trial (January 2006). State v.
Hibl, 

, 201–05 (Wis. 2006) (clarifying that even
for spontaneous identifications, courts should exclude undu‐
ly prejudicial identifications). (Whether Hibl simply clarified
state law or introduced a new limitation is an interesting
question, but not one that we need to resolve.)
    Langan’s performance in these respects was far from ide‐
al. Had this been the only flaw, we would have a different
case, but as is clear by now, it was not. In one respect, how‐
ever, we find nothing to criticize: although the trial court
could have, and probably would have, ordered the jury to
disregard Harper’s identification as unduly prejudicial if
Langan had made a proper motion, it is unclear whether his
failure to do so made any difference under then‐prevailing
state law. See 

. More importantly,
as we noted earlier, Langan did attempt to impeach Harper
in other ways.
    Cook also argues that the identification was so unreliable
that any competent attorney would have argued that it vio‐
lated due process. Courts have addressed due‐process con‐
cerns with in‐court identifications, particularly where, as
here, the defendant is the only person of his race in the
courtroom and is seated at the defense table. See, e.g., Lee v.
Foster, 

, 691 (7th Cir. 2014). But at the time of tri‐
al, only a handful of cases would have supported a due‐
process objection. This is not enough. See 

(counsel not deficient for lack of foresight or failing to
prepare for remote possibilities).
     5. Withdrawal of question to Sadowski about Hall’s
        possession of a gun immediately before the crimes
No. 18‐2214                                                17

     At one point, Langan was attempting to establish the fact
that Hall was carrying a gun on the night of the home inva‐
sion. When Sadowski was on the stand, he asked her wheth‐
er she saw Hall with a gun that night. While his words were
still hanging in the air, the prosecutor objected to the ques‐
tion. Without missing a beat, and without giving the court a
chance to rule on the objection, Langan withdrew the ques‐
tion. The judge later indicated (at the post‐conviction hear‐
ing) that he would have overruled the state’s objection and
allowed Langan to proceed. Post‐conviction counsel was
able to get Sadowski to admit that Hall frequently carried a
gun similar to the one used in the crimes. We can think of no
strategic reason why Langan would have wanted to fold so
quickly, and so we regard this as another instance of defi‐
cient performance.
   6. Failure to object to testimony that Cook temporarily
      discontinued his interrogation by police
   Detective Baldwin testified that after Cook initially
waived his Miranda rights, he identified Egerson from a pho‐
to while saying that he had not been in town that day. Bald‐
win then stated that the officers then asked Cook if he want‐
ed to give a written statement but Cook declined and “indi‐
cated that he didn’t want to talk anymore.” Langan did not
object to the latter testimony nor to the introduction of a
signed statement form with “refused” written on it. Alt‐
hough the form was introduced as an exhibit, the jury never
saw it.
    The detectives interviewed Cook again the next day. Af‐
ter again waiving his Miranda rights, Cook admitted that he
was in the area with Egerson on the night of the robbery. Ac‐
cording to Cook, the two went to Babic’s house for a few
18                                                No. 18‐2214

hours before returning to Milwaukee. Langan himself elicit‐
ed the testimony that Cook again invoked his right to silence
at the end of the second interview.
    We see nothing worth pursuing in these events. Cook
contends that the state appellate court unreasonably deter‐
mined that he did not invoke his Fifth Amendment right
against self‐incrimination and unreasonably applied Strick‐
land when it ruled that he was not prejudiced by Langan’s
failure to object to this part of Baldwin’s testimony. But Lan‐
gan’s failure to object to the detective’s testimony was not
deficient. Even though Doyle v. Ohio, 

, 619 (1976),
holds that silence following Miranda warnings may not be
used to impeach a defendant’s testimony at trial, it does not
bar all testimony about a defendant’s silence. See Anderson v.
Charles, 

, 408–09 (1980) (questions about post‐
Miranda silence permissible to “elicit an explanation for a
prior inconsistent statement”). Whether questions about a
defendant’s post‐Miranda silence violate due process de‐
pends on “the particular use to which the post‐arrest silence
is being put.” Lindgren v. Lane, 

, 202 (7th Cir.
1991).
    The state appellate court reasonably found that the pros‐
ecution here did not improperly use Cook’s silence as evi‐
dence of his guilt or to impeach him. See Bieghler v. McBride,

, 705, 707 (7th Cir. 2004). The state did not focus
on this aspect of Baldwin’s testimony. Cook’s termination of
questioning during his first interview came up only as part
of his summary of the sequence of events in the investigation
leading up to Cook’s admissions. See Splunge v. Parke, 

, 371–73 (7th Cir. 1998) (testimony on defendant’s
silence did not violate due process where prosecutor men‐
No. 18‐2214                                                19

tioned silence in sequence of events as a “prologue” to the
introduction of statements defendant did make). Counsel
was not deficient for failing to object to this testimony.
                              B
    Before turning to the question of prejudice, we add a few
facts that came out at the postconviction evidentiary hearing.
They shed additional light on Langan’s performance.
   Sadowski testified that she and Hall had heard a rumor
that the Harpers had a large amount of marijuana in their
garage. Based on that rumor, she and Hall had scoped out
the garage a few times with an ex‐boyfriend of the Harpers’
daughter. Although she claimed not to have seen Hall with a
gun on the night of the robbery, Sadowski admitted that
Hall had his gun on him “a lot” and it “[c]ould have been”
the same type of gun as the one used in the robbery.
    The court also heard evidence from the state and Hall
himself about Hall’s whereabouts at the time of Cook’s trial.
The state established that Hall was then on parole but in
“absconder status” after failing to show up for an appoint‐
ment with his parole officer; a warrant was also out for his
arrest. Hall’s parole officer had tried to conduct a home visit,
but Hall’s roommates reported that he had been kicked out.
Hall (who was brought in from prison for the hearing) testi‐
fied that he had been living and working at his usual loca‐
tions and that he would have complied with a subpoena to
testify at Cook’s trial. He also admitted to stopping by
Babic’s the night of the robbery but denied any involvement
in the crime; he further denied having a gun that night or
previously casing the Harpers’ garage.
20                                                No. 18‐2214

    Finally, Langan testified. He could not remember why he
did not question Babic on immunity or concessions in front
of the jury, saying only that perhaps he “forgot” to do so. He
explained that he did not request an accomplice jury instruc‐
tion—which would have told the jury to consider the wom‐
en’s testimony with “caution and great care”—because he
was satisfied with the general instruction about witness
credibility. He also testified that he knew that the state did
not disclose a witness who could authenticate Cook’s cell‐
phone records. But he did not object to the detective’s testi‐
mony about Cook’s cellphone being used the night of the
robbery because he had reviewed the records and believed
they were authentic and because he did not want to bog
down the jury with cumulative evidence. The court stated
that it thought it would have sustained an objection, had
Langan made one, and that the state could not have “gotten
that testimony in front of the jury if [counsel] had been do‐
ing his job.”
                              IV
    Although we (along with the state appellate court) have
found a number of examples of deficient performance, that
is not enough to win the day for Cook. He can prevail only if
he shows both substandard performance and prejudice. We
must now ascertain whether the state court unreasonably
concluded that Cook was not prejudiced by his lawyer’s fail‐
ings. In conducting that inquiry, we bear in mind that feder‐
al review of a state court’s application of Strickland is “dou‐
bly deferential”: the federal court must give first the defense
attorney and then the last state court to rule on the matter
“the benefit of the doubt.” Burt v. Titlow, 

, 15
(2013).
No. 18‐2214                                                  21

    Rather than ask how prejudicial each individual error
was, we evaluate Langan’s performance as a whole, or as
Cook has put it, we consider the cumulative effect of coun‐
sel’s errors. See Goodman v. Bertrand, 

, 1030 (7th
Cir. 2006). For this purpose, we set aside any alleged error
for which Langan’s performance did not fall below the con‐
stitutional minimum; we look only at the question whether
areas in which his performance was deficient, taken as a
whole, led to a reasonable probability of a different result.
    Particularly bearing in mind some of the critical factual
misapprehensions (perhaps triggered by the state’s over‐
statements in its briefs) that the state appellate court had, we
find (as did the experienced state trial judge) that counsel’s
performance, taken as a whole, undermines our confidence
in the outcome of Cook’s trial.
    We rely in particular on three significant errors: (1) Lan‐
gan’s failure to investigate Hall’s location; (2) his failure to
question Babic about immunity; and (3) his failure to object
to or otherwise seek to exclude the hearsay testimony about
the location and use of Cook’s cellphone. Taken together,
these instances of deficient performance undermined the tri‐
al judge’s confidence in the result of the trial, and as an ob‐
jective matter we come to the same conclusion. See, e.g.,

(counsel’s failure to subpoena im‐
portant defense witness because he thought the state would,
was prejudicial when combined with “catalog” of other er‐
rors).
   In deciding whether there is a reasonable probability that
the errors changed the outcome of the trial, the court must
consider all of the evidence. Logically, a verdict weakly sup‐
ported by the record is more likely to have been affected by
22                                                  No. 18‐2214

errors than one with overwhelming record support. See
Blackmon v. Williams, 

, 1105 (7th Cir. 2016). We
recognize that even at a hypothetical trial where counsel did
not make these errors, the jury would have learned of Cook’s
limited admissions and would have heard Babic and Sa‐
dowski implicate him in the robbery. But the jury would not
have learned that Cook’s phone was used to call Egerson’s
immediately after the robbery from a location near the
Harpers’ home. That evidence not only independently
linked Cook to the robbery; it also corroborated the women’s
identification of him as a participant. Further, the state’s case
relied almost entirely on Babic and Sadowski’s testimony,
yet counsel failed to pursue an important opportunity to im‐
peach Babic with her de facto immunity. He compounded
that error by failing to ask for instructions specifically un‐
dermining the value of both women’s testimony. If the jury
found the women not credible and the cellphone testimony
were excluded, the state’s evidence would have been re‐
duced to Cook’s admission that he was with the women and
Egerson that night. But so was Hall. And had counsel taken
reasonable steps to find Hall, the testimony at the post‐
conviction hearing indicates he would have succeeded, and
Hall would have complied with a trial subpoena. The jury—
and Margaret Harper, for that matter—would have seen
Hall, assessed his credibility, and heard his testimony, which
conflicted in some respects with Sadowski’s. Most im‐
portantly, the possibility that a different man was Egerson’s
accomplice would have been a concrete reality for the jury,
not just talk.
   To establish prejudice, Cook did not need to prove “that
counsel’s deficient conduct more likely than not altered the
outcome in the case.” 

. Instead, he
No. 18‐2214                                                 23

had to “show that there is a reasonable probability that, but
for counsel’s unprofessional errors, the result of the proceed‐
ing would have been different. A reasonable probability is a
probability sufficient to undermine confidence in the out‐
come.” 

We conclude that the state appellate court
unreasonably applied that standard when it concluded that
the result in Cook’s case could stand, despite counsel’s sub‐
par performance.
   We therefore REVERSE the judgment of the district court
and order that Cook’s petition for a writ of habeas corpus be
granted, subject to the state’s right to decide, within 120
days, whether to proceed with a new trial.