Li's deep blue eyes gazed at him amicably, but Rubin could see she'd have no reservation about carrying out her threat. The woman scared him.

'I think you get my drift.' Li thumped him on the shoulder and joined the others. 'OK. Let's talk about damage limitation. Did the drug work?'

'We injected ten mills,' said Peak. 'Any more than that would have really knocked him sideways, and we need his brain. The drug's supposed to work like an eraser on the mind. But there's no guarantee that his memory won't come back.'

'What kind of risk are we talking?'

'It's hard to say. A word, a colour, a smell could do it. Once the brain finds a trigger, it's capable of remembering exactly what happened.'

'Well, that's quite some risk.' Vanderbilt scowled. 'No drug can suppress a memory entirely. We still don't know enough about the workings of the brain.'

'We'll have to keep him under observation,' said Li. 'What do you think, Mick? How much longer are we going to need him?'

'Oh, we're going great guns,' Rubin said eagerly. Here was his chance to regain lost ground. 'Weaver and Anawak are working on the idea of pheromone-induced aggregation. Oliviera and Johanson think it might be scent-based too. This afternoon we'll be running some phase tests, and we should get our proof. If we're right about aggregation being triggered by scent, we'll soon be in a position to proceed as intended.'

'If, should, could.' Vanderbilt snorted. 'How long until you come up with a goddamned formula?'

'This is scientific research, Jack,' said Rubin. 'No one stood over Alexander Fleming and kept telling him to hurry up and discover penicillin.'

Vanderbilt was on the point of responding when a woman stood up and walked over.

'They've decoded Scratch in the CIC,' she said.

'Scratch?'

'Seems that way. Crowe said to Shankar that they'd figured it out.'

Li turned towards the desk where the audio and video footage from the CIC was being processed. A view from the overhead camera showed Shankar, Crowe and Anawak in conversation. Weaver had just walked in.

'They'll call us in a minute,' she said. 'Good. Don't forget to look surprised.'

COMBAT INFORMATION CENTER

Everyone was crowding round Crowe and Shankar, trying to get a look at the message. What they were seeing wasn't a spectrogram but a graphic representation of the transmission they'd received the day before.

'Is it a reply?' asked Li.

'Good question,' said Crowe.

'But what is Scratch anyway?' asked Greywolf who'd just arrived with Delaware. 'A language?'

'Well, Scratch itself might be a language, but this signal isn't part of it, not in the way it's been coded here,' said Shankar. 'It's like the Arecibo message. I mean, humans don't usually communicate in binary code. If you think about it, we didn't send that message into space. Our computers did.'

'The good news,' said Crowe, 'is that we've worked out its structure. You know how Scratch sounds as though a needle's being dragged across a record? Well, it's a staccato vibration of a very low frequency, ideally suited for propagating across the ocean. Infrasonic waves can travel incredible distances. And in this case the wavelength is extremely short. The trouble with infrasound is that we have to speed up any sound with a frequency of less than a hundred hertz to make it audible, but that would speed up the staccato. The trick to understanding this signal lies in slowing it down.'

'We had to stretch it,' said Shankar, 'to be able to identify the individual components. So we slowed it right down until the scratching noise became a sequence of individual pulses varying in length and intensity.'

'Sounds like Morse code,' said Weaver.

'It seems to work like it too.'

'How are you transcribing it?' asked Li. 'With spectrograms?'

'Yes, but they aren't enough. When it's a question of listening to something, it's always better to hear it. So we used an acoustic trick. It's a bit like false colour being added to radar images to show up the detail. In this instance we took each individual sound and replaced it with a frequency that we can hear, while keeping its original length and intensity. Whenever the original signal switched frequencies, we modified ours. That's how we handled Scratch.'

Crowe punched something into the keyboard. 'The sound we detected is like this.'

There was a rumble, like an underwater drum. The beats followed in quick succession, almost too fast to tell apart, but there could be no doubt that they were listening to a sequence of noises that varied in volume and duration.

'Well, it sounds like code,' said Anawak, 'but what does it mean?'

'We don't know.'

'You don't know?' echoed Vanderbilt. 'But I thought you'd cracked it.'

'What we don't know,' Crowe explained patiently, 'is how their language might work when they're using it normally. We can't make head or tail of the previous Scratch signals. But that's beside the point.' Smoke curled from her nostrils. 'We've got something better. We've got contact. Murray, show them the first part.'

Shankar clicked on an icon. The screen was lined with rows of numbers. Some columns seemed identical.

'We sent them some homework, as you know,' said Shankar. 'Math questions. Like an IQ test. We asked them to continue decimal sequences, work out some logarithms, fill in the missing numbers, that kind of thing. If it worked, we were hoping they'd find it kind of fun and send us a reply. It would be their way of telling us that they'd heard us, that they really exist, that they know about math and can manipulate numbers.' He pointed to the rows of figures on the screen. 'This is their answer. Grade A. They got everything right.'

'Christ,' whispered Weaver.

'That tells us two things,' said Crowe. 'First, Scratch is indeed a kind of language. In all probability, each of the Scratch signals contains complex information. Second, and this is the decisive point, it proves that they're capable of adapting Scratch so that we can understand it. That's an achievement of the highest order. It tells us that they're every bit as smart as we are. They're capable of decoding our messages; but they can also code their own.'

For a while they just stared at the columns of figures, admiration mixed with fear.

'But what does it prove?' asked Johanson, breaking the silence.

'I should have thought that was obvious,' retorted Delaware. 'It proves something's down there. Something that can think.'

'OK, but couldn't a computer generate the same results?'

'You don't think we're talking to a computer, do you?'

'He's got a point, you know,' said Anawak. 'All it proves is that our math questions have been answered. That's impressive, but it's not evidence of conscious intelligence.'

'But what else could be sending us messages?' asked Greywolf, disbelievingly. 'Mackerel?'

'Nonsense. Think about it. What we're seeing here is the work of a creature that can manipulate symbols. That's not proof of higher intelligence per se. Take chameleons, for instance. They solve a highly complex processing problem every time they change colour, but they've got no idea they're doing it. If you weren't acquainted with the IQ of chameleons, you might suppose they're very clever – after all, they can use a program that allows them to resemble foliage one day and a rock face the next. You'd probably credit them with enormous insight because they're reading the code of their surroundings. And you'd assume they were creative because they change their code to match.'

'So what are we looking at?' asked Delaware, helplessly.

Crowe smiled. 'Leon's right,' she said. 'Just because someone can manipulate symbols doesn't mean they understand them. The real proof of intelligence and creativity resides in a creature's ability to understand and conceptualise conditions in the real world. That requires a deeper understanding. Even the most highly powered computer doesn't deal in rules of thumb or counter-intuitive decisions. It can't engage with its environment or experience the world. I imagine the yrr had the same considerations in mind when they formulated their reply. They tried to find something that would signal to us they're capable of real understanding.' Crowe pointed to the screen. 'These are the results of the two math problems. If you look closely, you'll see that the first answer appears eleven times in a row, then you get three repetitions of answer number two, a single occurrence of answer number one, then nine times number two and so on. At one point the second answer appears nearly thirty thousand times. But why? It makes sense to send us the results more than once, of course, even if only to make sure that the message is long enough to be detected. But why would they mix them all together?'